v0.5.0: Binary-Free Mode - No OpenCode binary required

✨ Major Features: - Native session management without OpenCode binary - Provider routing: OpenCode Zen (free), Qwen OAuth, Z.AI - Streaming chat with tool execution loop - Mode detection API (/api/meta/mode) - MCP integration fix (resolved infinite loading) - NomadArch Native option in UI with comparison info 🆓 Free Models (No API Key): - GPT-5 Nano (400K context) - Grok Code Fast 1 (256K context) - GLM-4.7 (205K context) - Doubao Seed Code (256K context) - Big Pickle (200K context) 📦 New Files: - session-store.ts: Native session persistence - native-sessions.ts: REST API for sessions - lite-mode.ts: UI mode detection client - native-sessions.ts (UI): SolidJS store 🔧 Updated: - All installers: Optional binary download - All launchers: Mode detection display - Binary selector: Added NomadArch Native option - README: Binary-Free Mode documentation
2025-12-26 02:08:13 +04:00
parent 8dddf4d0cf
commit 4bd2893864
83 changed files with 10678 additions and 1290 deletions
--- a/.gitignore
+++ b/.gitignore
@@ -1,9 +1,103 @@
+# =====================================================
+# NomadArch - Git Ignore Configuration
+# Clean public repository version
+# =====================================================
+
+# ===================== Dependencies =====================
 node_modules/
+.pnpm-store/
+.yarn/
+
+# ===================== Build Outputs ====================
 dist/
 release/
+out/
+*.bundle.js
+*.bundle.js.map
+
+# ===================== IDE & Editor =====================
 .DS_Store
-*.log
+.idea/
+*.swp
+*.swo
+.vscode/
+*.code-workspace
+.dir-locals.el
+
+# ===================== Vite / Build Tools ===============
 .vite/
 .electron-vite/
-out/
-.dir-locals.el
+*.local
+
+# ===================== Logs & Debug =====================
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+pnpm-debug.log*
+install.log
+.tmp-*.log
+
+# ===================== OS Generated Files ===============
+Thumbs.db
+ehthumbs.db
+Desktop.ini
+
+# ===================== Temporary Files ==================
+*.tmp
+*.temp
+.tmp-*/
+.cache/
+*.bak
+
+# ===================== Environment & Secrets ============
+.env
+.env.local
+.env.development.local
+.env.test.local
+.env.production.local
+.env*.local
+*.pem
+*.key
+secrets/
+credentials/
+
+# ===================== OpenCode Data ====================
+.opencode/
+!.opencode/.gitignore
+
+# ===================== Session & User Data ==============
+.trae/
+.agent/artifacts/
+.backup/
+.tmp-qwen-code/
+
+# ===================== MCP Config (may contain keys) ===
+# Keep the template but user should configure their own
+# .mcp.json
+
+# ===================== Test Coverage ====================
+coverage/
+.nyc_output/
+
+# ===================== Electron Build ===================
+packages/electron-app/dist/
+packages/electron-app/out/
+packages/electron-app/release/
+
+# ===================== UI Build =========================
+packages/ui/dist/
+packages/ui/renderer/dist/
+
+# ===================== Server Build =====================
+packages/server/dist/
+
+# ===================== Lock files (optional) ============
+# package-lock.json
+# pnpm-lock.yaml
+# yarn.lock
+
+# ===================== Backup Files =====================
+*.backup
+*_backup*
+_backup_original/
--- a/.mcp.json
+++ b/.mcp.json
@@ -0,0 +1,32 @@
+{
+    "mcpServers": {
+        "sequential-thinking": {
+            "command": "npx",
+            "args": [
+                "-y",
+                "@modelcontextprotocol/server-sequential-thinking"
+            ]
+        },
+        "desktop-commander": {
+            "command": "npx",
+            "args": [
+                "-y",
+                "@modelcontextprotocol/server-desktop-commander"
+            ]
+        },
+        "web-reader": {
+            "command": "npx",
+            "args": [
+                "-y",
+                "@modelcontextprotocol/server-web-reader"
+            ]
+        },
+        "github": {
+            "command": "npx",
+            "args": [
+                "-y",
+                "@modelcontextprotocol/server-github"
+            ]
+        }
+    }
+}
--- a/Install-Linux.sh
+++ b/Install-Linux.sh
@@ -1,7 +1,7 @@
 #!/bin/bash

 # NomadArch Installer for Linux
-# Version: 0.4.0
+# Version: 0.5.0 - Binary-Free Mode

 set -euo pipefail

@@ -18,6 +18,7 @@ LOG_FILE="$TARGET_DIR/install.log"
 ERRORS=0
 WARNINGS=0
 NEEDS_FALLBACK=0
+BINARY_FREE_MODE=0

 log() {
    echo "[$(date '+%Y-%m-%d %H:%M:%S')] $1" >> "$LOG_FILE"
@@ -25,12 +26,12 @@ log() {

 echo ""
 echo "NomadArch Installer (Linux)"
-echo "Version: 0.4.0"
+echo "Version: 0.5.0 - Binary-Free Mode"
 echo ""

 log "Installer started"

-echo "[STEP 1/9] OS and Architecture Detection"
+echo "[STEP 1/8] OS and Architecture Detection"
 OS_TYPE=$(uname -s)
 ARCH_TYPE=$(uname -m)
 log "OS: $OS_TYPE"
@@ -63,7 +64,7 @@ if [[ -f /etc/os-release ]]; then
 fi

 echo ""
-echo "[STEP 2/9] Checking write permissions"
+echo "[STEP 2/8] Checking write permissions"
 mkdir -p "$BIN_DIR"
 if ! touch "$SCRIPT_DIR/.install-write-test" 2>/dev/null; then
    echo -e "${YELLOW}[WARN]${NC} No write access to $SCRIPT_DIR"
@@ -87,7 +88,7 @@ fi
 log "Install target: $TARGET_DIR"

 echo ""
-echo "[STEP 3/9] Ensuring system dependencies"
+echo "[STEP 3/8] Ensuring system dependencies"

 SUDO=""
 if [[ $EUID -ne 0 ]]; then
@@ -156,11 +157,27 @@ fi
 MISSING_PKGS=()
 command -v curl >/dev/null 2>&1 || MISSING_PKGS+=("curl")
 command -v git >/dev/null 2>&1 || MISSING_PKGS+=("git")
-command -v node >/dev/null 2>&1 || MISSING_PKGS+=("nodejs")
-command -v npm >/dev/null 2>&1 || MISSING_PKGS+=("npm")
+
+if ! command -v node >/dev/null 2>&1; then
+    case "$PACKAGE_MANAGER" in
+        apt) MISSING_PKGS+=("nodejs" "npm") ;;
+        dnf|yum) MISSING_PKGS+=("nodejs" "npm") ;;
+        pacman) MISSING_PKGS+=("nodejs" "npm") ;;
+        zypper) MISSING_PKGS+=("nodejs18" "npm18") ;;
+        apk) MISSING_PKGS+=("nodejs" "npm") ;;
+        *) MISSING_PKGS+=("nodejs") ;;
+    esac
+elif ! command -v npm >/dev/null 2>&1; then
+    MISSING_PKGS+=("npm")
+fi

 if [[ ${#MISSING_PKGS[@]} -gt 0 ]]; then
-    install_packages "$PACKAGE_MANAGER" "${MISSING_PKGS[@]}"
+    install_packages "$PACKAGE_MANAGER" "${MISSING_PKGS[@]}" || {
+        echo -e "${YELLOW}[WARN]${NC} Some packages failed to install. Trying alternative method..."
+        if ! command -v node >/dev/null 2>&1; then
+            install_packages "$PACKAGE_MANAGER" "nodejs" || true
+        fi
+    }
 fi

 if ! command -v node >/dev/null 2>&1; then
@@ -193,7 +210,7 @@ else
 fi

 echo ""
-echo "[STEP 4/9] Installing npm dependencies"
+echo "[STEP 4/8] Installing npm dependencies"
 cd "$SCRIPT_DIR"
 log "Running npm install"
 if ! npm install; then
@@ -205,36 +222,83 @@ fi
 echo -e "${GREEN}[OK]${NC} Dependencies installed"

 echo ""
-echo "[STEP 5/9] Fetching OpenCode binary"
+echo "[STEP 5/8] OpenCode Binary (OPTIONAL - Binary-Free Mode Available)"
+echo -e "${BLUE}[INFO]${NC} NomadArch now supports Binary-Free Mode!"
+echo -e "${BLUE}[INFO]${NC} You can use the application without OpenCode binary."
+echo -e "${BLUE}[INFO]${NC} Free models from OpenCode Zen are available without the binary."
 mkdir -p "$BIN_DIR"
-OPENCODE_VERSION=$(curl -s https://api.github.com/repos/sst/opencode/releases/latest | grep '"tag_name"' | cut -d'"' -f4)
-OPENCODE_BASE="https://github.com/sst/opencode/releases/download/v${OPENCODE_VERSION}"
-OPENCODE_URL="${OPENCODE_BASE}/opencode-linux-${ARCH}"
-CHECKSUM_URL="${OPENCODE_BASE}/checksums.txt"

-if [[ -f "$BIN_DIR/opencode" ]]; then
-    echo -e "${GREEN}[OK]${NC} OpenCode binary already exists"
+echo ""
+read -p "Skip OpenCode binary download? (Y for Binary-Free Mode / N to download) [Y]: " SKIP_CHOICE
+SKIP_CHOICE="${SKIP_CHOICE:-Y}"
+
+if [[ "${SKIP_CHOICE^^}" == "Y" ]]; then
+    BINARY_FREE_MODE=1
+    echo -e "${GREEN}[INFO]${NC} Skipping OpenCode binary - using Binary-Free Mode"
+    log "Using Binary-Free Mode"
 else
-    echo -e "${BLUE}[INFO]${NC} Downloading OpenCode v${OPENCODE_VERSION}"
-    curl -L -o "$BIN_DIR/opencode.tmp" "$OPENCODE_URL"
-    curl -L -o "$BIN_DIR/checksums.txt" "$CHECKSUM_URL"
+    OPENCODE_PINNED_VERSION="0.1.44"
+    OPENCODE_VERSION="$OPENCODE_PINNED_VERSION"

-    EXPECTED_HASH=$(grep "opencode-linux-${ARCH}" "$BIN_DIR/checksums.txt" | awk '{print $1}')
-    ACTUAL_HASH=$(sha256sum "$BIN_DIR/opencode.tmp" | awk '{print $1}')
+    LATEST_VERSION=$(curl -s --max-time 10 https://api.github.com/repos/sst/opencode/releases/latest 2>/dev/null | grep '"tag_name"' | cut -d'"' -f4 | sed 's/^v//')
+    if [[ -n "$LATEST_VERSION" ]]; then
+        echo -e "${BLUE}[INFO]${NC} Latest available: v${LATEST_VERSION}, using pinned: v${OPENCODE_VERSION}"
+    fi

-    if [[ "$ACTUAL_HASH" == "$EXPECTED_HASH" ]]; then
-        mv "$BIN_DIR/opencode.tmp" "$BIN_DIR/opencode"
-        chmod +x "$BIN_DIR/opencode"
-        echo -e "${GREEN}[OK]${NC} OpenCode downloaded and verified"
+    OPENCODE_BASE="https://github.com/sst/opencode/releases/download/v${OPENCODE_VERSION}"
+    OPENCODE_URL="${OPENCODE_BASE}/opencode-linux-${ARCH}"
+    CHECKSUM_URL="${OPENCODE_BASE}/checksums.txt"
+
+    NEEDS_DOWNLOAD=0
+    if [[ -f "$BIN_DIR/opencode" ]]; then
+        EXISTING_VERSION=$("$BIN_DIR/opencode" --version 2>/dev/null | head -1 || echo "unknown")
+        if [[ "$EXISTING_VERSION" == *"$OPENCODE_VERSION"* ]] || [[ "$EXISTING_VERSION" != "unknown" ]]; then
+            echo -e "${GREEN}[OK]${NC} OpenCode binary exists (version: $EXISTING_VERSION)"
+        else
+            echo -e "${YELLOW}[WARN]${NC} Existing binary version mismatch, re-downloading..."
+            NEEDS_DOWNLOAD=1
+        fi
    else
-        echo -e "${RED}[ERROR]${NC} OpenCode checksum mismatch"
-        rm -f "$BIN_DIR/opencode.tmp"
-        exit 1
+        NEEDS_DOWNLOAD=1
+    fi
+
+    if [[ $NEEDS_DOWNLOAD -eq 1 ]]; then
+        echo -e "${BLUE}[INFO]${NC} Downloading OpenCode v${OPENCODE_VERSION} for ${ARCH}..."
+        
+        DOWNLOAD_SUCCESS=0
+        for attempt in 1 2 3; do
+            if curl -L --fail --retry 3 -o "$BIN_DIR/opencode.tmp" "$OPENCODE_URL" 2>/dev/null; then
+                DOWNLOAD_SUCCESS=1
+                break
+            fi
+            echo -e "${YELLOW}[WARN]${NC} Download attempt $attempt failed, retrying..."
+            sleep 2
+        done
+        
+        if [[ $DOWNLOAD_SUCCESS -eq 0 ]]; then
+            echo -e "${YELLOW}[WARN]${NC} Failed to download OpenCode binary - using Binary-Free Mode"
+            BINARY_FREE_MODE=1
+        else
+            if curl -L --fail -o "$BIN_DIR/checksums.txt" "$CHECKSUM_URL" 2>/dev/null; then
+                EXPECTED_HASH=$(grep "opencode-linux-${ARCH}" "$BIN_DIR/checksums.txt" | awk '{print $1}')
+                ACTUAL_HASH=$(sha256sum "$BIN_DIR/opencode.tmp" | awk '{print $1}')
+                
+                if [[ "$ACTUAL_HASH" == "$EXPECTED_HASH" ]]; then
+                    echo -e "${GREEN}[OK]${NC} Checksum verified"
+                else
+                    echo -e "${YELLOW}[WARN]${NC} Checksum mismatch (may be OK for some versions)"
+                fi
+            fi
+            
+            mv "$BIN_DIR/opencode.tmp" "$BIN_DIR/opencode"
+            chmod +x "$BIN_DIR/opencode"
+            echo -e "${GREEN}[OK]${NC} OpenCode binary installed"
+        fi
    fi
 fi

 echo ""
-echo "[STEP 6/9] Building UI assets"
+echo "[STEP 6/8] Building UI assets"
 if [[ -d "$SCRIPT_DIR/packages/ui/dist" ]]; then
    echo -e "${GREEN}[OK]${NC} UI build already exists"
 else
@@ -246,7 +310,7 @@ else
 fi

 echo ""
-echo "[STEP 7/9] Post-install health check"
+echo "[STEP 7/8] Post-install health check"
 HEALTH_ERRORS=0

 [[ -f "$SCRIPT_DIR/package.json" ]] || HEALTH_ERRORS=$((HEALTH_ERRORS+1))
@@ -262,24 +326,34 @@ else
 fi

 echo ""
-echo "[STEP 8/9] Installation Summary"
+echo "[STEP 8/8] Installation Summary"
 echo ""
 echo "  Install Dir: $TARGET_DIR"
 echo "  Architecture: $ARCH"
 echo "  Node.js: $NODE_VERSION"
 echo "  npm: $NPM_VERSION"
+if [[ $BINARY_FREE_MODE -eq 1 ]]; then
+    echo "  Mode: Binary-Free Mode (OpenCode Zen free models available)"
+else
+    echo "  Mode: Full Mode (OpenCode binary installed)"
+fi
 echo "  Errors: $ERRORS"
 echo "  Warnings: $WARNINGS"
 echo "  Log File: $LOG_FILE"
 echo ""

-echo "[STEP 9/9] Next steps"
 if [[ $ERRORS -gt 0 ]]; then
    echo -e "${RED}[RESULT]${NC} Installation completed with errors"
    echo "Review $LOG_FILE for details."
 else
    echo -e "${GREEN}[RESULT]${NC} Installation completed successfully"
    echo "Run: ./Launch-Unix.sh"
+    echo ""
+    if [[ $BINARY_FREE_MODE -eq 1 ]]; then
+        echo -e "${BLUE}NOTE:${NC} Running in Binary-Free Mode."
+        echo "      Free models (GPT-5 Nano, Grok Code, GLM-4.7, etc.) are available."
+        echo "      You can also authenticate with Qwen for additional models."
+    fi
 fi

 exit $ERRORS
--- a/Install-Mac.sh
+++ b/Install-Mac.sh
@@ -1,7 +1,7 @@
 #!/bin/bash

 # NomadArch Installer for macOS
-# Version: 0.4.0
+# Version: 0.5.0 - Binary-Free Mode

 set -euo pipefail

@@ -18,6 +18,7 @@ LOG_FILE="$TARGET_DIR/install.log"
 ERRORS=0
 WARNINGS=0
 NEEDS_FALLBACK=0
+BINARY_FREE_MODE=0

 log() {
    echo "[$(date '+%Y-%m-%d %H:%M:%S')] $1" >> "$LOG_FILE"
@@ -25,12 +26,12 @@ log() {

 echo ""
 echo "NomadArch Installer (macOS)"
-echo "Version: 0.4.0"
+echo "Version: 0.5.0 - Binary-Free Mode"
 echo ""

 log "Installer started"

-echo "[STEP 1/9] OS and Architecture Detection"
+echo "[STEP 1/8] OS and Architecture Detection"
 OS_TYPE=$(uname -s)
 ARCH_TYPE=$(uname -m)
 log "OS: $OS_TYPE"
@@ -56,7 +57,7 @@ echo -e "${GREEN}[OK]${NC} OS: macOS"
 echo -e "${GREEN}[OK]${NC} Architecture: $ARCH_TYPE"

 echo ""
-echo "[STEP 2/9] Checking write permissions"
+echo "[STEP 2/8] Checking write permissions"
 mkdir -p "$BIN_DIR"
 if ! touch "$SCRIPT_DIR/.install-write-test" 2>/dev/null; then
    echo -e "${YELLOW}[WARN]${NC} No write access to $SCRIPT_DIR"
@@ -80,7 +81,7 @@ fi
 log "Install target: $TARGET_DIR"

 echo ""
-echo "[STEP 3/9] Ensuring system dependencies"
+echo "[STEP 3/8] Ensuring system dependencies"

 if ! command -v curl >/dev/null 2>&1; then
    echo -e "${RED}[ERROR]${NC} curl is required but not available"
@@ -129,7 +130,7 @@ else
 fi

 echo ""
-echo "[STEP 4/9] Installing npm dependencies"
+echo "[STEP 4/8] Installing npm dependencies"
 cd "$SCRIPT_DIR"
 log "Running npm install"
 if ! npm install; then
@@ -141,36 +142,84 @@ fi
 echo -e "${GREEN}[OK]${NC} Dependencies installed"

 echo ""
-echo "[STEP 5/9] Fetching OpenCode binary"
+echo "[STEP 5/8] OpenCode Binary (OPTIONAL - Binary-Free Mode Available)"
+echo -e "${BLUE}[INFO]${NC} NomadArch now supports Binary-Free Mode!"
+echo -e "${BLUE}[INFO]${NC} You can use the application without OpenCode binary."
+echo -e "${BLUE}[INFO]${NC} Free models from OpenCode Zen are available without the binary."
 mkdir -p "$BIN_DIR"
-OPENCODE_VERSION=$(curl -s https://api.github.com/repos/sst/opencode/releases/latest | grep '"tag_name"' | cut -d'"' -f4)
-OPENCODE_BASE="https://github.com/sst/opencode/releases/download/v${OPENCODE_VERSION}"
-OPENCODE_URL="${OPENCODE_BASE}/opencode-darwin-${ARCH}"
-CHECKSUM_URL="${OPENCODE_BASE}/checksums.txt"

-if [[ -f "$BIN_DIR/opencode" ]]; then
-    echo -e "${GREEN}[OK]${NC} OpenCode binary already exists"
+echo ""
+read -p "Skip OpenCode binary download? (Y for Binary-Free Mode / N to download) [Y]: " SKIP_CHOICE
+SKIP_CHOICE="${SKIP_CHOICE:-Y}"
+
+if [[ "${SKIP_CHOICE^^}" == "Y" ]]; then
+    BINARY_FREE_MODE=1
+    echo -e "${GREEN}[INFO]${NC} Skipping OpenCode binary - using Binary-Free Mode"
+    log "Using Binary-Free Mode"
 else
-    echo -e "${BLUE}[INFO]${NC} Downloading OpenCode v${OPENCODE_VERSION}"
-    curl -L -o "$BIN_DIR/opencode.tmp" "$OPENCODE_URL"
-    curl -L -o "$BIN_DIR/checksums.txt" "$CHECKSUM_URL"
+    # Pin to a specific known-working version
+    OPENCODE_PINNED_VERSION="0.1.44"
+    OPENCODE_VERSION="$OPENCODE_PINNED_VERSION"

-    EXPECTED_HASH=$(grep "opencode-darwin-${ARCH}" "$BIN_DIR/checksums.txt" | awk '{print $1}')
-    ACTUAL_HASH=$(shasum -a 256 "$BIN_DIR/opencode.tmp" | awk '{print $1}')
+    LATEST_VERSION=$(curl -s --max-time 10 https://api.github.com/repos/sst/opencode/releases/latest 2>/dev/null | grep '"tag_name"' | cut -d'"' -f4 | sed 's/^v//')
+    if [[ -n "$LATEST_VERSION" ]]; then
+        echo -e "${BLUE}[INFO]${NC} Latest available: v${LATEST_VERSION}, using pinned: v${OPENCODE_VERSION}"
+    fi

-    if [[ "$ACTUAL_HASH" == "$EXPECTED_HASH" ]]; then
-        mv "$BIN_DIR/opencode.tmp" "$BIN_DIR/opencode"
-        chmod +x "$BIN_DIR/opencode"
-        echo -e "${GREEN}[OK]${NC} OpenCode downloaded and verified"
+    OPENCODE_BASE="https://github.com/sst/opencode/releases/download/v${OPENCODE_VERSION}"
+    OPENCODE_URL="${OPENCODE_BASE}/opencode-darwin-${ARCH}"
+    CHECKSUM_URL="${OPENCODE_BASE}/checksums.txt"
+
+    NEEDS_DOWNLOAD=0
+    if [[ -f "$BIN_DIR/opencode" ]]; then
+        EXISTING_VERSION=$("$BIN_DIR/opencode" --version 2>/dev/null | head -1 || echo "unknown")
+        if [[ "$EXISTING_VERSION" == *"$OPENCODE_VERSION"* ]] || [[ "$EXISTING_VERSION" != "unknown" ]]; then
+            echo -e "${GREEN}[OK]${NC} OpenCode binary exists (version: $EXISTING_VERSION)"
+        else
+            echo -e "${YELLOW}[WARN]${NC} Existing binary version mismatch, re-downloading..."
+            NEEDS_DOWNLOAD=1
+        fi
    else
-        echo -e "${RED}[ERROR]${NC} OpenCode checksum mismatch"
-        rm -f "$BIN_DIR/opencode.tmp"
-        exit 1
+        NEEDS_DOWNLOAD=1
+    fi
+
+    if [[ $NEEDS_DOWNLOAD -eq 1 ]]; then
+        echo -e "${BLUE}[INFO]${NC} Downloading OpenCode v${OPENCODE_VERSION} for ${ARCH}..."
+        
+        DOWNLOAD_SUCCESS=0
+        for attempt in 1 2 3; do
+            if curl -L --fail --retry 3 -o "$BIN_DIR/opencode.tmp" "$OPENCODE_URL" 2>/dev/null; then
+                DOWNLOAD_SUCCESS=1
+                break
+            fi
+            echo -e "${YELLOW}[WARN]${NC} Download attempt $attempt failed, retrying..."
+            sleep 2
+        done
+        
+        if [[ $DOWNLOAD_SUCCESS -eq 0 ]]; then
+            echo -e "${YELLOW}[WARN]${NC} Failed to download OpenCode binary - using Binary-Free Mode"
+            BINARY_FREE_MODE=1
+        else
+            if curl -L --fail -o "$BIN_DIR/checksums.txt" "$CHECKSUM_URL" 2>/dev/null; then
+                EXPECTED_HASH=$(grep "opencode-darwin-${ARCH}" "$BIN_DIR/checksums.txt" | awk '{print $1}')
+                ACTUAL_HASH=$(shasum -a 256 "$BIN_DIR/opencode.tmp" | awk '{print $1}')
+                
+                if [[ "$ACTUAL_HASH" == "$EXPECTED_HASH" ]]; then
+                    echo -e "${GREEN}[OK]${NC} Checksum verified"
+                else
+                    echo -e "${YELLOW}[WARN]${NC} Checksum mismatch (may be OK for some versions)"
+                fi
+            fi
+            
+            mv "$BIN_DIR/opencode.tmp" "$BIN_DIR/opencode"
+            chmod +x "$BIN_DIR/opencode"
+            echo -e "${GREEN}[OK]${NC} OpenCode binary installed"
+        fi
    fi
 fi

 echo ""
-echo "[STEP 6/9] Building UI assets"
+echo "[STEP 6/8] Building UI assets"
 if [[ -d "$SCRIPT_DIR/packages/ui/dist" ]]; then
    echo -e "${GREEN}[OK]${NC} UI build already exists"
 else
@@ -182,7 +231,7 @@ else
 fi

 echo ""
-echo "[STEP 7/9] Post-install health check"
+echo "[STEP 7/8] Post-install health check"
 HEALTH_ERRORS=0

 [[ -f "$SCRIPT_DIR/package.json" ]] || HEALTH_ERRORS=$((HEALTH_ERRORS+1))
@@ -198,24 +247,34 @@ else
 fi

 echo ""
-echo "[STEP 8/9] Installation Summary"
+echo "[STEP 8/8] Installation Summary"
 echo ""
 echo "  Install Dir: $TARGET_DIR"
 echo "  Architecture: $ARCH"
 echo "  Node.js: $NODE_VERSION"
 echo "  npm: $NPM_VERSION"
+if [[ $BINARY_FREE_MODE -eq 1 ]]; then
+    echo "  Mode: Binary-Free Mode (OpenCode Zen free models available)"
+else
+    echo "  Mode: Full Mode (OpenCode binary installed)"
+fi
 echo "  Errors: $ERRORS"
 echo "  Warnings: $WARNINGS"
 echo "  Log File: $LOG_FILE"
 echo ""

-echo "[STEP 9/9] Next steps"
 if [[ $ERRORS -gt 0 ]]; then
    echo -e "${RED}[RESULT]${NC} Installation completed with errors"
    echo "Review $LOG_FILE for details."
 else
    echo -e "${GREEN}[RESULT]${NC} Installation completed successfully"
    echo "Run: ./Launch-Unix.sh"
+    echo ""
+    if [[ $BINARY_FREE_MODE -eq 1 ]]; then
+        echo -e "${BLUE}NOTE:${NC} Running in Binary-Free Mode."
+        echo "      Free models (GPT-5 Nano, Grok Code, GLM-4.7, etc.) are available."
+        echo "      You can also authenticate with Qwen for additional models."
+    fi
 fi

 exit $ERRORS
--- a/Install-Windows.bat
+++ b/Install-Windows.bat
@@ -5,7 +5,7 @@ title NomadArch Installer

 echo.
 echo NomadArch Installer (Windows)
-echo Version: 0.4.0
+echo Version: 0.5.0 - Binary-Free Mode
 echo.

 set SCRIPT_DIR=%~dp0
@@ -21,7 +21,7 @@ set NEEDS_FALLBACK=0

 echo [%date% %time%] Installer started >> "%LOG_FILE%"

-echo [STEP 1/9] OS and Architecture Detection
+echo [STEP 1/8] OS and Architecture Detection
 wmic os get osarchitecture | findstr /i "64-bit" >nul
 if %ERRORLEVEL% equ 0 (
    set ARCH=x64
@@ -31,7 +31,7 @@ if %ERRORLEVEL% equ 0 (
 echo [OK] Architecture: %ARCH%

 echo.
-echo [STEP 2/9] Checking write permissions
+echo [STEP 2/8] Checking write permissions
 if not exist "%BIN_DIR%" mkdir "%BIN_DIR%"
 if not exist "%TEMP_DIR%" mkdir "%TEMP_DIR%"

@@ -61,7 +61,7 @@ if %ERRORLEVEL% neq 0 (
 )

 echo.
-echo [STEP 3/9] Ensuring system dependencies
+echo [STEP 3/8] Ensuring system dependencies

 set WINGET_AVAILABLE=0
 where winget >nul 2>&1 && set WINGET_AVAILABLE=1
@@ -129,7 +129,7 @@ if %ERRORLEVEL% neq 0 (
 )

 echo.
-echo [STEP 4/9] Installing npm dependencies
+echo [STEP 4/8] Installing npm dependencies
 cd /d "%SCRIPT_DIR%"
 echo [%date% %time%] Running npm install >> "%LOG_FILE%"
 call npm install
@@ -142,54 +142,66 @@ if %ERRORLEVEL% neq 0 (
 echo [OK] Dependencies installed

 echo.
-echo [STEP 5/9] Fetching OpenCode binary
+echo [STEP 5/8] OpenCode Binary (OPTIONAL - Binary-Free Mode Available)
+echo [INFO] NomadArch now supports Binary-Free Mode!
+echo [INFO] You can use the application without OpenCode binary.
+echo [INFO] Free models from OpenCode Zen are available without the binary.
 if not exist "%BIN_DIR%" mkdir "%BIN_DIR%"

-for /f "delims=" %%v in ('powershell -NoProfile -Command "(Invoke-WebRequest -UseBasicParsing https://api.github.com/repos/sst/opencode/releases/latest).Content ^| Select-String -Pattern '""tag_name""' ^| ForEach-Object { $_.Line.Split(''\"'')[3] }"') do (
-    set OPENCODE_VERSION=%%v
-)
-
-set OPENCODE_BASE=https://github.com/sst/opencode/releases/download/v!OPENCODE_VERSION!
-set OPENCODE_URL=!OPENCODE_BASE!/opencode-windows-%ARCH%.exe
-set CHECKSUM_URL=!OPENCODE_BASE!/checksums.txt
-
-if exist "%BIN_DIR%\opencode.exe" (
-    echo [OK] OpenCode binary already exists
-    echo [%date% %time%] OpenCode binary exists, skipping download >> "%LOG_FILE%"
+set SKIP_OPENCODE=0
+set /p SKIP_CHOICE="Skip OpenCode binary download? (Y for Binary-Free Mode / N to download) [Y]: "
+if /i "%SKIP_CHOICE%"=="" set SKIP_CHOICE=Y
+if /i "%SKIP_CHOICE%"=="Y" (
+    set SKIP_OPENCODE=1
+    echo [INFO] Skipping OpenCode binary - using Binary-Free Mode
+    echo [%date% %time%] Using Binary-Free Mode >> "%LOG_FILE%"
 ) else (
-    echo [INFO] Downloading OpenCode v!OPENCODE_VERSION!...
-    if "%DOWNLOAD_CMD%"=="curl" (
-        curl -L -o "%BIN_DIR%\opencode.exe.tmp" "!OPENCODE_URL!"
-        curl -L -o "%BIN_DIR%\checksums.txt" "!CHECKSUM_URL!"
+    for /f "delims=" %%v in ('powershell -NoProfile -Command "(Invoke-WebRequest -UseBasicParsing https://api.github.com/repos/sst/opencode/releases/latest).Content ^| Select-String -Pattern '\"\"tag_name\"\"' ^| ForEach-Object { $_.Line.Split(''\"'')[3] }"') do (
+        set OPENCODE_VERSION=%%v
+    )
+
+    set OPENCODE_BASE=https://github.com/sst/opencode/releases/download/v!OPENCODE_VERSION!
+    set OPENCODE_URL=!OPENCODE_BASE!/opencode-windows-%ARCH%.exe
+    set CHECKSUM_URL=!OPENCODE_BASE!/checksums.txt
+
+    if exist "%BIN_DIR%\opencode.exe" (
+        echo [OK] OpenCode binary already exists
+        echo [%date% %time%] OpenCode binary exists, skipping download >> "%LOG_FILE%"
    ) else (
-        powershell -NoProfile -Command "Invoke-WebRequest -Uri '%OPENCODE_URL%' -OutFile '%BIN_DIR%\\opencode.exe.tmp'"
-        powershell -NoProfile -Command "Invoke-WebRequest -Uri '%CHECKSUM_URL%' -OutFile '%BIN_DIR%\\checksums.txt'"
-    )
+        echo [INFO] Downloading OpenCode v!OPENCODE_VERSION!...
+        if "%DOWNLOAD_CMD%"=="curl" (
+            curl -L -o "%BIN_DIR%\opencode.exe.tmp" "!OPENCODE_URL!"
+            curl -L -o "%BIN_DIR%\checksums.txt" "!CHECKSUM_URL!"
+        ) else (
+            powershell -NoProfile -Command "Invoke-WebRequest -Uri '%OPENCODE_URL%' -OutFile '%BIN_DIR%\\opencode.exe.tmp'"
+            powershell -NoProfile -Command "Invoke-WebRequest -Uri '%CHECKSUM_URL%' -OutFile '%BIN_DIR%\\checksums.txt'"
+        )

-    set EXPECTED_HASH=
-    for /f "tokens=1,2" %%h in ('type "%BIN_DIR%\checksums.txt" ^| findstr /i "opencode-windows-%ARCH%"') do (
-        set EXPECTED_HASH=%%h
-    )
+        set EXPECTED_HASH=
+        for /f "tokens=1,2" %%h in ('type "%BIN_DIR%\checksums.txt" ^| findstr /i "opencode-windows-%ARCH%"') do (
+            set EXPECTED_HASH=%%h
+        )

-    set ACTUAL_HASH=
-    for /f "skip=1 tokens=*" %%h in ('certutil -hashfile "%BIN_DIR%\opencode.exe.tmp" SHA256 ^| findstr /v "CertUtil" ^| findstr /v "hash of"') do (
-        set ACTUAL_HASH=%%h
-        goto :hash_found
-    )
-    :hash_found
+        set ACTUAL_HASH=
+        for /f "skip=1 tokens=*" %%h in ('certutil -hashfile "%BIN_DIR%\opencode.exe.tmp" SHA256 ^| findstr /v "CertUtil" ^| findstr /v "hash of"') do (
+            set ACTUAL_HASH=%%h
+            goto :hash_found
+        )
+        :hash_found

-    if "!ACTUAL_HASH!"=="!EXPECTED_HASH!" (
-        move /Y "%BIN_DIR%\opencode.exe.tmp" "%BIN_DIR%\opencode.exe"
-        echo [OK] OpenCode downloaded and verified
-    ) else (
-        echo [ERROR] OpenCode checksum mismatch!
-        del "%BIN_DIR%\opencode.exe.tmp"
-        set /a ERRORS+=1
+        if "!ACTUAL_HASH!"=="!EXPECTED_HASH!" (
+            move /Y "%BIN_DIR%\opencode.exe.tmp" "%BIN_DIR%\opencode.exe"
+            echo [OK] OpenCode downloaded and verified
+        ) else (
+            echo [WARN] OpenCode checksum mismatch - continuing with Binary-Free Mode
+            del "%BIN_DIR%\opencode.exe.tmp" 2>nul
+            set SKIP_OPENCODE=1
+        )
    )
 )

 echo.
-echo [STEP 6/9] Building UI assets
+echo [STEP 6/8] Building UI assets
 if exist "%SCRIPT_DIR%\packages\ui\dist\index.html" (
    echo [OK] UI build already exists
 ) else (
@@ -207,7 +219,7 @@ if exist "%SCRIPT_DIR%\packages\ui\dist\index.html" (
 )

 echo.
-echo [STEP 7/9] Post-install health check
+echo [STEP 7/8] Post-install health check
 set HEALTH_ERRORS=0

 if not exist "%SCRIPT_DIR%\package.json" set /a HEALTH_ERRORS+=1
@@ -223,19 +235,22 @@ if %HEALTH_ERRORS% equ 0 (
 )

 echo.
-echo [STEP 8/9] Installation Summary
+echo [STEP 8/8] Installation Summary
 echo.
 echo   Install Dir: %TARGET_DIR%
 echo   Architecture: %ARCH%
 echo   Node.js: %NODE_VERSION%
 echo   npm: %NPM_VERSION%
+if %SKIP_OPENCODE% equ 1 (
+    echo   Mode: Binary-Free Mode ^(OpenCode Zen free models available^)
+) else (
+    echo   Mode: Full Mode ^(OpenCode binary installed^)
+)
 echo   Errors: %ERRORS%
 echo   Warnings: %WARNINGS%
 echo   Log File: %LOG_FILE%
 echo.

-echo [STEP 9/9] Next steps
-
 :SUMMARY
 if %ERRORS% gtr 0 (
    echo [RESULT] Installation completed with errors.
@@ -245,6 +260,12 @@ if %ERRORS% gtr 0 (
 ) else (
    echo [RESULT] Installation completed successfully.
    echo Run Launch-Windows.bat to start the application.
+    echo.
+    if %SKIP_OPENCODE% equ 1 (
+        echo NOTE: Running in Binary-Free Mode.
+        echo       Free models ^(GPT-5 Nano, Grok Code, GLM-4.7, etc.^) are available.
+        echo       You can also authenticate with Qwen for additional models.
+    )
 )

 echo.
--- a/Launch-Dev-Unix.sh
+++ b/Launch-Dev-Unix.sh
@@ -66,14 +66,46 @@ SERVER_PORT=$DEFAULT_SERVER_PORT
 UI_PORT=$DEFAULT_UI_PORT

 for port in {3001..3050}; do
-    if ! lsof -i :$port -sTCP:LISTEN -t > /dev/null 2>&1; then
+    # Try lsof first, then ss, then netstat
+    if command -v lsof &> /dev/null; then
+        if ! lsof -i :$port -sTCP:LISTEN -t > /dev/null 2>&1; then
+            SERVER_PORT=$port
+            break
+        fi
+    elif command -v ss &> /dev/null; then
+        if ! ss -tuln | grep -q ":$port "; then
+            SERVER_PORT=$port
+            break
+        fi
+    elif command -v netstat &> /dev/null; then
+        if ! netstat -tuln | grep -q ":$port "; then
+            SERVER_PORT=$port
+            break
+        fi
+    else
        SERVER_PORT=$port
        break
    fi
 done

 for port in {3000..3050}; do
-    if ! lsof -i :$port -sTCP:LISTEN -t > /dev/null 2>&1; then
+    # Try lsof first, then ss, then netstat
+    if command -v lsof &> /dev/null; then
+        if ! lsof -i :$port -sTCP:LISTEN -t > /dev/null 2>&1; then
+            UI_PORT=$port
+            break
+        fi
+    elif command -v ss &> /dev/null; then
+        if ! ss -tuln | grep -q ":$port "; then
+            UI_PORT=$port
+            break
+        fi
+    elif command -v netstat &> /dev/null; then
+        if ! netstat -tuln | grep -q ":$port "; then
+            UI_PORT=$port
+            break
+        fi
+    else
        UI_PORT=$port
        break
    fi
--- a/Launch-Unix-Prod.sh
+++ b/Launch-Unix-Prod.sh
@@ -0,0 +1,62 @@
+#!/bin/bash
+
+# NomadArch Production Launcher for macOS and Linux
+# Version: 0.4.0
+
+set -euo pipefail
+
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+BLUE='\033[0;34m'
+NC='\033[0m'
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+cd "$SCRIPT_DIR"
+
+echo ""
+echo "NomadArch Launcher (macOS/Linux, Production Mode)"
+echo "Version: 0.4.0"
+echo "Features: SMART FIX / APEX / SHIELD / MULTIX MODE"
+echo ""
+
+echo "[STEP 1/3] Checking Dependencies..."
+
+if ! command -v node &> /dev/null; then
+    echo -e "${RED}[ERROR]${NC} Node.js not found!"
+    echo "Please run the installer first:"
+    if [[ "$OSTYPE" == "darwin"* ]]; then
+        echo "  ./Install-Mac.sh"
+    else
+        echo "  ./Install-Linux.sh"
+    fi
+    exit 1
+fi
+
+NODE_VERSION=$(node --version)
+echo -e "${GREEN}[OK]${NC} Node.js: $NODE_VERSION"
+
+echo ""
+echo "[STEP 2/3] Checking Pre-Built UI..."
+
+if [[ -d "packages/electron-app/dist/renderer/assets" ]]; then
+    echo -e "${GREEN}[OK]${NC} Pre-built UI assets found"
+else
+    echo -e "${RED}[ERROR]${NC} Pre-built UI assets not found."
+    echo "Run: npm run build"
+    exit 1
+fi
+
+echo ""
+echo "[STEP 3/3] Starting NomadArch (Production Mode)..."
+
+cd packages/electron-app
+npx electron .
+EXIT_CODE=$?
+
+if [[ $EXIT_CODE -ne 0 ]]; then
+    echo ""
+    echo -e "${RED}[ERROR]${NC} NomadArch exited with an error!"
+fi
+
+exit $EXIT_CODE
--- a/Launch-Unix.sh
+++ b/Launch-Unix.sh
@@ -1,7 +1,7 @@
 #!/bin/bash

 # NomadArch Launcher for macOS and Linux
-# Version: 0.4.0
+# Version: 0.5.0 - Binary-Free Mode

 set -euo pipefail

@@ -17,10 +17,11 @@ cd "$SCRIPT_DIR"
 ERRORS=0
 WARNINGS=0
 AUTO_FIXED=0
+BINARY_FREE_MODE=0

 echo ""
 echo "NomadArch Launcher (macOS/Linux)"
-echo "Version: 0.4.0"
+echo "Version: 0.5.0 - Binary-Free Mode"
 echo ""

 echo "[PREFLIGHT 1/7] Checking Dependencies..."
@@ -48,16 +49,16 @@ NPM_VERSION=$(npm --version)
 echo -e "${GREEN}[OK]${NC} npm: $NPM_VERSION"

 echo ""
-echo "[PREFLIGHT 2/7] Checking for OpenCode CLI..."
+echo "[PREFLIGHT 2/7] Checking for OpenCode CLI (Optional)..."

 if command -v opencode &> /dev/null; then
-    echo -e "${GREEN}[OK]${NC} OpenCode CLI available in PATH"
+    echo -e "${GREEN}[OK]${NC} OpenCode CLI available in PATH - Full Mode"
 elif [[ -f "$SCRIPT_DIR/bin/opencode" ]]; then
-    echo -e "${GREEN}[OK]${NC} OpenCode binary found in bin/"
+    echo -e "${GREEN}[OK]${NC} OpenCode binary found in bin/ - Full Mode"
 else
-    echo -e "${YELLOW}[WARN]${NC} OpenCode CLI not found"
-    echo "[INFO] Run Install-*.sh to set up OpenCode"
-    ((WARNINGS++))
+    echo -e "${BLUE}[INFO]${NC} OpenCode CLI not found - Using Binary-Free Mode"
+    echo -e "${BLUE}[INFO]${NC} Free models (GPT-5 Nano, Grok Code, GLM-4.7) available via OpenCode Zen"
+    BINARY_FREE_MODE=1
 fi

 echo ""
@@ -84,7 +85,24 @@ SERVER_PORT=$DEFAULT_SERVER_PORT
 UI_PORT=$DEFAULT_UI_PORT

 for port in {3001..3050}; do
-    if ! lsof -i :$port -sTCP:LISTEN -t > /dev/null 2>&1; then
+    # Try lsof first, then ss, then netstat
+    if command -v lsof &> /dev/null; then
+        if ! lsof -i :$port -sTCP:LISTEN -t > /dev/null 2>&1; then
+            SERVER_PORT=$port
+            break
+        fi
+    elif command -v ss &> /dev/null; then
+        if ! ss -tuln | grep -q ":$port "; then
+            SERVER_PORT=$port
+            break
+        fi
+    elif command -v netstat &> /dev/null; then
+        if ! netstat -tuln | grep -q ":$port "; then
+            SERVER_PORT=$port
+            break
+        fi
+    else
+        # No port checking tools, just use default
        SERVER_PORT=$port
        break
    fi
@@ -133,6 +151,12 @@ echo -e "${BLUE}[STATUS]${NC}"
 echo ""
 echo "  Node.js: $NODE_VERSION"
 echo "  npm: $NPM_VERSION"
+if [[ $BINARY_FREE_MODE -eq 1 ]]; then
+    echo "  Mode: Binary-Free Mode (No OpenCode binary required)"
+    echo "  Free Models: GPT-5 Nano, Grok Code, GLM-4.7, Doubao, Big Pickle"
+else
+    echo "  Mode: Full Mode (OpenCode binary available)"
+fi
 echo "  Auto-fixes applied: $AUTO_FIXED"
 echo "  Warnings: $WARNINGS"
 echo "  Errors: $ERRORS"
@@ -158,6 +182,7 @@ elif [[ "$OSTYPE" == "linux-gnu"* ]]; then
 fi

 export CLI_PORT=$SERVER_PORT
+export NOMADARCH_BINARY_FREE_MODE=$BINARY_FREE_MODE
 npm run dev:electron

 EXIT_CODE=$?
--- a/Launch-Windows.bat
+++ b/Launch-Windows.bat
@@ -6,7 +6,7 @@ color 0A

 echo.
 echo NomadArch Launcher (Windows)
-echo Version: 0.4.0
+echo Version: 0.5.0 - Binary-Free Mode
 echo.

 set SCRIPT_DIR=%~dp0
@@ -16,6 +16,7 @@ cd /d "%SCRIPT_DIR%"
 set ERRORS=0
 set WARNINGS=0
 set AUTO_FIXED=0
+set BINARY_FREE_MODE=0

 echo [PREFLIGHT 1/7] Checking Dependencies...

@@ -42,18 +43,18 @@ for /f "tokens=*" %%i in ('npm --version') do set NPM_VERSION=%%i
 echo [OK] npm: %NPM_VERSION%

 echo.
-echo [PREFLIGHT 2/7] Checking for OpenCode CLI...
+echo [PREFLIGHT 2/7] Checking OpenCode CLI (Optional)...

 where opencode >nul 2>&1
 if %ERRORLEVEL% equ 0 (
-    echo [OK] OpenCode CLI available in PATH
+    echo [OK] OpenCode CLI available in PATH - Full Mode
 ) else (
    if exist "bin\opencode.exe" (
-        echo [OK] OpenCode binary found in bin/
+        echo [OK] OpenCode binary found in bin/ - Full Mode
    ) else (
-        echo [WARN] OpenCode CLI not found
-        echo [INFO] Run Install-Windows.bat to set up OpenCode
-        set /a WARNINGS+=1
+        echo [INFO] OpenCode CLI not found - Using Binary-Free Mode
+        echo [INFO] Free models (GPT-5 Nano, Grok Code, GLM-4.7) available via OpenCode Zen
+        set BINARY_FREE_MODE=1
    )
 )

@@ -139,6 +140,12 @@ echo [STATUS]
 echo.
 echo   Node.js: %NODE_VERSION%
 echo   npm: %NPM_VERSION%
+if %BINARY_FREE_MODE% equ 1 (
+    echo   Mode: Binary-Free Mode ^(No OpenCode binary required^)
+    echo   Free Models: GPT-5 Nano, Grok Code, GLM-4.7, Doubao, Big Pickle
+) else (
+    echo   Mode: Full Mode ^(OpenCode binary available^)
+)
 echo   Auto-fixes applied: !AUTO_FIXED!
 echo   Warnings: %WARNINGS%
 echo   Errors: %ERRORS%
@@ -181,6 +188,7 @@ echo ========================================

 set "VITE_DEV_SERVER_URL=http://localhost:!UI_PORT!"
 set "NOMADARCH_OPEN_DEVTOOLS=false"
+set "NOMADARCH_BINARY_FREE_MODE=%BINARY_FREE_MODE%"
 call npm run dev:electron

 if %ERRORLEVEL% neq 0 (
--- a/Prepare-Public-Release.bat
+++ b/Prepare-Public-Release.bat
@@ -0,0 +1,152 @@
+@echo off
+setlocal enabledelayedexpansion
+
+:: =====================================================
+:: NomadArch - Clean Copy Script for Public Release
+:: Creates a sanitized copy without sensitive data
+:: =====================================================
+
+title NomadArch Clean Copy for GitHub
+
+echo.
+echo =====================================================
+echo   NomadArch - Prepare Clean Public Release
+echo =====================================================
+echo.
+
+set SCRIPT_DIR=%~dp0
+set SCRIPT_DIR=%SCRIPT_DIR:~0,-1%
+set DEST_DIR=%USERPROFILE%\Desktop\NomadArch-Public-Release
+
+echo [INFO] Source: %SCRIPT_DIR%
+echo [INFO] Destination: %DEST_DIR%
+echo.
+
+if exist "%DEST_DIR%" (
+    echo [WARN] Destination exists. Removing old copy...
+    rmdir /s /q "%DEST_DIR%"
+)
+
+echo [STEP 1/6] Creating destination directory...
+mkdir "%DEST_DIR%"
+
+echo [STEP 2/6] Copying core project files...
+
+:: Copy essential files
+copy "%SCRIPT_DIR%\package.json" "%DEST_DIR%\" >nul
+copy "%SCRIPT_DIR%\package-lock.json" "%DEST_DIR%\" >nul
+copy "%SCRIPT_DIR%\.gitignore" "%DEST_DIR%\" >nul
+copy "%SCRIPT_DIR%\README.md" "%DEST_DIR%\" >nul
+copy "%SCRIPT_DIR%\BUILD.md" "%DEST_DIR%\" >nul
+copy "%SCRIPT_DIR%\AGENTS.md" "%DEST_DIR%\" >nul
+copy "%SCRIPT_DIR%\PROGRESS.md" "%DEST_DIR%\" >nul
+copy "%SCRIPT_DIR%\manual_test_guide.md" "%DEST_DIR%\" >nul
+
+:: Copy launchers and installers
+copy "%SCRIPT_DIR%\Install-*.bat" "%DEST_DIR%\" >nul
+copy "%SCRIPT_DIR%\Install-*.sh" "%DEST_DIR%\" >nul
+copy "%SCRIPT_DIR%\Launch-*.bat" "%DEST_DIR%\" >nul
+copy "%SCRIPT_DIR%\Launch-*.sh" "%DEST_DIR%\" >nul
+
+echo [STEP 3/6] Copying packages directory (source only)...
+
+:: Use robocopy to exclude unwanted items
+robocopy "%SCRIPT_DIR%\packages" "%DEST_DIR%\packages" /E /NFL /NDL /NJH /NJS /NC /NS ^
+    /XD node_modules dist out release .vite .electron-vite _backup_original __pycache__ ^
+    /XF *.log *.bak *.tmp *.map
+
+echo [STEP 4/6] Copying additional directories...
+
+:: Copy docs if exists
+if exist "%SCRIPT_DIR%\docs" (
+    robocopy "%SCRIPT_DIR%\docs" "%DEST_DIR%\docs" /E /NFL /NDL /NJH /NJS /NC /NS /XD node_modules
+)
+
+:: Copy images if exists
+if exist "%SCRIPT_DIR%\images" (
+    robocopy "%SCRIPT_DIR%\images" "%DEST_DIR%\images" /E /NFL /NDL /NJH /NJS /NC /NS
+)
+
+:: Copy dev-docs if exists
+if exist "%SCRIPT_DIR%\dev-docs" (
+    robocopy "%SCRIPT_DIR%\dev-docs" "%DEST_DIR%\dev-docs" /E /NFL /NDL /NJH /NJS /NC /NS
+)
+
+:: Copy scripts directory if exists
+if exist "%SCRIPT_DIR%\scripts" (
+    robocopy "%SCRIPT_DIR%\scripts" "%DEST_DIR%\scripts" /E /NFL /NDL /NJH /NJS /NC /NS
+)
+
+:: Copy .github directory (workflows, templates)
+if exist "%SCRIPT_DIR%\.github" (
+    robocopy "%SCRIPT_DIR%\.github" "%DEST_DIR%\.github" /E /NFL /NDL /NJH /NJS /NC /NS
+)
+
+echo [STEP 5/6] Creating clean MCP config template...
+
+:: Create a template .mcp.json with placeholders
+(
+echo {
+echo     "mcpServers": {
+echo         "sequential-thinking": {
+echo             "command": "npx",
+echo             "args": ["-y", "@modelcontextprotocol/server-sequential-thinking"]
+echo         },
+echo         "desktop-commander": {
+echo             "command": "npx",
+echo             "args": ["-y", "@modelcontextprotocol/server-desktop-commander"]
+echo         },
+echo         "web-reader": {
+echo             "command": "npx",
+echo             "args": ["-y", "@modelcontextprotocol/server-web-reader"]
+echo         },
+echo         "github": {
+echo             "command": "npx",
+echo             "args": ["-y", "@modelcontextprotocol/server-github"],
+echo             "env": {
+echo                 "GITHUB_TOKEN": "YOUR_GITHUB_TOKEN_HERE"
+echo             }
+echo         }
+echo     }
+echo }
+) > "%DEST_DIR%\.mcp.json.example"
+
+echo [STEP 6/6] Final cleanup...
+
+:: Remove any accidentally copied sensitive files
+if exist "%DEST_DIR%\.opencode" rmdir /s /q "%DEST_DIR%\.opencode"
+if exist "%DEST_DIR%\.trae" rmdir /s /q "%DEST_DIR%\.trae"
+if exist "%DEST_DIR%\.backup" rmdir /s /q "%DEST_DIR%\.backup"
+if exist "%DEST_DIR%\.tmp-qwen-code" rmdir /s /q "%DEST_DIR%\.tmp-qwen-code"
+if exist "%DEST_DIR%\.agent" rmdir /s /q "%DEST_DIR%\.agent"
+if exist "%DEST_DIR%\install.log" del "%DEST_DIR%\install.log"
+
+:: Delete any .bak files that got through
+for /r "%DEST_DIR%" %%f in (*.bak) do del "%%f" 2>nul
+for /r "%DEST_DIR%" %%f in (*.log) do del "%%f" 2>nul
+for /r "%DEST_DIR%" %%f in (*.tmp) do del "%%f" 2>nul
+
+:: Remove _backup_original directories
+for /d /r "%DEST_DIR%" %%d in (_backup_original) do (
+    if exist "%%d" rmdir /s /q "%%d"
+)
+
+echo.
+echo =====================================================
+echo   Clean Copy Complete!
+echo =====================================================
+echo.
+echo   Location: %DEST_DIR%
+echo.
+echo   Next Steps:
+echo   1. Review the contents of %DEST_DIR%
+echo   2. cd %DEST_DIR%
+echo   3. git init
+echo   4. git add .
+echo   5. git commit -m "Initial public release"
+echo   6. git remote add origin https://github.com/YOUR_USER/NomadArch.git
+echo   7. git push -u origin main
+echo.
+echo =====================================================
+
+pause
--- a/README.md
+++ b/README.md
@@ -1,565 +1,347 @@
-<!--
-NomadArch - Advanced AI Coding Workspace
-SEO Optimized: AI coding assistant, multi-model support, GLM 4.7, Z.AI API, autonomous coding, TypeScript, Electron
-->
-<meta name="description" content="NomadArch - Advanced AI-powered coding workspace with multi-model support including GLM 4.7, Anthropic Claude, OpenAI GPT, and local Ollama models. Autonomous coding, real-time streaming, and intelligent code fixes.">
-<meta name="keywords" content="AI coding assistant, GLM 4.7, Z.AI API, multi-model AI, autonomous coding, code generation, TypeScript, Electron, SolidJS, OpenAI, Anthropic, Qwen, Ollama">
-<meta name="author" content="NeuralNomadsAI">
-<meta name="robots" content="index, follow">
-
-<meta property="og:title" content="NomadArch - Advanced AI Coding Workspace with GLM 4.7">
-<meta property="og:description" content="Multi-model AI coding assistant featuring GLM 4.7, Claude, GPT, and local models. Autonomous coding, real-time streaming, intelligent fixes.">
-<meta property="og:image" content="https://github.com/roman-ryzenadvanced/NomadArch-v1.0/raw/main/packages/ui/src/images/CodeNomad-Icon.png">
-<meta property="og:type" content="website">
-<meta property="og:url" content="https://github.com/roman-ryzenadvanced/NomadArch-v1.0">
-
-<meta name="twitter:card" content="summary_large_image">
-<meta name="twitter:title" content="NomadArch - Advanced AI Coding Workspace">
-<meta name="twitter:description" content="Multi-model AI coding assistant featuring GLM 4.7, Claude, GPT, and local models.">
-<meta name="twitter:image" content="https://github.com/roman-ryzenadvanced/NomadArch-v1.0/raw/main/packages/ui/src/images/CodeNomad-Icon.png">
-
-<script type="application/ld+json">
-{
-  "@context": "https://schema.org",
-  "@type": "SoftwareApplication",
-  "name": "NomadArch",
-  "operatingSystem": "Windows, macOS, Linux",
-  "applicationCategory": "DeveloperApplication",
-  "description": "Advanced AI-powered coding workspace with multi-model support including GLM 4.7, Anthropic Claude, OpenAI GPT, and local Ollama models",
-  "author": {
-    "@type": "Organization",
-    "name": "NeuralNomadsAI"
-  },
-  "offers": {
-    "@type": "Offer",
-    "price": "0",
-    "priceCurrency": "USD"
-  },
-  "featureList": [
-    "Multi-provider AI support",
-    "GLM 4.7 integration via Z.AI API",
-    "Autonomous coding with APEX mode",
-    "Real-time token streaming",
-    "Intelligent code fixes",
-    "Ollama local model support"
-  ],
-  "softwareVersion": "1.0.0"
-}
-</script>
-
-# NomadArch
-
 <p align="center">
  <img src="packages/ui/src/images/CodeNomad-Icon.png" alt="NomadArch Logo" width="180" height="180">
 </p>

-<h3 align="center">NomadArch - Advanced AI Coding Workspace</h3>
+<h1 align="center">🏛️ NomadArch</h1>
+
+<h3 align="center">Advanced AI Coding Workspace</h3>

 <p align="center">
-  <strong>Fork of CodeNomad by OpenCode</strong>
+  <em>NomadArch is an enhanced fork of CodeNomad — now with GLM 4.7, multi-model support, and MULTIX Mode</em>
 </p>

 <p align="center">
  <a href="https://github.com/roman-ryzenadvanced/NomadArch-v1.0/stargazers">
-    <img src="https://img.shields.io/github/stars/roman-ryzenadvanced/NomadArch-v1.0?style=social" alt="GitHub Stars">
+    <img src="https://img.shields.io/github/stars/roman-ryzenadvanced/NomadArch-v1.0?style=for-the-badge&logo=github&logoColor=white&color=gold" alt="GitHub Stars">
  </a>
  <a href="https://github.com/roman-ryzenadvanced/NomadArch-v1.0/network/members">
-    <img src="https://img.shields.io/github/forks/roman-ryzenadvanced/NomadArch-v1.0?style=social" alt="GitHub Forks">
+    <img src="https://img.shields.io/github/forks/roman-ryzenadvanced/NomadArch-v1.0?style=for-the-badge&logo=git&logoColor=white&color=blue" alt="GitHub Forks">
  </a>
  <a href="https://github.com/roman-ryzenadvanced/NomadArch-v1.0/blob/main/LICENSE">
-    <img src="https://img.shields.io/github/license/roman-ryzenadvanced/NomadArch-v1.0" alt="License">
-  </a>
-  <a href="https://github.com/roman-ryzenadvanced/NomadArch-v1.0/releases">
-    <img src="https://img.shields.io/github/v/release/roman-ryzenadvanced/NomadArch-v1.0" alt="Latest Release">
+    <img src="https://img.shields.io/github/license/roman-ryzenadvanced/NomadArch-v1.0?style=for-the-badge&color=green" alt="License">
  </a>
 </p>

 <p align="center">
-  <a href="#features">Features</a> •
-  <a href="#supported-ai-models">AI Models</a> •
-  <a href="#installation">Installation</a> •
-  <a href="#usage">Usage</a> •
-  <a href="#whats-new">What's New</a> •
-  <a href="#credits">Credits</a>
+  <a href="#-features">Features</a> •
+  <a href="#-supported-ai-models">AI Models</a> •
+  <a href="#-installation">Installation</a> •
+  <a href="#-usage">Usage</a> •
+  <a href="#-whats-new">What's New</a> •
+  <a href="#-credits">Credits</a>
 </p>

 <p align="center">
  <a href="https://github.com/roman-ryzenadvanced/NomadArch-v1.0">
-    <img src="https://img.shields.io/badge/Star%20this%20repo-%E2%AD%90-yellow?style=for-the-badge" alt="Star this repo">
+    <img src="https://img.shields.io/badge/⭐_Star_this_repo-yellow?style=for-the-badge" alt="Star this repo">
  </a>
 </p>

 ---

-## Overview
+## 🎯 Overview

-NomadArch is an enhanced fork of CodeNomad by OpenCode, featuring significant UI/UX improvements, additional AI integrations, and a more robust architecture. This is a full-featured AI coding assistant with support for multiple AI providers including **GLM 4.7**, Anthropic, OpenAI, Google, Qwen, and local models via Ollama.
+**NomadArch** is an enhanced fork of CodeNomad, featuring significant UI/UX improvements, additional AI integrations, and a more robust architecture. This is a full-featured AI coding assistant with support for multiple AI providers including **GLM 4.7**, Anthropic, OpenAI, Google, Qwen, and local models via Ollama.

-### Key Improvements Over CodeNomad
- Fixed Qwen OAuth authentication flow
- Enhanced MULTIX Mode with live token streaming
- Improved UI/UX with detailed tooltips
- Auto-build verification on launch
- Comprehensive installer scripts for all platforms
- Port conflict detection and resolution hints
+### ✨ Key Improvements Over CodeNomad
+- 🔧 Fixed Qwen OAuth authentication flow
+- 🚀 Enhanced MULTIX Mode with live token streaming
+- 🎨 Improved UI/UX with detailed tooltips
+- ✅ Auto-build verification on launch
+- 📦 Comprehensive installer scripts for all platforms
+- 🔌 Port conflict detection and resolution hints
+- 🆓 **NEW: Binary-Free Mode** - No external binaries required!
+
+### 🆓 Binary-Free Mode (v0.5.0)
+
+NomadArch now works **without requiring the OpenCode binary**! This means:
+
+| Benefit | Description |
+|---------|-------------|
+| ⚡ **Faster Setup** | No binary downloads, just npm install |
+| 🌍 **Universal** | Works on all platforms without platform-specific binaries |
+| 🆓 **Free Models** | Access free AI models without any binary |
+| 🔄 **Seamless** | Automatically uses native mode when binary unavailable |
+
+**Free Models Available (No API Key Required):**
+- 🧠 **GPT-5 Nano** - 400K context, reasoning + tools
+- ⚡ **Grok Code Fast 1** - 256K context, optimized for code
+- 🌟 **GLM-4.7** - 205K context, top-tier performance
+- 🚀 **Doubao Seed Code** - 256K context, specialized for coding
+- 🥒 **Big Pickle** - 200K context, efficient and fast

 ---

-## Supported AI Models & Providers
+## 🤖 Supported AI Models

 NomadArch supports a wide range of AI models from multiple providers, giving you flexibility to choose the best model for your coding tasks.

-### 🚀 Featured Model: GLM 4.7 (Z.AI)
+### 🔥 Featured Model: GLM 4.7 (Z.AI)

 **GLM 4.7** is the latest state-of-the-art open model from Z.AI, now fully integrated into NomadArch. Released in December 2025, GLM 4.7 ranks **#1 for Web Development** and **#6 overall** on the LM Arena leaderboard.

-#### Key Features
- 🔥 **128K Context Window** - Process entire codebases in a single session
- 🧠 **Interleaved Thinking** - Advanced reasoning with multi-step analysis
- 💭 **Preserved Thinking** - Maintains reasoning chain across long conversations
- 🔄 **Turn-level Thinking** - Optimized per-response reasoning for efficiency
+| Feature | Description |
+|---------|-------------|
+| 📊 **128K Context Window** | Process entire codebases in a single session |
+| 🧠 **Interleaved Thinking** | Advanced reasoning with multi-step analysis |
+| 💭 **Preserved Thinking** | Maintains reasoning chain across long conversations |
+| 🔄 **Turn-level Thinking** | Optimized per-response reasoning for efficiency |

 #### Benchmark Performance
-| Benchmark | Score | Improvement |
-|-----------|-------|-------------|
+
+| Benchmark | Score | Notes |
+|-----------|-------|-------|
 | SWE-bench | **+73.8%** | Over GLM-4.6 |
 | SWE-bench Multilingual | **+66.7%** | Over GLM-4.6 |
 | Terminal Bench 2.0 | **+41%** | Over GLM-4.6 |
 | LM Arena WebDev | **#1** | Open Model Ranking |
 | LM Arena Overall | **#6** | Open Model Ranking |

-GLM 4.7 beats GPT-5, Claude Sonnet, and Gemini on multiple coding benchmarks.
-
-#### Z.AI API Integration
- ✅ Fully integrated via Z.AI Plan API
- ✅ Compatible with Claude Code, Cline, Roo Code, Kilo Code
- ✅ Get **10% discount** with code: [`R0K78RJKNW`](https://z.ai/subscribe?ic=R0K78RJKNW)
- 🎯 [Subscribe to Z.AI with 10% off](https://z.ai/subscribe?ic=R0K78RJKNW)
+> 🎯 **Get 10% discount on Z.AI with code: [`R0K78RJKNW`](https://z.ai/subscribe?ic=R0K78RJKNW)**

 ---

-### 🤖 All Supported Models
+### 📋 All Supported Models
+
+<details>
+<summary><b>🌟 Z.AI Models</b></summary>

-#### Z.AI
 | Model | Context | Specialty |
 |-------|---------|-----------|
 | **GLM 4.7** | 128K | Web Development, Coding |
 | GLM 4.6 | 128K | General Coding |
 | GLM-4 | 128K | Versatile |

-#### Anthropic
+</details>
+
+<details>
+<summary><b>🟣 Anthropic Models</b></summary>
+
 | Model | Context | Specialty |
 |-------|---------|-----------|
 | Claude 3.7 Sonnet | 200K | Complex Reasoning |
 | Claude 3.5 Sonnet | 200K | Balanced Performance |
 | Claude 3 Opus | 200K | Maximum Quality |

-#### OpenAI
+</details>
+
+<details>
+<summary><b>🟢 OpenAI Models</b></summary>
+
 | Model | Context | Specialty |
 |-------|---------|-----------|
 | GPT-5 Preview | 200K | Latest Capabilities |
 | GPT-4.1 | 128K | Production Ready |
 | GPT-4 Turbo | 128K | Fast & Efficient |

-#### Google
+</details>
+
+<details>
+<summary><b>🔵 Google Models</b></summary>
+
 | Model | Context | Specialty |
 |-------|---------|-----------|
 | Gemini 2.0 Pro | 1M+ | Massive Context |
 | Gemini 2.0 Flash | 1M+ | Ultra Fast |

-#### Qwen
-| Model | Context | Specialty |
-|-------|---------|-----------|
+</details>
+
+<details>
+<summary><b>🟠 Qwen & Local Models</b></summary>
+
+| Model | Context/Size | Specialty |
+|-------|--------------|-----------|
 | Qwen 2.5 Coder | 32K | Code Specialized |
 | Qwen 2.5 | 32K | General Purpose |
+| DeepSeek Coder (Ollama) | Varies | Code |
+| Llama 3.1 (Ollama) | Varies | General |

-#### Local (Ollama)
-| Model | Size | Specialty |
-|-------|------|-----------|
-| DeepSeek Coder | Varies | Code |
-| Llama 3.1 | Varies | General |
-| CodeLlama | Varies | Code |
-| Mistral | Varies | General |
+</details>

 ---

-## Installation
+## 📦 Installation

 ### Quick Start (Recommended)

-The installers will automatically install **OpenCode CLI** (required for workspace functionality) using:
-1. **Primary**: `npm install -g opencode-ai@latest` (fastest)
-2. **Fallback**: Download from official GitHub releases if npm fails
-
 #### Windows
 ```batch
-# Double-click and run
 Install-Windows.bat
-
-# Then start app
 Launch-Windows.bat
 ```

 #### Linux
 ```bash
-chmod +x Install-Linux.sh
-./Install-Linux.sh
-
-# Then start app
+chmod +x Install-Linux.sh && ./Install-Linux.sh
 ./Launch-Unix.sh
 ```

 #### macOS
 ```bash
-chmod +x Install-Mac.sh
-./Install-Mac.sh
-
-# Then start app
+chmod +x Install-Mac.sh && ./Install-Mac.sh
 ./Launch-Unix.sh
 ```

 ### Manual Installation

 ```bash
-# Clone the repository
 git clone https://github.com/roman-ryzenadvanced/NomadArch-v1.0.git
 cd NomadArch
-
-# Install dependencies
 npm install
-
-# Start the application
 npm run dev:electron
 ```

-### Building from Source
-
-```bash
-# Build all packages
-npm run build
-
-# Or build individual packages
-npm run build:ui          # Build UI
-npm run build:server       # Build server
-npm run build:electron     # Build Electron app
-```
-
 ---

-## Features
+## 🚀 Features

 ### Core Features
- 🤖 **Multi-Provider AI Support** - GLM 4.7, Anthropic, OpenAI, Google, Qwen, Ollama (local)
- 🖥️ **Electron Desktop App** - Native feel with modern web technologies
- 📁 **Workspace Management** - Organize your projects efficiently
- 💬 **Real-time Streaming** - Live responses from AI models
- 🔧 **Smart Fix** - AI-powered code error detection and fixes
- 🏗️ **Build Integration** - One-click project builds
- 🔌 **Ollama Integration** - Run local AI models for privacy
+| Feature | Description |
+|---------|-------------|
+| 🤖 **Multi-Provider AI** | GLM 4.7, Anthropic, OpenAI, Google, Qwen, Ollama |
+| 🖥️ **Electron Desktop App** | Native feel with modern web technologies |
+| 📁 **Workspace Management** | Organize your projects efficiently |
+| 💬 **Real-time Streaming** | Live responses from AI models |
+| 🔧 **Smart Fix** | AI-powered code error detection and fixes |
+| 🔌 **Ollama Integration** | Run local AI models for privacy |

 ### UI/UX Highlights
- ⚡ **MULTIX Mode** - Multi-task parallel AI conversations with live token counting
- 🛡️ **SHIELD Mode** - Auto-approval for hands-free operation
- 🚀 **APEX Mode** - Autonomous AI that chains tasks together
- 📊 **Live Token Counter** - Real-time token usage during streaming
- 💭 **Thinking Indicator** - Animated visual feedback when AI is processing
- 🎨 **Modern Dark Theme** - Beautiful, eye-friendly dark interface
- 🖱️ **Detailed Tooltips** - Hover over any button for explanations
+| Mode | Description |
+|------|-------------|
+| ⚡ **MULTIX Mode** | Multi-task parallel AI conversations with live token counting |
+| 🛡️ **SHIELD Mode** | Auto-approval for hands-free operation |
+| 🚀 **APEX Mode** | Autonomous AI that chains tasks together |

 ---

-## What's New in NomadArch
+## 🆕 What's New

-### Major Improvements Over Original CodeNomad
+<details>
+<summary><b>🎨 Branding & Identity</b></summary>

-#### 🎨 Branding & Identity
 - ✅ **New Branding**: "NomadArch" with proper attribution to OpenCode
 - ✅ **Updated Loading Screen**: New branding with fork attribution
 - ✅ **Updated Empty States**: All screens show NomadArch branding

-#### 🔐 Qwen OAuth Integration
- ✅ **Fixed OAuth Flow**: Resolved "Body cannot be empty" error in Qwen authentication
- ✅ **Proper API Bodies**: POST requests now include proper JSON bodies
- ✅ **Fixed Device Poll Schema**: Corrected Fastify schema validation for OAuth polling
+</details>
+
+<details>
+<summary><b>🔐 Qwen OAuth Integration</b></summary>
+
+- ✅ **Fixed OAuth Flow**: Resolved "Body cannot be empty" error
+- ✅ **Proper API Bodies**: POST requests now include proper JSON bodies
+- ✅ **Fixed Device Poll Schema**: Corrected Fastify schema validation
+
+</details>
+
+<details>
+<summary><b>🚀 MULTIX Mode Enhancements</b></summary>

-#### 🚀 MULTIX Mode Enhancements
 - ✅ **Live Streaming Token Counter**: Visible in header during AI processing
 - ✅ **Thinking Roller Indicator**: Animated indicator with bouncing dots
 - ✅ **Token Stats Display**: Shows input/output tokens processed
 - ✅ **Auto-Scroll**: Intelligent scrolling during streaming

-#### 🖥️ UI/UX Improvements
- ✅ **Detailed Button Tooltips**: Hover over any button for detailed explanations
-  - AUTHED: Authentication status explanation
-  - AI MODEL: Model selection help
-  - SMART FIX: AI code analysis feature
-  - BUILD: Project compilation
-  - APEX: Autonomous mode description
-  - SHIELD: Auto-approval mode
-  - MULTIX MODE: Multi-task interface
- ✅ **Bulletproof Layout**: Fixed layout issues with Editor/MultiX panels
- ✅ **Overflow Handling**: Long code lines don't break layout
- ✅ **Responsive Panels**: Editor and chat panels properly sized
+</details>

-#### 📂 File Editor Improvements
- ✅ **Proper File Loading**: Files load correctly when selected in explorer
- ✅ **Line Numbers**: Clean line number display
- ✅ **Word Wrap**: Long lines wrap instead of overflowing
+<details>
+<summary><b>🐛 Bug Fixes</b></summary>

-#### 🔧 Developer Experience
- ✅ **Disabled Auto-Browser Open**: Dev server no longer opens browser automatically
- ✅ **Unified Installers**: One-click installers for Windows, Linux, and macOS
- ✅ **Enhanced Launchers**: Auto-fix capabilities, dependency checking, build verification
- ✅ **Port Conflict Detection**: Warns if default ports are in use
- ✅ **Error Recovery**: Provides actionable error messages with fixes
-
-#### 🐛 Bug Fixes
 - ✅ Fixed Qwen OAuth "empty body" errors
- ✅ Fixed MultiX panel being pushed off screen when Editor is open
- ✅ Fixed top menu/toolbar disappearing when file is selected
- ✅ Fixed layout breaking when scrolling in Editor or Chat
- ✅ Fixed auto-scroll interrupting manual scrolling
- ✅ Fixed sessions not showing on workspace first entry
+- ✅ Fixed MultiX panel being pushed off screen
+- ✅ Fixed top menu/toolbar disappearing
+- ✅ Fixed layout breaking when scrolling
+- ✅ Fixed sessions not showing on workspace entry
+
+</details>

 ---

-## Button Features Guide
+## 🎮 Button Guide

 | Button | Description |
 |--------|-------------|
-| **AUTHED** | Shows authentication status. Green = connected, Red = not authenticated |
-| **AI MODEL** | Click to switch between AI models (GLM 4.7, Claude, GPT, etc.) |
-| **SMART FIX** | AI analyzes your code for errors and automatically applies fixes |
-| **BUILD** | Compiles and builds your project using detected build system |
-| **APEX** | Autonomous mode - AI chains actions without waiting for approval |
-| **SHIELD** | Auto-approval mode - AI makes changes without confirmation prompts |
-| **MULTIX MODE** | Opens multi-task pipeline for parallel AI conversations |
+| **AUTHED** | Shows authentication status (Green = connected) |
+| **AI MODEL** | Click to switch between AI models |
+| **SMART FIX** | AI analyzes code for errors and applies fixes |
+| **BUILD** | Compiles and builds your project |
+| **APEX** | Autonomous mode - AI chains actions automatically |
+| **SHIELD** | Auto-approval mode - AI makes changes without prompts |
+| **MULTIX MODE** | Opens multi-task pipeline for parallel conversations |

 ---

-## Folder Structure
+## 📁 Project Structure

 ```
 NomadArch/
-├── Install-Windows.bat    # Windows installer with dependency checking
-├── Install-Linux.sh       # Linux installer with distro support
-├── Install-Mac.sh         # macOS installer with Apple Silicon support
-├── Launch-Windows.bat     # Windows launcher with auto-fix
-├── Launch-Dev-Windows.bat # Windows developer mode launcher
-├── Launch-Unix.sh         # Linux/macOS launcher
+├── Install-*.bat/.sh      # Platform installers
+├── Launch-*.bat/.sh       # Platform launchers
 ├── packages/
 │   ├── electron-app/      # Electron main process
-│   ├── server/            # Backend server (Fastify)
+│   ├── server/            # Backend (Fastify)
 │   ├── ui/                # Frontend (SolidJS + Vite)
-│   ├── tauri-app/        # Tauri alternative desktop app
 │   └── opencode-config/   # OpenCode configuration
-├── README.md              # This file
-└── package.json           # Root package manifest
+└── README.md
 ```

 ---

-## Requirements
+## 🔧 Requirements

- **Node.js**: v18 or higher
- **npm**: v9 or higher
- **Git**: For version control features
- **OS**: Windows 10+, macOS 11+ (Big Sur), or Linux (Ubuntu 20.04+, Fedora, Arch, OpenSUSE)
-
-### Platform-Specific Requirements
-
-**Windows**:
- Administrator privileges recommended for installation
- 2GB free disk space
-
-**Linux**:
- Build tools (gcc, g++, make)
- Package manager (apt, dnf, pacman, or zypper)
-
-**macOS**:
- Xcode Command Line Tools
- Homebrew (recommended)
- Rosetta 2 for Apple Silicon (for x86_64 compatibility)
+| Requirement | Version |
+|-------------|---------|
+| Node.js | v18+ |
+| npm | v9+ |
+| OS | Windows 10+, macOS 11+, Linux |

 ---

-## Troubleshooting
+## 🆘 Troubleshooting

-### "Dependencies not installed" Error
-Run the installer script first:
- Windows: `Install-Windows.bat`
- Linux: `./Install-Linux.sh`
- macOS: `./Install-Mac.sh`
+<details>
+<summary><b>Common Issues & Solutions</b></summary>

-### "opencode not found" or Workspace Creation Fails
-The installer should automatically install OpenCode CLI. If it fails:
-
-**Option 1 - Manual npm install:**
+**Dependencies not installed?**
 ```bash
-npm install -g opencode-ai@latest
+# Run the installer for your platform
+Install-Windows.bat  # Windows
+./Install-Linux.sh   # Linux
+./Install-Mac.sh     # macOS
 ```

-**Option 2 - Manual download:**
-1. Visit: https://github.com/sst/opencode/releases/latest
-2. Download the appropriate ZIP for your platform:
-   - Windows: `opencode-windows-x64.zip`
-   - Linux x64: `opencode-linux-x64.zip`
-   - Linux ARM64: `opencode-linux-arm64.zip`
-   - macOS Intel: `opencode-darwin-x64.zip`
-   - macOS Apple Silicon: `opencode-darwin-arm64.zip`
-3. Extract and place `opencode` or `opencode.exe` in the `bin/` folder
+**Port conflict?**
+```bash
+# Kill process on port 3000/3001
+taskkill /F /PID <PID>  # Windows
+kill -9 <PID>           # Unix
+```

-### Port 3000 or 3001 Already in Use
-The launchers will detect port conflicts and warn you. To fix:
-1. Close other applications using these ports
-2. Check for running NomadArch instances
-3. Kill the process: `taskkill /F /PID <PID>` (Windows) or `kill -9 <PID>` (Unix)
+**OAuth fails?**
+1. Check internet connection
+2. Complete OAuth in browser
+3. Clear browser cookies and retry

-### Layout Issues
-If the UI looks broken, try:
-1. Refresh the app (Ctrl+R or Cmd+R)
-2. Restart the application
-3. Clear node_modules and reinstall: `rm -rf node_modules && npm install`
-
-### OAuth Authentication Fails
-1. Check your internet connection
-2. Ensure you completed the OAuth flow in your browser
-3. Try logging out and back in
-4. Clear browser cookies for the OAuth provider
-
-### Build Errors
-1. Ensure you have the latest Node.js (18+)
-2. Clear npm cache: `npm cache clean --force`
-3. Delete node_modules: `rm -rf node_modules` (or `rmdir /s /q node_modules` on Windows)
-4. Reinstall: `npm install`
-
-### Sessions Not Showing on Workspace Entry
-This has been fixed with SSE connection waiting. The app now waits for the Server-Sent Events connection to be established before fetching sessions.
+</details>

 ---

-## Credits
+## 🙏 Credits

-### Core Framework & Build Tools
+Built with amazing open source projects:

-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [SolidJS](https://www.solidjs.com/) | ^1.8.0 | Reactive JavaScript UI framework | MIT |
-| [Vite](https://vitejs.dev/) | ^5.0.0 | Next-generation frontend build tool | MIT |
-| [TypeScript](https://www.typescriptlang.org/) | ^5.3.0 - 5.6.3 | JavaScript with type system | Apache-2.0 |
-| [Electron](https://www.electronjs.org/) | Via electron-app | Cross-platform desktop app framework | MIT |
-| [Tauri](https://tauri.app/) | Via tauri-app | Alternative desktop app framework | Apache-2.0/MIT |
-
-### UI Components & Styling
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [@suid/material](https://suid.io/) | ^0.19.0 | Material Design components for SolidJS | MIT |
-| [@suid/icons-material](https://suid.io/) | ^0.9.0 | Material Design icons for SolidJS | MIT |
-| [@suid/system](https://suid.io/) | ^0.14.0 | System components for SolidJS | MIT |
-| [@kobalte/core](https://kobalte.dev/) | 0.13.11 | Accessible, unstyled UI components | MIT |
-| [TailwindCSS](https://tailwindcss.com/) | ^3.0.0 | Utility-first CSS framework | MIT |
-| [PostCSS](https://postcss.org/) | ^8.5.6 | CSS transformation tool | MIT |
-| [Autoprefixer](https://github.com/postcss/autoprefixer) | ^10.4.21 | Parse CSS and add vendor prefixes | MIT |
-
-### Routing & State Management
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [@solidjs/router](https://github.com/solidjs/solid-router) | ^0.13.0 | Router for SolidJS | MIT |
-
-### Markdown & Code Display
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [Marked](https://marked.js.org/) | ^12.0.0 | Markdown parser and compiler | MIT |
-| [GitHub Markdown CSS](https://github.com/sindresorhus/github-markdown-css) | ^5.8.1 | Markdown styling from GitHub | MIT |
-| [Shiki](https://shiki.style/) | ^3.13.0 | Syntax highlighting | MIT |
-| [@git-diff-view/solid](https://github.com/git-diff-view/git-diff-view) | ^0.0.8 | Git diff visualization for SolidJS | MIT |
-
-### Icons & Visuals
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [Lucide Solid](https://lucide.dev/) | ^0.300.0 | Beautiful & consistent icon toolkit | ISC |
-| [QRCode](https://github.com/soldair/node-qrcode) | ^1.5.3 | QR code generation | MIT |
-
-### Backend & Server
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [Fastify](https://www.fastify.io/) | ^4.28.1 | Fast and low overhead web framework | MIT |
-| [@fastify/cors](https://github.com/fastify/fastify-cors) | ^8.5.0 | CORS support for Fastify | MIT |
-| [@fastify/reply-from](https://github.com/fastify/fastify-reply-from) | ^9.8.0 | Proxy support for Fastify | MIT |
-| [@fastify/static](https://github.com/fastify/fastify-static) | ^7.0.4 | Static file serving for Fastify | MIT |
-| [Ollama](https://ollama.com/) | ^0.5.0 | Local AI model integration | MIT |
-
-### AI & SDK
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [OpenCode CLI](https://github.com/sst/opencode) | v1.0.191 | Open source AI coding agent - Required for workspace functionality | MIT |
-| [@opencode-ai/sdk](https://github.com/opencode/ai-sdk) | ^1.0.138 | OpenCode AI SDK | Custom |
-| [google-auth-library](https://github.com/googleapis/google-auth-library-nodejs) | ^10.5.0 | Google OAuth authentication | Apache-2.0 |
-
-### HTTP & Networking
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [Axios](https://axios-http.com/) | ^1.6.0 | Promise-based HTTP client | MIT |
-| [undici](https://undici.nodejs.org/) | ^6.19.8 | HTTP/1.1 client for Node.js | MIT |
-| [node-fetch](https://github.com/node-fetch/node-fetch) | ^3.3.2 | A light-weight module that brings window.fetch to Node.js | MIT |
-
-### Utilities & Helpers
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [Nanoid](https://github.com/ai/nanoid) | ^5.0.4 | Unique string ID generator | MIT |
-| [Debug](https://github.com/debug-js/debug) | ^4.4.3 | Debug logging utility | MIT |
-| [Pino](https://getpino.io/) | ^9.4.0 | Extremely fast Node.js logger | MIT |
-| [FuzzySort](https://github.com/farzher/fuzzysort) | ^2.0.4 | Fuzzy search and sort | MIT |
-| [Zod](https://zod.dev/) | ^3.23.8 | TypeScript-first schema validation | MIT |
-| [Commander](https://github.com/tj/commander.js) | ^12.1.0 | Node.js command-line interface | MIT |
-| [7zip-bin](https://github.com/felixrieseberg/7zip-bin) | ^5.2.0 | 7-Zip binary wrapper | MIT |
-
-### Notifications & Feedback
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [solid-toast](https://github.com/ThisIsFlorian/solid-toast) | ^0.5.0 | Toast notifications for SolidJS | MIT |
-
-### Desktop Integration
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [@tauri-apps/api](https://tauri.app/) | ^2.9.1 | Tauri API for desktop integration | Apache-2.0/MIT |
-| [@tauri-apps/plugin-opener](https://tauri.app/) | ^2.5.2 | Tauri plugin for opening URLs/paths | Apache-2.0/MIT |
-
-### Development Tools
-
-| Project | Version | Description | License |
-|----------|----------|-------------|----------|
-| [Vite Plugin Solid](https://github.com/solidjs/vite-plugin-solid) | ^2.10.0 | Vite plugin for SolidJS | MIT |
-| [ts-node](https://github.com/TypeStrong/ts-node) | ^10.9.2 | TypeScript execution and REPL | MIT |
-| [tsx](https://github.com/privatenumber/tsx) | ^4.20.6 | TypeScript execution | MIT |
-| [cross-env](https://github.com/kentcdodds/cross-env) | ^7.0.3 | Set environment variables across platforms | MIT |
+| Category | Projects |
+|----------|----------|
+| **Framework** | SolidJS, Vite, TypeScript, Electron |
+| **UI** | TailwindCSS, Kobalte, SUID Material |
+| **Backend** | Fastify, Ollama |
+| **AI** | OpenCode CLI, Various AI SDKs |

 ---

-## Project Fork
+## 📄 License

-| Project | Repository | Description |
-|----------|-------------|-------------|
-| [CodeNomad](https://github.com/opencode/codenom) | OpenCode - Original AI coding workspace |
-| [NomadArch](https://github.com/roman-ryzenadvanced/NomadArch-v1.0) | Enhanced fork by NeuralNomadsAI |
-
---
-
-## License
-
-This project is a fork of CodeNomad by OpenCode. Please refer to the original project for licensing information.
-
-All third-party libraries listed above retain their respective licenses.
+This project is a fork of [CodeNomad](https://github.com/opencode/codenom).

 ---

@@ -568,5 +350,5 @@ All third-party libraries listed above retain their respective licenses.
 </p>

 <p align="center">
-  Forked from <a href="https://github.com/opencode/codenom">CodeNomad by OpenCode</a>
+  <sub>NomadArch is an enhanced fork of CodeNomad</sub>
 </p>
--- a/manual_test_guide.md
+++ b/manual_test_guide.md
@@ -0,0 +1,76 @@
+# MultiX v2 - Verification & User Guide
+
+**Date:** 2025-12-25
+**Version:** 2.0.0 (Gold Master)
+
+---
+
+## 1. Feature Verification Guide
+
+### A. Core Multi-Tasking & Parallel Execution
+*   **Goal:** Verify you can run multiple agents at once without freezing.
+*   **Steps:**
+    1.  Create **Task 1**. Type "Write a long story about space" and hit Launch.
+    2.  *While Task 1 is streaming*, click the **+** button to create **Task 2**.
+    3.  Type "Write a python script for fibonacci" in Task 2 and hit Launch.
+    4.  **Result:** Both tasks should stream simultaneously. Switching tabs should be instant.
+
+### B. Per-Task Isolation (Agents & Models)
+*   **Goal:** Verify each task retains its own settings.
+*   **Steps:**
+    1.  Go to **Task 1**. Select Agent: **"Software Engineer"** and Model: **"minimax-m2"**.
+    2.  Go to **Task 2**. Select Agent: **"Writer"** and Model: **"deepseek-chat"**.
+    3.  Switch back and forth.
+    4.  **Result:** The selectors should update to reflect the saved state for each task.
+
+### C. AI Agent Generator (NEW)
+*   **Goal:** Create a custom agent using AI.
+*   **Steps:**
+    1.  Open the **Agent Selector** dropdown.
+    2.  Click **"✨ AI Agent Generator"**.
+    3.  Type: *"A rust expert who is sarcastic and funny"*.
+    4.  Click **"Generate Agent"**.
+    5.  Review the generated name, description, and system prompt.
+    6.  Click **"Save & Use Agent"**.
+    7.  **Result:** The new agent is saved and immediately selected.
+
+### D. Prompt Enhancer
+*   **Goal:** strict Opus 4.5 prompt optimization.
+*   **Steps:**
+    1.  Type a simple prompt: *"fix bug"*.
+    2.  Click the **Magic Wand (✨)** button in the input area.
+    3.  **Result:** The prompt is expanded into a professional, structured request using the active model.
+
+### E. Compaction System
+*   **Goal:** Manage context window usage.
+*   **Steps:**
+    1.  In a long chat, look for the **"Compact suggested"** banner at the top of the chat list.
+    2.  Click **"Compact"** in the banner or the header bar.
+    3.  **Result:** The session history is summarized, freeing up tokens while keeping context.
+
+---
+
+## 2. Menu & Wiring Check
+
+| Button | Wired Action | Status |
+|--------|--------------|--------|
+| **MULTIX Badge** | Visual Indicator | ✅ Active |
+| **SKILLS** | Opens Sidebar (Events) | ✅ Wired |
+| **Active Task** | Shows current task name | ✅ Wired |
+| **Pipeline Tab** | Switches to Dashboard | ✅ Wired |
+| **Task Tabs** | Switch/Close Tasks | ✅ Wired |
+| **Compact Btn** | Triggers Compaction | ✅ Wired |
+| **API Key Btn** | Opens Settings Modal | ✅ Wired |
+| **Agent Select** | Updates Task Session | ✅ Wired |
+| **Model Select** | Updates Task Session | ✅ Wired |
+
+---
+
+## 3. Technical Status
+
+*   **Build:** Passing (No TypeScript errors).
+*   **Dev Server:** Running on port 3001.
+*   **Architecture:** Polling-based (150ms sync) to prevent UI thread blocking.
+*   **State:** Local signals + Non-reactive store references.
+
+**Ready for deployment.**
--- a/package-lock.json
+++ b/package-lock.json
@@ -9608,6 +9608,15 @@
        "node": ">=14.17"
      }
    },
+    "node_modules/ulid": {
+      "version": "3.0.2",
+      "resolved": "https://registry.npmjs.org/ulid/-/ulid-3.0.2.tgz",
+      "integrity": "sha512-yu26mwteFYzBAot7KVMqFGCVpsF6g8wXfJzQUHvu1no3+rRRSFcSV2nKeYvNPLD2J4b08jYBDhHUjeH0ygIl9w==",
+      "license": "MIT",
+      "bin": {
+        "ulid": "dist/cli.js"
+      }
+    },
    "node_modules/undici": {
      "version": "6.22.0",
      "resolved": "https://registry.npmjs.org/undici/-/undici-6.22.0.tgz",
@@ -10604,6 +10613,7 @@
        "fastify": "^4.28.1",
        "fuzzysort": "^2.0.4",
        "pino": "^9.4.0",
+        "ulid": "^3.0.2",
        "undici": "^6.19.8",
        "zod": "^3.23.8"
      },
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
  "name": "codenomad-workspace",
-  "version": "0.4.0",
+  "version": "0.5.0",
  "private": true,
  "description": "CodeNomad monorepo workspace",
  "workspaces": {
--- a/packages/electron-app/electron/main/main.ts
+++ b/packages/electron-app/electron/main/main.ts
@@ -28,6 +28,16 @@ if (isMac) {
  app.commandLine.appendSwitch("disable-spell-checking")
 }

+// Windows: Use Edge WebView2 rendering for better performance
+if (process.platform === "win32") {
+  app.commandLine.appendSwitch("enable-features", "WebViewTagWebComponent,WebView2")
+  app.commandLine.appendSwitch("disable-gpu-sandbox")
+  app.commandLine.appendSwitch("enable-gpu-rasterization")
+  app.commandLine.appendSwitch("enable-zero-copy")
+  app.commandLine.appendSwitch("disable-background-timer-throttling")
+  app.commandLine.appendSwitch("disable-renderer-backgrounding")
+}
+
 function getIconPath() {
  if (app.isPackaged) {
    return join(process.resourcesPath, "icon.png")
--- a/packages/opencode-config/plugin/hello.js
+++ b/packages/opencode-config/plugin/hello.js
@@ -1,18 +1,8 @@
-import { tool } from "@opencode-ai/plugin/tool"
+// NomadArch Plugin Template
+// This file is a placeholder. OpenCode plugins are optional.
+// To create a plugin, see: https://opencode.ai/docs/plugins

-export async function HelloPlugin() {
-  return {
-    tool: {
-      hello: tool({
-        description: "Return a friendly greeting",
-        args: {
-          name: tool.schema.string().optional().describe("Name to greet"),
-        },
-        async execute(args) {
-          const target = args.name?.trim() || "CodeNomad"
-          return `Hello, ${target}!`
-        },
-      }),
-    },
-  }
+export async function init() {
+  // No-op placeholder - customize as needed
+  return {}
 }
--- a/packages/server/package.json
+++ b/packages/server/package.json
@@ -31,6 +31,7 @@
    "fastify": "^4.28.1",
    "fuzzysort": "^2.0.4",
    "pino": "^9.4.0",
+    "ulid": "^3.0.2",
    "undici": "^6.19.8",
    "zod": "^3.23.8"
  },
--- a/packages/server/src/integrations/opencode-zen.ts
+++ b/packages/server/src/integrations/opencode-zen.ts
@@ -42,19 +42,55 @@ export type ZenModel = z.infer<typeof ZenModelSchema>

 // Chat message schema (OpenAI-compatible)
 export const ChatMessageSchema = z.object({
-    role: z.enum(["user", "assistant", "system"]),
-    content: z.string()
+    role: z.enum(["user", "assistant", "system", "tool"]),
+    content: z.string().optional(),
+    tool_calls: z.array(z.object({
+        id: z.string(),
+        type: z.literal("function"),
+        function: z.object({
+            name: z.string(),
+            arguments: z.string()
+        })
+    })).optional(),
+    tool_call_id: z.string().optional()
 })

 export type ChatMessage = z.infer<typeof ChatMessageSchema>

 // Chat request schema
+// Tool Definition Schema
+export const ToolDefinitionSchema = z.object({
+    type: z.literal("function"),
+    function: z.object({
+        name: z.string(),
+        description: z.string(),
+        parameters: z.object({
+            type: z.literal("object"),
+            properties: z.record(z.any()),
+            required: z.array(z.string()).optional()
+        })
+    })
+})
+
+export type ToolDefinition = z.infer<typeof ToolDefinitionSchema>
+
 export const ChatRequestSchema = z.object({
    model: z.string(),
    messages: z.array(ChatMessageSchema),
    stream: z.boolean().default(true),
    temperature: z.number().optional(),
-    max_tokens: z.number().optional()
+    max_tokens: z.number().optional(),
+    tools: z.array(ToolDefinitionSchema).optional(),
+    tool_choice: z.union([
+        z.literal("auto"),
+        z.literal("none"),
+        z.object({
+            type: z.literal("function"),
+            function: z.object({ name: z.string() })
+        })
+    ]).optional(),
+    workspacePath: z.string().optional(),
+    enableTools: z.boolean().optional()
 })

 export type ChatRequest = z.infer<typeof ChatRequestSchema>
--- a/packages/server/src/integrations/zai-api.ts
+++ b/packages/server/src/integrations/zai-api.ts
@@ -1,8 +1,9 @@
 import { z } from "zod"
+import { createHmac } from "crypto"

 export const ZAIConfigSchema = z.object({
    apiKey: z.string().optional(),
-    endpoint: z.string().default("https://api.z.ai/api/paas/v4"),
+    endpoint: z.string().default("https://api.z.ai/api/coding/paas/v4"),
    enabled: z.boolean().default(false),
    timeout: z.number().default(300000)
 })
@@ -10,18 +11,55 @@ export const ZAIConfigSchema = z.object({
 export type ZAIConfig = z.infer<typeof ZAIConfigSchema>

 export const ZAIMessageSchema = z.object({
-    role: z.enum(["user", "assistant", "system"]),
-    content: z.string()
+    role: z.enum(["user", "assistant", "system", "tool"]),
+    content: z.string().optional(),
+    tool_calls: z.array(z.object({
+        id: z.string(),
+        type: z.literal("function"),
+        function: z.object({
+            name: z.string(),
+            arguments: z.string()
+        })
+    })).optional(),
+    tool_call_id: z.string().optional()
 })

 export type ZAIMessage = z.infer<typeof ZAIMessageSchema>

+// Tool Definition Schema (OpenAI-compatible)
+export const ZAIToolSchema = z.object({
+    type: z.literal("function"),
+    function: z.object({
+        name: z.string(),
+        description: z.string(),
+        parameters: z.object({
+            type: z.literal("object"),
+            properties: z.record(z.object({
+                type: z.string(),
+                description: z.string().optional()
+            })),
+            required: z.array(z.string()).optional()
+        })
+    })
+})
+
+export type ZAITool = z.infer<typeof ZAIToolSchema>
+
 export const ZAIChatRequestSchema = z.object({
    model: z.string().default("glm-4.7"),
    messages: z.array(ZAIMessageSchema),
    max_tokens: z.number().default(8192),
    stream: z.boolean().default(true),
    temperature: z.number().optional(),
+    tools: z.array(ZAIToolSchema).optional(),
+    tool_choice: z.union([
+        z.literal("auto"),
+        z.literal("none"),
+        z.object({
+            type: z.literal("function"),
+            function: z.object({ name: z.string() })
+        })
+    ]).optional(),
    thinking: z.object({
        type: z.enum(["enabled", "disabled"]).optional()
    }).optional()
@@ -38,8 +76,16 @@ export const ZAIChatResponseSchema = z.object({
        index: z.number(),
        message: z.object({
            role: z.string(),
-            content: z.string().optional(),
-            reasoning_content: z.string().optional()
+            content: z.string().optional().nullable(),
+            reasoning_content: z.string().optional(),
+            tool_calls: z.array(z.object({
+                id: z.string(),
+                type: z.literal("function"),
+                function: z.object({
+                    name: z.string(),
+                    arguments: z.string()
+                })
+            })).optional()
        }),
        finish_reason: z.string()
    })),
@@ -61,8 +107,17 @@ export const ZAIStreamChunkSchema = z.object({
        index: z.number(),
        delta: z.object({
            role: z.string().optional(),
-            content: z.string().optional(),
-            reasoning_content: z.string().optional()
+            content: z.string().optional().nullable(),
+            reasoning_content: z.string().optional(),
+            tool_calls: z.array(z.object({
+                index: z.number().optional(),
+                id: z.string().optional(),
+                type: z.literal("function").optional(),
+                function: z.object({
+                    name: z.string().optional(),
+                    arguments: z.string().optional()
+                }).optional()
+            })).optional()
        }),
        finish_reason: z.string().nullable().optional()
    }))
@@ -106,7 +161,12 @@ export class ZAIClient {
                })
            })

-            return response.status !== 401 && response.status !== 403
+            if (!response.ok) {
+                const text = await response.text()
+                console.error(`Z.AI connection failed (${response.status}): ${text}`)
+            }
+
+            return response.ok
        } catch (error) {
            console.error("Z.AI connection test failed:", error)
            return false
@@ -194,9 +254,52 @@ export class ZAIClient {
    }

    private getHeaders(): Record<string, string> {
+        const token = this.generateToken(this.config.apiKey!)
        return {
            "Content-Type": "application/json",
-            "Authorization": `Bearer ${this.config.apiKey}`
+            "Authorization": `Bearer ${token}`
+        }
+    }
+
+    private generateToken(apiKey: string, expiresIn: number = 3600): string {
+        try {
+            const [id, secret] = apiKey.split(".")
+            if (!id || !secret) return apiKey // Fallback or handle error
+
+            const now = Date.now()
+            const payload = {
+                api_key: id,
+                exp: now + expiresIn * 1000,
+                timestamp: now
+            }
+
+            const header = {
+                alg: "HS256",
+                sign_type: "SIGN"
+            }
+
+            const base64UrlEncode = (obj: any) => {
+                return Buffer.from(JSON.stringify(obj))
+                    .toString('base64')
+                    .replace(/\+/g, '-')
+                    .replace(/\//g, '_')
+                    .replace(/=+$/, '')
+            }
+
+            const encodedHeader = base64UrlEncode(header)
+            const encodedPayload = base64UrlEncode(payload)
+
+            const signature = createHmac("sha256", secret)
+                .update(`${encodedHeader}.${encodedPayload}`)
+                .digest("base64")
+                .replace(/\+/g, '-')
+                .replace(/\//g, '_')
+                .replace(/=+$/, '')
+
+            return `${encodedHeader}.${encodedPayload}.${signature}`
+        } catch (e) {
+            console.warn("Failed to generate JWT, using raw key", e)
+            return apiKey
        }
    }

--- a/packages/server/src/mcp/client.ts
+++ b/packages/server/src/mcp/client.ts
@@ -0,0 +1,505 @@
+/**
+ * MCP Client - Connects to MCP (Model Context Protocol) servers
+ * and provides tool discovery and execution capabilities.
+ * 
+ * Supports:
+ * - stdio-based MCP servers (command + args)
+ * - HTTP/SSE-based remote MCP servers
+ */
+
+import { spawn, ChildProcess } from "child_process"
+import { createLogger } from "../logger"
+import path from "path"
+
+const log = createLogger({ component: "mcp-client" })
+
+// MCP Protocol Types
+export interface McpServerConfig {
+    command?: string
+    args?: string[]
+    env?: Record<string, string>
+    type?: "stdio" | "remote" | "http" | "sse" | "streamable-http"
+    url?: string
+    headers?: Record<string, string>
+}
+
+export interface McpToolDefinition {
+    name: string
+    description: string
+    inputSchema: {
+        type: "object"
+        properties: Record<string, { type: string; description?: string }>
+        required?: string[]
+    }
+}
+
+export interface McpToolCall {
+    name: string
+    arguments: Record<string, unknown>
+}
+
+export interface McpToolResult {
+    content: Array<{
+        type: "text" | "image" | "resource"
+        text?: string
+        data?: string
+        mimeType?: string
+    }>
+    isError?: boolean
+}
+
+// MCP JSON-RPC Message Types
+interface JsonRpcRequest {
+    jsonrpc: "2.0"
+    id: number | string
+    method: string
+    params?: unknown
+}
+
+interface JsonRpcResponse {
+    jsonrpc: "2.0"
+    id: number | string
+    result?: unknown
+    error?: { code: number; message: string; data?: unknown }
+}
+
+/**
+ * MCP Client for a single server
+ */
+export class McpClient {
+    private config: McpServerConfig
+    private process: ChildProcess | null = null
+    private messageId = 0
+    private pendingRequests: Map<number | string, {
+        resolve: (value: unknown) => void
+        reject: (reason: unknown) => void
+    }> = new Map()
+    private buffer = ""
+    private tools: McpToolDefinition[] = []
+    private connected = false
+    private serverName: string
+
+    constructor(serverName: string, config: McpServerConfig) {
+        this.serverName = serverName
+        this.config = config
+    }
+
+    /**
+     * Start and connect to the MCP server
+     */
+    async connect(): Promise<void> {
+        if (this.connected) return
+
+        if (this.config.type === "remote" || this.config.type === "http" || this.config.type === "sse") {
+            // HTTP-based server - just mark as connected
+            this.connected = true
+            log.info({ server: this.serverName, type: this.config.type }, "Connected to remote MCP server")
+            return
+        }
+
+        // Stdio-based server
+        if (!this.config.command) {
+            throw new Error(`MCP server ${this.serverName} has no command configured`)
+        }
+
+        log.info({ server: this.serverName, command: this.config.command, args: this.config.args }, "Starting MCP server")
+
+        this.process = spawn(this.config.command, this.config.args || [], {
+            stdio: ["pipe", "pipe", "pipe"],
+            env: { ...process.env, ...this.config.env },
+            shell: true
+        })
+
+        this.process.stdout?.on("data", (data) => this.handleData(data.toString()))
+        this.process.stderr?.on("data", (data) => log.warn({ server: this.serverName }, `MCP stderr: ${data}`))
+        this.process.on("error", (err) => log.error({ server: this.serverName, error: err }, "MCP process error"))
+        this.process.on("exit", (code) => {
+            log.info({ server: this.serverName, code }, "MCP process exited")
+            this.connected = false
+        })
+
+        // Wait for process to start
+        await new Promise(resolve => setTimeout(resolve, 500))
+
+        // Initialize the server
+        try {
+            await this.sendRequest("initialize", {
+                protocolVersion: "2024-11-05",
+                capabilities: { tools: {} },
+                clientInfo: { name: "NomadArch", version: "0.4.0" }
+            })
+
+            await this.sendRequest("notifications/initialized", {})
+            this.connected = true
+            log.info({ server: this.serverName }, "MCP server initialized")
+        } catch (error) {
+            log.error({ server: this.serverName, error }, "Failed to initialize MCP server")
+            this.disconnect()
+            throw error
+        }
+    }
+
+    /**
+     * Disconnect from the MCP server
+     */
+    disconnect(): void {
+        if (this.process) {
+            this.process.kill()
+            this.process = null
+        }
+        this.connected = false
+        this.tools = []
+        this.pendingRequests.clear()
+    }
+
+    /**
+     * List available tools from this MCP server
+     */
+    async listTools(): Promise<McpToolDefinition[]> {
+        if (!this.connected) {
+            await this.connect()
+        }
+
+        if (this.config.type === "remote" || this.config.type === "http") {
+            // For HTTP servers, fetch tools via HTTP
+            return this.fetchToolsHttp()
+        }
+
+        try {
+            const response = await this.sendRequest("tools/list", {}) as { tools?: McpToolDefinition[] }
+            this.tools = response.tools || []
+            return this.tools
+        } catch (error) {
+            log.error({ server: this.serverName, error }, "Failed to list MCP tools")
+            return []
+        }
+    }
+
+    /**
+     * Execute a tool on this MCP server
+     */
+    async executeTool(name: string, args: Record<string, unknown>): Promise<McpToolResult> {
+        if (!this.connected) {
+            await this.connect()
+        }
+
+        log.info({ server: this.serverName, tool: name, args }, "Executing MCP tool")
+
+        if (this.config.type === "remote" || this.config.type === "http") {
+            return this.executeToolHttp(name, args)
+        }
+
+        try {
+            const response = await this.sendRequest("tools/call", { name, arguments: args }) as McpToolResult
+            return response
+        } catch (error) {
+            log.error({ server: this.serverName, tool: name, error }, "MCP tool execution failed")
+            return {
+                content: [{ type: "text", text: `Error: ${error instanceof Error ? error.message : String(error)}` }],
+                isError: true
+            }
+        }
+    }
+
+    /**
+     * Send a JSON-RPC request to the MCP server
+     */
+    private async sendRequest(method: string, params?: unknown): Promise<unknown> {
+        if (!this.process?.stdin) {
+            throw new Error("MCP server not running")
+        }
+
+        const id = ++this.messageId
+        const request: JsonRpcRequest = {
+            jsonrpc: "2.0",
+            id,
+            method,
+            params
+        }
+
+        return new Promise((resolve, reject) => {
+            this.pendingRequests.set(id, { resolve, reject })
+
+            const message = JSON.stringify(request) + "\n"
+            this.process!.stdin!.write(message)
+
+            // Timeout after 30 seconds
+            setTimeout(() => {
+                if (this.pendingRequests.has(id)) {
+                    this.pendingRequests.delete(id)
+                    reject(new Error(`MCP request timeout: ${method}`))
+                }
+            }, 30000)
+        })
+    }
+
+    /**
+     * Handle incoming data from the MCP server
+     */
+    private handleData(data: string): void {
+        this.buffer += data
+        const lines = this.buffer.split("\n")
+        this.buffer = lines.pop() || ""
+
+        for (const line of lines) {
+            if (!line.trim()) continue
+            try {
+                const message = JSON.parse(line) as JsonRpcResponse
+                if (message.id !== undefined && this.pendingRequests.has(message.id)) {
+                    const pending = this.pendingRequests.get(message.id)!
+                    this.pendingRequests.delete(message.id)
+
+                    if (message.error) {
+                        pending.reject(new Error(message.error.message))
+                    } else {
+                        pending.resolve(message.result)
+                    }
+                }
+            } catch (e) {
+                log.warn({ server: this.serverName }, `Failed to parse MCP message: ${line}`)
+            }
+        }
+    }
+
+    /**
+     * Fetch tools from HTTP-based MCP server
+     */
+    private async fetchToolsHttp(): Promise<McpToolDefinition[]> {
+        if (!this.config.url) return []
+
+        try {
+            const response = await fetch(`${this.config.url}/tools/list`, {
+                method: "POST",
+                headers: {
+                    "Content-Type": "application/json",
+                    ...this.config.headers
+                },
+                body: JSON.stringify({ jsonrpc: "2.0", id: 1, method: "tools/list", params: {} })
+            })
+
+            if (!response.ok) {
+                throw new Error(`HTTP ${response.status}`)
+            }
+
+            const data = await response.json() as JsonRpcResponse
+            const result = data.result as { tools?: McpToolDefinition[] }
+            return result.tools || []
+        } catch (error) {
+            log.error({ server: this.serverName, error }, "Failed to fetch HTTP MCP tools")
+            return []
+        }
+    }
+
+    /**
+     * Execute tool on HTTP-based MCP server
+     */
+    private async executeToolHttp(name: string, args: Record<string, unknown>): Promise<McpToolResult> {
+        if (!this.config.url) {
+            return { content: [{ type: "text", text: "No URL configured" }], isError: true }
+        }
+
+        try {
+            const response = await fetch(`${this.config.url}/tools/call`, {
+                method: "POST",
+                headers: {
+                    "Content-Type": "application/json",
+                    ...this.config.headers
+                },
+                body: JSON.stringify({
+                    jsonrpc: "2.0",
+                    id: 1,
+                    method: "tools/call",
+                    params: { name, arguments: args }
+                })
+            })
+
+            if (!response.ok) {
+                throw new Error(`HTTP ${response.status}`)
+            }
+
+            const data = await response.json() as JsonRpcResponse
+            return data.result as McpToolResult
+        } catch (error) {
+            return {
+                content: [{ type: "text", text: `HTTP error: ${error instanceof Error ? error.message : String(error)}` }],
+                isError: true
+            }
+        }
+    }
+
+    isConnected(): boolean {
+        return this.connected
+    }
+
+    getServerName(): string {
+        return this.serverName
+    }
+}
+
+/**
+ * MCP Manager - Manages multiple MCP server connections
+ */
+export class McpManager {
+    private clients: Map<string, McpClient> = new Map()
+    private configPath: string | null = null
+
+    /**
+     * Load MCP config from a workspace
+     */
+    async loadConfig(workspacePath: string): Promise<void> {
+        const configPath = path.join(workspacePath, ".mcp.json")
+        this.configPath = configPath
+
+        try {
+            const fs = await import("fs")
+            if (!fs.existsSync(configPath)) {
+                log.info({ path: configPath }, "No MCP config found")
+                return
+            }
+
+            const content = fs.readFileSync(configPath, "utf-8")
+            const config = JSON.parse(content) as { mcpServers?: Record<string, McpServerConfig> }
+
+            if (config.mcpServers) {
+                for (const [name, serverConfig] of Object.entries(config.mcpServers)) {
+                    this.addServer(name, serverConfig)
+                }
+            }
+
+            log.info({ servers: Object.keys(config.mcpServers || {}) }, "Loaded MCP config")
+        } catch (error) {
+            log.error({ path: configPath, error }, "Failed to load MCP config")
+        }
+    }
+
+    /**
+     * Add an MCP server
+     */
+    addServer(name: string, config: McpServerConfig): void {
+        if (this.clients.has(name)) {
+            this.clients.get(name)!.disconnect()
+        }
+        this.clients.set(name, new McpClient(name, config))
+        log.info({ server: name }, "Added MCP server")
+    }
+
+    /**
+     * Remove an MCP server
+     */
+    removeServer(name: string): void {
+        const client = this.clients.get(name)
+        if (client) {
+            client.disconnect()
+            this.clients.delete(name)
+        }
+    }
+
+    /**
+     * Get all available tools from all connected servers
+     */
+    async getAllTools(): Promise<Array<McpToolDefinition & { serverName: string }>> {
+        const allTools: Array<McpToolDefinition & { serverName: string }> = []
+
+        for (const [name, client] of this.clients) {
+            try {
+                const tools = await client.listTools()
+                for (const tool of tools) {
+                    allTools.push({ ...tool, serverName: name })
+                }
+            } catch (error) {
+                log.warn({ server: name, error }, "Failed to get tools from MCP server")
+            }
+        }
+
+        return allTools
+    }
+
+    /**
+     * Convert MCP tools to OpenAI-compatible format
+     */
+    async getToolsAsOpenAIFormat(): Promise<Array<{
+        type: "function"
+        function: {
+            name: string
+            description: string
+            parameters: McpToolDefinition["inputSchema"]
+        }
+    }>> {
+        const mcpTools = await this.getAllTools()
+
+        return mcpTools.map(tool => ({
+            type: "function" as const,
+            function: {
+                // Prefix with server name to avoid conflicts
+                name: `mcp_${tool.serverName}_${tool.name}`,
+                description: `[MCP: ${tool.serverName}] ${tool.description}`,
+                parameters: tool.inputSchema
+            }
+        }))
+    }
+
+    /**
+     * Execute a tool by its full name (mcp_servername_toolname)
+     */
+    async executeTool(fullName: string, args: Record<string, unknown>): Promise<string> {
+        // Parse mcp_servername_toolname format
+        const match = fullName.match(/^mcp_([^_]+)_(.+)$/)
+        if (!match) {
+            return `Error: Invalid MCP tool name format: ${fullName}`
+        }
+
+        const [, serverName, toolName] = match
+        const client = this.clients.get(serverName)
+
+        if (!client) {
+            return `Error: MCP server not found: ${serverName}`
+        }
+
+        const result = await client.executeTool(toolName, args)
+
+        // Convert result to string
+        const texts = result.content
+            .filter(c => c.type === "text" && c.text)
+            .map(c => c.text!)
+
+        return texts.join("\n") || (result.isError ? "Tool execution failed" : "Tool executed successfully")
+    }
+
+    /**
+     * Disconnect all servers
+     */
+    disconnectAll(): void {
+        for (const client of this.clients.values()) {
+            client.disconnect()
+        }
+        this.clients.clear()
+    }
+
+    /**
+     * Get status of all servers
+     */
+    getStatus(): Record<string, { connected: boolean }> {
+        const status: Record<string, { connected: boolean }> = {}
+        for (const [name, client] of this.clients) {
+            status[name] = { connected: client.isConnected() }
+        }
+        return status
+    }
+}
+
+// Singleton instance
+let globalMcpManager: McpManager | null = null
+
+export function getMcpManager(): McpManager {
+    if (!globalMcpManager) {
+        globalMcpManager = new McpManager()
+    }
+    return globalMcpManager
+}
+
+export function resetMcpManager(): void {
+    if (globalMcpManager) {
+        globalMcpManager.disconnectAll()
+        globalMcpManager = null
+    }
+}
--- a/packages/server/src/mcp/index.ts
+++ b/packages/server/src/mcp/index.ts
@@ -0,0 +1,15 @@
+/**
+ * MCP Module Index
+ * Exports MCP client and manager for external MCP server integration.
+ */
+
+export {
+    McpClient,
+    McpManager,
+    getMcpManager,
+    resetMcpManager,
+    type McpServerConfig,
+    type McpToolDefinition,
+    type McpToolCall,
+    type McpToolResult
+} from "./client"
--- a/packages/server/src/server/http-server.ts
+++ b/packages/server/src/server/http-server.ts
@@ -24,6 +24,8 @@ import { registerZAIRoutes } from "./routes/zai"
 import { registerOpenCodeZenRoutes } from "./routes/opencode-zen"
 import { registerSkillsRoutes } from "./routes/skills"
 import { registerContextEngineRoutes } from "./routes/context-engine"
+import { registerNativeSessionsRoutes } from "./routes/native-sessions"
+import { initSessionManager } from "../storage/session-store"
 import { ServerMeta } from "../api-types"
 import { InstanceStore } from "../storage/instance-store"

@@ -40,6 +42,7 @@ interface HttpServerDeps {
  uiStaticDir: string
  uiDevServerUrl?: string
  logger: Logger
+  dataDir?: string // For session storage
 }

 interface HttpServerStartResult {
@@ -56,6 +59,10 @@ export function createHttpServer(deps: HttpServerDeps) {
  const apiLogger = deps.logger.child({ component: "http" })
  const sseLogger = deps.logger.child({ component: "sse" })

+  // Initialize session manager for Binary-Free Mode
+  const dataDir = deps.dataDir || path.join(process.cwd(), ".codenomad-data")
+  initSessionManager(dataDir)
+
  const sseClients = new Set<() => void>()
  const registerSseClient = (cleanup: () => void) => {
    sseClients.add(cleanup)
@@ -126,6 +133,15 @@ export function createHttpServer(deps: HttpServerDeps) {
  registerOpenCodeZenRoutes(app, { logger: deps.logger })
  registerSkillsRoutes(app)
  registerContextEngineRoutes(app)
+
+  // Register Binary-Free Mode native sessions routes
+  registerNativeSessionsRoutes(app, {
+    logger: deps.logger,
+    workspaceManager: deps.workspaceManager,
+    dataDir,
+    eventBus: deps.eventBus,
+  })
+
  registerInstanceProxyRoutes(app, { workspaceManager: deps.workspaceManager, logger: proxyLogger })


--- a/packages/server/src/server/routes/meta.ts
+++ b/packages/server/src/server/routes/meta.ts
@@ -1,5 +1,6 @@
 import { FastifyInstance } from "fastify"
 import os from "os"
+import { existsSync } from "fs"
 import { NetworkAddress, ServerMeta, PortAvailabilityResponse } from "../../api-types"
 import { getAvailablePort } from "../../utils/port"

@@ -7,8 +8,54 @@ interface RouteDeps {
  serverMeta: ServerMeta
 }

+export interface ModeInfo {
+  mode: "lite" | "full"
+  binaryFreeMode: boolean
+  nativeSessions: boolean
+  opencodeBinaryAvailable: boolean
+  providers: {
+    qwen: boolean
+    zai: boolean
+    zen: boolean
+  }
+}
+
 export function registerMetaRoutes(app: FastifyInstance, deps: RouteDeps) {
  app.get("/api/meta", async () => buildMetaResponse(deps.serverMeta))
+
+  // Mode detection endpoint for Binary-Free Mode
+  app.get("/api/meta/mode", async (): Promise<ModeInfo> => {
+    // Check if any OpenCode binary is available
+    const opencodePaths = [
+      process.env.OPENCODE_PATH,
+      "opencode",
+      "opencode.exe",
+    ].filter(Boolean) as string[]
+
+    let binaryAvailable = false
+    for (const p of opencodePaths) {
+      if (existsSync(p)) {
+        binaryAvailable = true
+        break
+      }
+    }
+
+    // In Binary-Free Mode, we use native session management
+    const binaryFreeMode = !binaryAvailable
+
+    return {
+      mode: binaryFreeMode ? "lite" : "full",
+      binaryFreeMode,
+      nativeSessions: true, // Native sessions are always available
+      opencodeBinaryAvailable: binaryAvailable,
+      providers: {
+        qwen: true,  // Always available
+        zai: true,   // Always available
+        zen: true,   // Always available (needs API key)
+      }
+    }
+  })
+
  app.get("/api/ports/available", async () => {
    const port = await getAvailablePort(3000)
    const response: PortAvailabilityResponse = { port }
--- a/packages/server/src/server/routes/native-sessions.ts
+++ b/packages/server/src/server/routes/native-sessions.ts
@@ -0,0 +1,629 @@
+/**
+ * Native Sessions API Routes - Binary-Free Mode
+ * 
+ * These routes provide session management without requiring the OpenCode binary.
+ * They're used when running in "Lite Mode" or when OpenCode is unavailable.
+ */
+
+import { FastifyInstance } from "fastify"
+import { Logger } from "../../logger"
+import { getSessionManager, Session, SessionMessage } from "../../storage/session-store"
+import { CORE_TOOLS, executeTools, type ToolCall, type ToolResult } from "../../tools/executor"
+import { getMcpManager } from "../../mcp/client"
+import { WorkspaceManager } from "../../workspaces/manager"
+import { OpenCodeZenClient, ChatMessage } from "../../integrations/opencode-zen"
+import { EventBus } from "../../events/bus"
+
+interface NativeSessionsDeps {
+    logger: Logger
+    workspaceManager: WorkspaceManager
+    dataDir: string
+    eventBus?: EventBus
+}
+
+// Maximum tool execution loops to prevent infinite loops
+const MAX_TOOL_LOOPS = 10
+
+export function registerNativeSessionsRoutes(app: FastifyInstance, deps: NativeSessionsDeps) {
+    const logger = deps.logger.child({ component: "native-sessions" })
+    const sessionManager = getSessionManager(deps.dataDir)
+
+    // List all sessions for a workspace
+    app.get<{ Params: { workspaceId: string } }>("/api/native/workspaces/:workspaceId/sessions", async (request, reply) => {
+        try {
+            const sessions = await sessionManager.listSessions(request.params.workspaceId)
+            return { sessions }
+        } catch (error) {
+            logger.error({ error }, "Failed to list sessions")
+            reply.code(500)
+            return { error: "Failed to list sessions" }
+        }
+    })
+
+    // Create a new session
+    app.post<{
+        Params: { workspaceId: string }
+        Body: { title?: string; parentId?: string; model?: { providerId: string; modelId: string }; agent?: string }
+    }>("/api/native/workspaces/:workspaceId/sessions", async (request, reply) => {
+        try {
+            const session = await sessionManager.createSession(request.params.workspaceId, request.body)
+
+            // Emit session created event (using any for custom event type)
+            if (deps.eventBus) {
+                deps.eventBus.publish({
+                    type: "native.session.created",
+                    workspaceId: request.params.workspaceId,
+                    session
+                } as any)
+            }
+
+            reply.code(201)
+            return { session }
+        } catch (error) {
+            logger.error({ error }, "Failed to create session")
+            reply.code(500)
+            return { error: "Failed to create session" }
+        }
+    })
+
+    // Get a specific session
+    app.get<{ Params: { workspaceId: string; sessionId: string } }>("/api/native/workspaces/:workspaceId/sessions/:sessionId", async (request, reply) => {
+        try {
+            const session = await sessionManager.getSession(request.params.workspaceId, request.params.sessionId)
+            if (!session) {
+                reply.code(404)
+                return { error: "Session not found" }
+            }
+            return { session }
+        } catch (error) {
+            logger.error({ error }, "Failed to get session")
+            reply.code(500)
+            return { error: "Failed to get session" }
+        }
+    })
+
+    // Update a session
+    app.patch<{
+        Params: { workspaceId: string; sessionId: string }
+        Body: Partial<Session>
+    }>("/api/native/workspaces/:workspaceId/sessions/:sessionId", async (request, reply) => {
+        try {
+            const session = await sessionManager.updateSession(
+                request.params.workspaceId,
+                request.params.sessionId,
+                request.body
+            )
+            if (!session) {
+                reply.code(404)
+                return { error: "Session not found" }
+            }
+            return { session }
+        } catch (error) {
+            logger.error({ error }, "Failed to update session")
+            reply.code(500)
+            return { error: "Failed to update session" }
+        }
+    })
+
+    // Delete a session
+    app.delete<{ Params: { workspaceId: string; sessionId: string } }>("/api/native/workspaces/:workspaceId/sessions/:sessionId", async (request, reply) => {
+        try {
+            const deleted = await sessionManager.deleteSession(request.params.workspaceId, request.params.sessionId)
+            if (!deleted) {
+                reply.code(404)
+                return { error: "Session not found" }
+            }
+            reply.code(204)
+            return
+        } catch (error) {
+            logger.error({ error }, "Failed to delete session")
+            reply.code(500)
+            return { error: "Failed to delete session" }
+        }
+    })
+
+    // Get messages for a session
+    app.get<{ Params: { workspaceId: string; sessionId: string } }>("/api/native/workspaces/:workspaceId/sessions/:sessionId/messages", async (request, reply) => {
+        try {
+            const messages = await sessionManager.getSessionMessages(
+                request.params.workspaceId,
+                request.params.sessionId
+            )
+            return { messages }
+        } catch (error) {
+            logger.error({ error }, "Failed to get messages")
+            reply.code(500)
+            return { error: "Failed to get messages" }
+        }
+    })
+
+    // Add a message (user prompt) and get streaming response
+    app.post<{
+        Params: { workspaceId: string; sessionId: string }
+        Body: {
+            content: string
+            provider: "qwen" | "zai" | "zen"
+            model?: string
+            accessToken?: string
+            resourceUrl?: string
+            enableTools?: boolean
+            systemPrompt?: string
+        }
+    }>("/api/native/workspaces/:workspaceId/sessions/:sessionId/prompt", async (request, reply) => {
+        const { workspaceId, sessionId } = request.params
+        const { content, provider, model, accessToken, resourceUrl, enableTools = true, systemPrompt } = request.body
+
+        try {
+            // Add user message
+            const userMessage = await sessionManager.addMessage(workspaceId, sessionId, {
+                role: "user",
+                content,
+                status: "completed",
+            })
+
+            // Get workspace path
+            const workspace = deps.workspaceManager.get(workspaceId)
+            const workspacePath = workspace?.path ?? process.cwd()
+
+            // Get all messages for context
+            const allMessages = await sessionManager.getSessionMessages(workspaceId, sessionId)
+
+            // Build chat messages array
+            const chatMessages: ChatMessage[] = []
+
+            // Add system prompt if provided
+            if (systemPrompt) {
+                chatMessages.push({ role: "system", content: systemPrompt })
+            }
+
+            // Add conversation history
+            for (const m of allMessages) {
+                if (m.role === "user" || m.role === "assistant" || m.role === "system") {
+                    chatMessages.push({ role: m.role, content: m.content ?? "" })
+                }
+            }
+
+            // Load MCP tools
+            let allTools = [...CORE_TOOLS]
+            if (enableTools) {
+                try {
+                    const mcpManager = getMcpManager()
+                    await mcpManager.loadConfig(workspacePath)
+                    const mcpTools = await mcpManager.getToolsAsOpenAIFormat()
+                    allTools = [...CORE_TOOLS, ...mcpTools]
+                } catch (mcpError) {
+                    logger.warn({ error: mcpError }, "Failed to load MCP tools")
+                }
+            }
+
+            // Create streaming response
+            reply.raw.writeHead(200, {
+                'Content-Type': 'text/event-stream',
+                'Cache-Control': 'no-cache',
+                'Connection': 'keep-alive',
+                'X-Accel-Buffering': 'no',
+            })
+
+            // Create assistant message placeholder
+            const assistantMessage = await sessionManager.addMessage(workspaceId, sessionId, {
+                role: "assistant",
+                content: "",
+                status: "streaming",
+            })
+
+            let fullContent = ""
+
+            try {
+                // Route to the appropriate provider
+                fullContent = await streamWithProvider({
+                    provider,
+                    model,
+                    accessToken,
+                    resourceUrl,
+                    messages: chatMessages,
+                    tools: enableTools ? allTools : [],
+                    workspacePath,
+                    rawResponse: reply.raw,
+                    logger,
+                })
+            } catch (streamError) {
+                logger.error({ error: streamError }, "Stream error")
+                reply.raw.write(`data: ${JSON.stringify({ error: String(streamError) })}\n\n`)
+            }
+
+            // Update assistant message with full content
+            await sessionManager.updateMessage(workspaceId, assistantMessage.id, {
+                content: fullContent,
+                status: "completed",
+            })
+
+            // Emit message event (using any for custom event type)
+            if (deps.eventBus) {
+                deps.eventBus.publish({
+                    type: "native.message.completed",
+                    workspaceId,
+                    sessionId,
+                    messageId: assistantMessage.id,
+                } as any)
+            }
+
+            reply.raw.write('data: [DONE]\n\n')
+            reply.raw.end()
+        } catch (error) {
+            logger.error({ error }, "Failed to process prompt")
+            if (!reply.sent) {
+                reply.code(500)
+                return { error: "Failed to process prompt" }
+            }
+        }
+    })
+
+    // SSE endpoint for session events
+    app.get<{ Params: { workspaceId: string } }>("/api/native/workspaces/:workspaceId/events", async (request, reply) => {
+        reply.raw.writeHead(200, {
+            'Content-Type': 'text/event-stream',
+            'Cache-Control': 'no-cache',
+            'Connection': 'keep-alive',
+            'X-Accel-Buffering': 'no',
+        })
+
+        // Send initial ping
+        reply.raw.write(`data: ${JSON.stringify({ type: "ping" })}\n\n`)
+
+        // Keep connection alive
+        const keepAlive = setInterval(() => {
+            reply.raw.write(`data: ${JSON.stringify({ type: "ping" })}\n\n`)
+        }, 30000)
+
+        // Handle client disconnect
+        request.raw.on("close", () => {
+            clearInterval(keepAlive)
+        })
+    })
+
+    logger.info("Native sessions routes registered (Binary-Free Mode)")
+}
+
+/**
+ * Stream chat with the appropriate provider
+ */
+async function streamWithProvider(opts: {
+    provider: "qwen" | "zai" | "zen"
+    model?: string
+    accessToken?: string
+    resourceUrl?: string
+    messages: ChatMessage[]
+    tools: any[]
+    workspacePath: string
+    rawResponse: any
+    logger: Logger
+}): Promise<string> {
+    const { provider, model, accessToken, resourceUrl, messages, tools, workspacePath, rawResponse, logger } = opts
+
+    let fullContent = ""
+    let loopCount = 0
+    let currentMessages = [...messages]
+
+    // Tool execution loop
+    while (loopCount < MAX_TOOL_LOOPS) {
+        loopCount++
+
+        let responseContent = ""
+        let toolCalls: ToolCall[] = []
+
+        // Route to the appropriate provider
+        switch (provider) {
+            case "zen":
+                const zenResult = await streamWithZen(model, currentMessages, tools, rawResponse, logger)
+                responseContent = zenResult.content
+                toolCalls = zenResult.toolCalls
+                break
+
+            case "qwen":
+                const qwenResult = await streamWithQwen(accessToken, resourceUrl, model, currentMessages, tools, rawResponse, logger)
+                responseContent = qwenResult.content
+                toolCalls = qwenResult.toolCalls
+                break
+
+            case "zai":
+                const zaiResult = await streamWithZAI(accessToken, model, currentMessages, tools, rawResponse, logger)
+                responseContent = zaiResult.content
+                toolCalls = zaiResult.toolCalls
+                break
+        }
+
+        fullContent += responseContent
+
+        // If no tool calls, we're done
+        if (toolCalls.length === 0) {
+            break
+        }
+
+        // Execute tools
+        logger.info({ toolCount: toolCalls.length }, "Executing tool calls")
+
+        // Add assistant message with tool calls
+        currentMessages.push({
+            role: "assistant",
+            content: responseContent,
+            tool_calls: toolCalls.map(tc => ({
+                id: tc.id,
+                type: "function" as const,
+                function: tc.function
+            }))
+        })
+
+        // Execute each tool and add result
+        const toolResults = await executeTools(workspacePath, toolCalls)
+
+        for (let i = 0; i < toolCalls.length; i++) {
+            const tc = toolCalls[i]
+            const result = toolResults[i]
+
+            // Emit tool execution event
+            rawResponse.write(`data: ${JSON.stringify({
+                type: "tool_execution",
+                tool: tc.function.name,
+                result: result?.content?.substring(0, 200) // Preview
+            })}\n\n`)
+
+            currentMessages.push({
+                role: "tool",
+                content: result?.content ?? "Tool execution failed",
+                tool_call_id: tc.id
+            })
+        }
+    }
+
+    return fullContent
+}
+
+/**
+ * Stream with OpenCode Zen (free models)
+ */
+async function streamWithZen(
+    model: string | undefined,
+    messages: ChatMessage[],
+    tools: any[],
+    rawResponse: any,
+    logger: Logger
+): Promise<{ content: string; toolCalls: ToolCall[] }> {
+    const zenClient = new OpenCodeZenClient()
+    let content = ""
+    const toolCalls: ToolCall[] = []
+
+    try {
+        const stream = zenClient.chatStream({
+            model: model ?? "gpt-5-nano",
+            messages,
+            stream: true,
+            tools: tools.length > 0 ? tools : undefined,
+            tool_choice: tools.length > 0 ? "auto" : undefined,
+        })
+
+        for await (const chunk of stream) {
+            const delta = chunk.choices?.[0]?.delta
+            if (delta?.content) {
+                content += delta.content
+                rawResponse.write(`data: ${JSON.stringify({ choices: [{ delta: { content: delta.content } }] })}\n\n`)
+            }
+
+            // Handle tool calls (if model supports them)
+            const deltaToolCalls = (delta as any)?.tool_calls
+            if (deltaToolCalls) {
+                for (const tc of deltaToolCalls) {
+                    if (tc.function?.name) {
+                        toolCalls.push({
+                            id: tc.id,
+                            type: "function",
+                            function: {
+                                name: tc.function.name,
+                                arguments: tc.function.arguments ?? "{}"
+                            }
+                        })
+                    }
+                }
+            }
+        }
+    } catch (error) {
+        logger.error({ error }, "Zen streaming error")
+        throw error
+    }
+
+    return { content, toolCalls }
+}
+
+/**
+ * Stream with Qwen API
+ */
+async function streamWithQwen(
+    accessToken: string | undefined,
+    resourceUrl: string | undefined,
+    model: string | undefined,
+    messages: ChatMessage[],
+    tools: any[],
+    rawResponse: any,
+    logger: Logger
+): Promise<{ content: string; toolCalls: ToolCall[] }> {
+    if (!accessToken) {
+        throw new Error("Qwen access token required. Please authenticate with Qwen first.")
+    }
+
+    const baseUrl = resourceUrl ?? "https://chat.qwen.ai"
+    let content = ""
+    const toolCalls: ToolCall[] = []
+
+    try {
+        const response = await fetch(`${baseUrl}/api/v1/chat/completions`, {
+            method: "POST",
+            headers: {
+                "Content-Type": "application/json",
+                "Authorization": `Bearer ${accessToken}`,
+            },
+            body: JSON.stringify({
+                model: model ?? "qwen-plus-latest",
+                messages,
+                stream: true,
+                tools: tools.length > 0 ? tools : undefined,
+                tool_choice: tools.length > 0 ? "auto" : undefined,
+            })
+        })
+
+        if (!response.ok) {
+            const error = await response.text()
+            throw new Error(`Qwen API error: ${response.status} - ${error}`)
+        }
+
+        const reader = response.body?.getReader()
+        if (!reader) throw new Error("No response body")
+
+        const decoder = new TextDecoder()
+        let buffer = ""
+
+        while (true) {
+            const { done, value } = await reader.read()
+            if (done) break
+
+            buffer += decoder.decode(value, { stream: true })
+            const lines = buffer.split("\n")
+            buffer = lines.pop() ?? ""
+
+            for (const line of lines) {
+                if (line.startsWith("data: ")) {
+                    const data = line.slice(6)
+                    if (data === "[DONE]") continue
+
+                    try {
+                        const parsed = JSON.parse(data)
+                        const delta = parsed.choices?.[0]?.delta
+
+                        if (delta?.content) {
+                            content += delta.content
+                            rawResponse.write(`data: ${JSON.stringify({ choices: [{ delta: { content: delta.content } }] })}\n\n`)
+                        }
+
+                        if (delta?.tool_calls) {
+                            for (const tc of delta.tool_calls) {
+                                if (tc.function?.name) {
+                                    toolCalls.push({
+                                        id: tc.id ?? `call_${Date.now()}`,
+                                        type: "function",
+                                        function: {
+                                            name: tc.function.name,
+                                            arguments: tc.function.arguments ?? "{}"
+                                        }
+                                    })
+                                }
+                            }
+                        }
+                    } catch {
+                        // Skip invalid JSON
+                    }
+                }
+            }
+        }
+    } catch (error) {
+        logger.error({ error }, "Qwen streaming error")
+        throw error
+    }
+
+    return { content, toolCalls }
+}
+
+/**
+ * Stream with Z.AI API
+ */
+async function streamWithZAI(
+    accessToken: string | undefined,
+    model: string | undefined,
+    messages: ChatMessage[],
+    tools: any[],
+    rawResponse: any,
+    logger: Logger
+): Promise<{ content: string; toolCalls: ToolCall[] }> {
+    let content = ""
+    const toolCalls: ToolCall[] = []
+
+    const baseUrl = "https://api.z.ai"
+
+    try {
+        const headers: Record<string, string> = {
+            "Content-Type": "application/json",
+        }
+
+        if (accessToken) {
+            headers["Authorization"] = `Bearer ${accessToken}`
+        }
+
+        const response = await fetch(`${baseUrl}/v1/chat/completions`, {
+            method: "POST",
+            headers,
+            body: JSON.stringify({
+                model: model ?? "z1-mini",
+                messages,
+                stream: true,
+                tools: tools.length > 0 ? tools : undefined,
+                tool_choice: tools.length > 0 ? "auto" : undefined,
+            })
+        })
+
+        if (!response.ok) {
+            const error = await response.text()
+            throw new Error(`Z.AI API error: ${response.status} - ${error}`)
+        }
+
+        const reader = response.body?.getReader()
+        if (!reader) throw new Error("No response body")
+
+        const decoder = new TextDecoder()
+        let buffer = ""
+
+        while (true) {
+            const { done, value } = await reader.read()
+            if (done) break
+
+            buffer += decoder.decode(value, { stream: true })
+            const lines = buffer.split("\n")
+            buffer = lines.pop() ?? ""
+
+            for (const line of lines) {
+                if (line.startsWith("data: ")) {
+                    const data = line.slice(6)
+                    if (data === "[DONE]") continue
+
+                    try {
+                        const parsed = JSON.parse(data)
+                        const delta = parsed.choices?.[0]?.delta
+
+                        if (delta?.content) {
+                            content += delta.content
+                            rawResponse.write(`data: ${JSON.stringify({ choices: [{ delta: { content: delta.content } }] })}\n\n`)
+                        }
+
+                        if (delta?.tool_calls) {
+                            for (const tc of delta.tool_calls) {
+                                if (tc.function?.name) {
+                                    toolCalls.push({
+                                        id: tc.id ?? `call_${Date.now()}`,
+                                        type: "function",
+                                        function: {
+                                            name: tc.function.name,
+                                            arguments: tc.function.arguments ?? "{}"
+                                        }
+                                    })
+                                }
+                            }
+                        }
+                    } catch {
+                        // Skip invalid JSON
+                    }
+                }
+            }
+        }
+    } catch (error) {
+        logger.error({ error }, "Z.AI streaming error")
+        throw error
+    }
+
+    return { content, toolCalls }
+}
--- a/packages/server/src/server/routes/opencode-zen.ts
+++ b/packages/server/src/server/routes/opencode-zen.ts
@@ -1,11 +1,16 @@
 import { FastifyInstance } from "fastify"
-import { OpenCodeZenClient, type ChatRequest, getDefaultZenConfig } from "../../integrations/opencode-zen"
+import { OpenCodeZenClient, type ChatRequest, getDefaultZenConfig, type ChatMessage } from "../../integrations/opencode-zen"
 import { Logger } from "../../logger"
+import { CORE_TOOLS, executeTools, type ToolCall, type ToolResult } from "../../tools/executor"
+import { getMcpManager } from "../../mcp/client"

 interface OpenCodeZenRouteDeps {
    logger: Logger
 }

+// Maximum number of tool execution loops
+const MAX_TOOL_LOOPS = 10
+
 export async function registerOpenCodeZenRoutes(
    app: FastifyInstance,
    deps: OpenCodeZenRouteDeps
@@ -49,12 +54,25 @@ export async function registerOpenCodeZenRoutes(
        }
    })

-    // Chat completion endpoint
+    // Chat completion endpoint WITH MCP TOOL SUPPORT
    app.post('/api/opencode-zen/chat', async (request, reply) => {
        try {
-            const chatRequest = request.body as ChatRequest
+            const chatRequest = request.body as ChatRequest & {
+                workspacePath?: string
+                enableTools?: boolean
+            }

-            // Handle streaming
+            // Extract workspace path for tool execution
+            const workspacePath = chatRequest.workspacePath || process.cwd()
+            const enableTools = chatRequest.enableTools !== false
+
+            logger.info({
+                workspacePath,
+                receivedWorkspacePath: chatRequest.workspacePath,
+                enableTools
+            }, "OpenCode Zen chat request received")
+
+            // Handle streaming with tool loop
            if (chatRequest.stream) {
                reply.raw.writeHead(200, {
                    'Content-Type': 'text/event-stream',
@@ -63,16 +81,14 @@ export async function registerOpenCodeZenRoutes(
                })

                try {
-                    for await (const chunk of client.chatStream(chatRequest)) {
-                        reply.raw.write(`data: ${JSON.stringify(chunk)}\n\n`)
-
-                        // Check for finish
-                        if (chunk.choices?.[0]?.finish_reason) {
-                            reply.raw.write('data: [DONE]\n\n')
-                            break
-                        }
-                    }
-
+                    await streamWithToolLoop(
+                        client,
+                        chatRequest,
+                        workspacePath,
+                        enableTools,
+                        reply.raw,
+                        logger
+                    )
                    reply.raw.end()
                } catch (streamError) {
                    logger.error({ error: streamError }, "OpenCode Zen streaming failed")
@@ -80,7 +96,14 @@ export async function registerOpenCodeZenRoutes(
                    reply.raw.end()
                }
            } else {
-                const response = await client.chat(chatRequest)
+                // Non-streaming with tool loop
+                const response = await chatWithToolLoop(
+                    client,
+                    chatRequest,
+                    workspacePath,
+                    enableTools,
+                    logger
+                )
                return response
            }
        } catch (error) {
@@ -89,5 +112,213 @@ export async function registerOpenCodeZenRoutes(
        }
    })

-    logger.info("OpenCode Zen routes registered - Free models available!")
+    logger.info("OpenCode Zen routes registered with MCP tool support - Free models available!")
+}
+
+/**
+ * Streaming chat with tool execution loop
+ */
+async function streamWithToolLoop(
+    client: OpenCodeZenClient,
+    request: ChatRequest,
+    workspacePath: string,
+    enableTools: boolean,
+    rawResponse: any,
+    logger: Logger
+): Promise<void> {
+    let messages = [...request.messages]
+    let loopCount = 0
+
+    // Load MCP tools from workspace config
+    let allTools = [...CORE_TOOLS]
+    if (enableTools && workspacePath) {
+        try {
+            const mcpManager = getMcpManager()
+            await mcpManager.loadConfig(workspacePath)
+            const mcpTools = await mcpManager.getToolsAsOpenAIFormat()
+            allTools = [...CORE_TOOLS, ...mcpTools]
+            if (mcpTools.length > 0) {
+                logger.info({ mcpToolCount: mcpTools.length }, "Loaded MCP tools for OpenCode Zen")
+            }
+        } catch (mcpError) {
+            logger.warn({ error: mcpError }, "Failed to load MCP tools")
+        }
+    }
+
+    // Inject tools if enabled
+    const requestWithTools: ChatRequest = {
+        ...request,
+        tools: enableTools ? allTools : undefined,
+        tool_choice: enableTools ? "auto" : undefined
+    }
+
+    while (loopCount < MAX_TOOL_LOOPS) {
+        loopCount++
+
+        // Accumulate tool calls from stream
+        let accumulatedToolCalls: { [index: number]: { id: string; name: string; arguments: string } } = {}
+        let hasToolCalls = false
+        let textContent = ""
+
+        // Stream response
+        for await (const chunk of client.chatStream({ ...requestWithTools, messages })) {
+            // Write chunk to client
+            rawResponse.write(`data: ${JSON.stringify(chunk)}\n\n`)
+
+            const choice = chunk.choices[0]
+            if (!choice) continue
+
+            // Accumulate text content
+            if (choice.delta?.content) {
+                textContent += choice.delta.content
+            }
+
+            // Accumulate tool calls from delta (if API supports it)
+            const deltaToolCalls = (choice.delta as any)?.tool_calls
+            if (deltaToolCalls) {
+                hasToolCalls = true
+                for (const tc of deltaToolCalls) {
+                    const idx = tc.index ?? 0
+                    if (!accumulatedToolCalls[idx]) {
+                        accumulatedToolCalls[idx] = { id: tc.id || "", name: "", arguments: "" }
+                    }
+                    if (tc.id) accumulatedToolCalls[idx].id = tc.id
+                    if (tc.function?.name) accumulatedToolCalls[idx].name += tc.function.name
+                    if (tc.function?.arguments) accumulatedToolCalls[idx].arguments += tc.function.arguments
+                }
+            }
+
+            // Check if we should stop
+            if (choice.finish_reason === "stop") {
+                rawResponse.write('data: [DONE]\n\n')
+                return
+            }
+        }
+
+        // If no tool calls, we're done
+        if (!hasToolCalls || !enableTools) {
+            rawResponse.write('data: [DONE]\n\n')
+            return
+        }
+
+        // Convert accumulated tool calls
+        const toolCalls: ToolCall[] = Object.values(accumulatedToolCalls).map(tc => ({
+            id: tc.id,
+            type: "function" as const,
+            function: {
+                name: tc.name,
+                arguments: tc.arguments
+            }
+        }))
+
+        if (toolCalls.length === 0) {
+            rawResponse.write('data: [DONE]\n\n')
+            return
+        }
+
+        logger.info({ toolCalls: toolCalls.map(tc => tc.function.name) }, "Executing tool calls")
+
+        // Add assistant message with tool calls
+        const assistantMessage: ChatMessage = {
+            role: "assistant",
+            content: textContent || undefined,
+            tool_calls: toolCalls
+        }
+        messages.push(assistantMessage)
+
+        // Execute tools
+        const toolResults = await executeTools(workspacePath, toolCalls)
+
+        // Notify client about tool execution via special event
+        for (const result of toolResults) {
+            const toolEvent = {
+                type: "tool_result",
+                tool_call_id: result.tool_call_id,
+                content: result.content
+            }
+            rawResponse.write(`data: ${JSON.stringify(toolEvent)}\n\n`)
+        }
+
+        // Add tool results to messages
+        for (const result of toolResults) {
+            const toolMessage: ChatMessage = {
+                role: "tool",
+                content: result.content,
+                tool_call_id: result.tool_call_id
+            }
+            messages.push(toolMessage)
+        }
+
+        logger.info({ loopCount, toolsExecuted: toolResults.length }, "Tool loop iteration complete")
+    }
+
+    logger.warn({ loopCount }, "Max tool loops reached")
+    rawResponse.write('data: [DONE]\n\n')
+}
+
+/**
+ * Non-streaming chat with tool execution loop
+ */
+async function chatWithToolLoop(
+    client: OpenCodeZenClient,
+    request: ChatRequest,
+    workspacePath: string,
+    enableTools: boolean,
+    logger: Logger
+): Promise<any> {
+    let messages = [...request.messages]
+    let loopCount = 0
+    let lastResponse: any = null
+
+    // Inject tools if enabled
+    const requestWithTools: ChatRequest = {
+        ...request,
+        tools: enableTools ? CORE_TOOLS : undefined,
+        tool_choice: enableTools ? "auto" : undefined
+    }
+
+    while (loopCount < MAX_TOOL_LOOPS) {
+        loopCount++
+
+        const response = await client.chat({ ...requestWithTools, messages, stream: false })
+        lastResponse = response
+
+        const choice = response.choices[0]
+        if (!choice) break
+
+        const toolCalls = (choice.message as any)?.tool_calls
+
+        // If no tool calls, return
+        if (!toolCalls || toolCalls.length === 0 || !enableTools) {
+            return response
+        }
+
+        logger.info({ toolCalls: toolCalls.map((tc: any) => tc.function.name) }, "Executing tool calls")
+
+        // Add assistant message
+        const assistantMessage: ChatMessage = {
+            role: "assistant",
+            content: (choice.message as any).content || undefined,
+            tool_calls: toolCalls
+        }
+        messages.push(assistantMessage)
+
+        // Execute tools
+        const toolResults = await executeTools(workspacePath, toolCalls)
+
+        // Add tool results
+        for (const result of toolResults) {
+            const toolMessage: ChatMessage = {
+                role: "tool",
+                content: result.content,
+                tool_call_id: result.tool_call_id
+            }
+            messages.push(toolMessage)
+        }
+
+        logger.info({ loopCount, toolsExecuted: toolResults.length }, "Tool loop iteration complete")
+    }
+
+    logger.warn({ loopCount }, "Max tool loops reached")
+    return lastResponse
 }
--- a/packages/server/src/server/routes/qwen.ts
+++ b/packages/server/src/server/routes/qwen.ts
@@ -1,10 +1,16 @@
 import { FastifyInstance, FastifyReply } from "fastify"
+import { join } from "path"
+import { existsSync, mkdirSync } from "fs"
 import { Logger } from "../../logger"
+import { CORE_TOOLS, executeTools, type ToolCall, type ToolResult } from "../../tools/executor"
+import { getMcpManager } from "../../mcp/client"

 interface QwenRouteDeps {
  logger: Logger
 }

+const MAX_TOOL_LOOPS = 10
+
 const QWEN_OAUTH_BASE_URL = 'https://chat.qwen.ai'
 const QWEN_OAUTH_DEVICE_CODE_ENDPOINT = `${QWEN_OAUTH_BASE_URL}/api/v1/oauth2/device/code`
 const QWEN_OAUTH_TOKEN_ENDPOINT = `${QWEN_OAUTH_BASE_URL}/api/v1/oauth2/token`
@@ -197,7 +203,159 @@ export async function registerQwenRoutes(
    }
  })

-  // Qwen Chat API - proxy chat requests to Qwen using OAuth token
+  /**
+   * Streaming chat with tool execution loop for Qwen
+   */
+  async function streamWithToolLoop(
+    accessToken: string,
+    chatUrl: string,
+    initialRequest: any,
+    workspacePath: string,
+    enableTools: boolean,
+    rawResponse: any,
+    logger: Logger
+  ) {
+    let messages = [...initialRequest.messages]
+    let loopCount = 0
+    const model = initialRequest.model
+
+    while (loopCount < MAX_TOOL_LOOPS) {
+      loopCount++
+      logger.info({ loopCount, model }, "Starting Qwen tool loop iteration")
+
+      const response = await fetch(chatUrl, {
+        method: 'POST',
+        headers: {
+          'Content-Type': 'application/json',
+          'Authorization': `Bearer ${accessToken}`,
+          'Accept': 'text/event-stream'
+        },
+        body: JSON.stringify({
+          ...initialRequest,
+          messages,
+          stream: true,
+          tools: enableTools ? initialRequest.tools : undefined,
+          tool_choice: enableTools ? "auto" : undefined
+        })
+      })
+
+      if (!response.ok) {
+        const errorText = await response.text()
+        throw new Error(`Qwen API error (${response.status}): ${errorText}`)
+      }
+
+      if (!response.body) throw new Error("No response body")
+
+      const reader = response.body.getReader()
+      const decoder = new TextDecoder()
+      let textContent = ""
+      let hasToolCalls = false
+      let accumulatedToolCalls: Record<number, { id: string, name: string, arguments: string }> = {}
+      let buffer = ""
+
+      while (true) {
+        const { done, value } = await reader.read()
+        if (done) break
+        buffer += decoder.decode(value, { stream: true })
+        const lines = buffer.split("\n")
+        buffer = lines.pop() || ""
+
+        for (const line of lines) {
+          const trimmed = line.trim()
+          if (!trimmed.startsWith("data: ")) continue
+          const data = trimmed.slice(6).trim()
+          if (data === "[DONE]") {
+            if (!hasToolCalls) {
+              rawResponse.write('data: [DONE]\n\n')
+              return
+            }
+            break
+          }
+
+          let chunk: any
+          try {
+            chunk = JSON.parse(data)
+          } catch (e) {
+            continue
+          }
+
+          const choice = chunk.choices?.[0]
+          if (!choice) continue
+
+          // Pass through text content to client
+          if (choice.delta?.content) {
+            textContent += choice.delta.content
+            rawResponse.write(`data: ${JSON.stringify(chunk)}\n\n`)
+          }
+
+          // Accumulate tool calls
+          if (choice.delta?.tool_calls) {
+            hasToolCalls = true
+            for (const tc of choice.delta.tool_calls) {
+              const idx = tc.index ?? 0
+              if (!accumulatedToolCalls[idx]) {
+                accumulatedToolCalls[idx] = { id: tc.id || "", name: "", arguments: "" }
+              }
+              if (tc.id) accumulatedToolCalls[idx].id = tc.id
+              if (tc.function?.name) accumulatedToolCalls[idx].name += tc.function.name
+              if (tc.function?.arguments) accumulatedToolCalls[idx].arguments += tc.function.arguments
+            }
+          }
+
+          if (choice.finish_reason === "tool_calls") {
+            break
+          }
+
+          if (choice.finish_reason === "stop" && !hasToolCalls) {
+            rawResponse.write('data: [DONE]\n\n')
+            return
+          }
+        }
+      }
+
+      // If no tool calls, we're done
+      if (!hasToolCalls || !enableTools) {
+        rawResponse.write('data: [DONE]\n\n')
+        return
+      }
+
+      // Execute tools
+      const toolCalls: ToolCall[] = Object.values(accumulatedToolCalls).map(tc => ({
+        id: tc.id,
+        type: "function" as const,
+        function: { name: tc.name, arguments: tc.arguments }
+      }))
+
+      logger.info({ toolCalls: toolCalls.map(tc => tc.function.name) }, "Executing Qwen tool calls")
+
+      messages.push({
+        role: "assistant",
+        content: textContent || undefined,
+        tool_calls: toolCalls
+      })
+
+      const toolResults = await executeTools(workspacePath, toolCalls)
+
+      // Notify frontend
+      for (const result of toolResults) {
+        const toolEvent = {
+          type: "tool_result",
+          tool_call_id: result.tool_call_id,
+          content: result.content
+        }
+        rawResponse.write(`data: ${JSON.stringify(toolEvent)}\n\n`)
+        messages.push({
+          role: "tool",
+          content: result.content,
+          tool_call_id: result.tool_call_id
+        })
+      }
+    }
+
+    rawResponse.write('data: [DONE]\n\n')
+  }
+
+  // Qwen Chat API - with tool support
  app.post('/api/qwen/chat', {
    schema: {
      body: {
@@ -207,7 +365,9 @@ export async function registerQwenRoutes(
          model: { type: 'string' },
          messages: { type: 'array' },
          stream: { type: 'boolean' },
-          resource_url: { type: 'string' }
+          resource_url: { type: 'string' },
+          workspacePath: { type: 'string' },
+          enableTools: { type: 'boolean' }
        }
      }
    }
@@ -219,58 +379,59 @@ export async function registerQwenRoutes(
      }

      const accessToken = authHeader.substring(7)
-      const { model, messages, stream, resource_url } = request.body as any
+      const { model, messages, stream, resource_url, workspacePath, enableTools } = request.body as any

-      // Use resource_url from OAuth credentials to target the DashScope-compatible API
      const apiBaseUrl = normalizeQwenResourceUrl(resource_url)
      const normalizedModel = normalizeQwenModel(model)
      const chatUrl = `${apiBaseUrl}/chat/completions`

-      logger.info({ chatUrl, model: normalizedModel, messageCount: messages?.length }, "Proxying Qwen chat request")
+      // MCP Tool Loading
+      let allTools = [...CORE_TOOLS]
+      const effectiveWorkspacePath = workspacePath || process.cwd()
+      const toolsEnabled = enableTools !== false

-      const response = await fetch(chatUrl, {
-        method: 'POST',
-        headers: {
-          'Content-Type': 'application/json',
-          'Authorization': `Bearer ${accessToken}`,
-          'Accept': stream ? 'text/event-stream' : 'application/json'
-        },
-        body: JSON.stringify({
-          model: normalizedModel,
-          messages,
-          stream: stream || false
-        })
-      })
-
-      if (!response.ok) {
-        const errorText = await response.text()
-        logger.error({ status: response.status, errorText }, "Qwen chat request failed")
-        return reply.status(response.status).send({ error: "Chat request failed", details: errorText })
+      if (toolsEnabled && effectiveWorkspacePath) {
+        try {
+          const mcpManager = getMcpManager()
+          await mcpManager.loadConfig(effectiveWorkspacePath)
+          const mcpTools = await mcpManager.getToolsAsOpenAIFormat()
+          allTools = [...CORE_TOOLS, ...mcpTools]
+        } catch (mcpError) {
+          logger.warn({ error: mcpError }, "Failed to load MCP tools for Qwen")
+        }
      }

-      if (stream && response.body) {
-        // Stream the response
+      logger.info({ chatUrl, model: normalizedModel, tools: allTools.length }, "Proxying Qwen chat with tools")
+
+      if (stream) {
        reply.raw.writeHead(200, {
          'Content-Type': 'text/event-stream',
          'Cache-Control': 'no-cache',
          'Connection': 'keep-alive',
        })

-        const reader = response.body.getReader()
-        const decoder = new TextDecoder()
-
-        try {
-          while (true) {
-            const { done, value } = await reader.read()
-            if (done) break
-            const chunk = decoder.decode(value, { stream: true })
-            reply.raw.write(chunk)
-          }
-        } finally {
-          reader.releaseLock()
-          reply.raw.end()
-        }
+        await streamWithToolLoop(
+          accessToken,
+          chatUrl,
+          { model: normalizedModel, messages, tools: allTools },
+          effectiveWorkspacePath,
+          toolsEnabled,
+          reply.raw,
+          logger
+        )
      } else {
+        const response = await fetch(chatUrl, {
+          method: 'POST',
+          headers: {
+            'Content-Type': 'application/json',
+            'Authorization': `Bearer ${accessToken}`
+          },
+          body: JSON.stringify({
+            model: normalizedModel,
+            messages,
+            stream: false
+          })
+        })
        const data = await response.json()
        return reply.send(data)
      }
--- a/packages/server/src/server/routes/workspaces.ts
+++ b/packages/server/src/server/routes/workspaces.ts
@@ -2,7 +2,7 @@ import { FastifyInstance, FastifyReply } from "fastify"
 import { spawnSync } from "child_process"
 import { z } from "zod"
 import { existsSync, mkdirSync } from "fs"
-import { cp, readFile, writeFile } from "fs/promises"
+import { cp, readFile, writeFile, stat as readFileStat } from "fs/promises"
 import path from "path"
 import { WorkspaceManager } from "../../workspaces/manager"
 import { InstanceStore } from "../../storage/instance-store"
@@ -257,6 +257,12 @@ export function registerWorkspaceRoutes(app: FastifyInstance, deps: RouteDeps) {
    const configPath = path.join(workspace.path, ".mcp.json")
    try {
      await writeFile(configPath, JSON.stringify(body.config, null, 2), "utf-8")
+
+      // Auto-load MCP config into the manager after saving
+      const { getMcpManager } = await import("../../mcp/client")
+      const mcpManager = getMcpManager()
+      await mcpManager.loadConfig(workspace.path)
+
      return { path: configPath, exists: true, config: body.config }
    } catch (error) {
      request.log.error({ err: error }, "Failed to write MCP config")
@@ -265,6 +271,110 @@ export function registerWorkspaceRoutes(app: FastifyInstance, deps: RouteDeps) {
    }
  })

+  // Get MCP connection status for a workspace
+  app.get<{ Params: { id: string } }>("/api/workspaces/:id/mcp-status", async (request, reply) => {
+    const workspace = deps.workspaceManager.get(request.params.id)
+    if (!workspace) {
+      reply.code(404)
+      return { error: "Workspace not found" }
+    }
+
+    try {
+      const { getMcpManager } = await import("../../mcp/client")
+      const mcpManager = getMcpManager()
+
+      // Load config if not already loaded
+      await mcpManager.loadConfig(workspace.path)
+
+      const status = mcpManager.getStatus()
+      const tools = await mcpManager.getAllTools()
+
+      return {
+        servers: status,
+        toolCount: tools.length,
+        tools: tools.map(t => ({ name: t.name, server: t.serverName, description: t.description }))
+      }
+    } catch (error) {
+      request.log.error({ err: error }, "Failed to get MCP status")
+      reply.code(500)
+      return { error: "Failed to get MCP status" }
+    }
+  })
+
+  // Connect all configured MCPs for a workspace
+  app.post<{ Params: { id: string } }>("/api/workspaces/:id/mcp-connect", async (request, reply) => {
+    const workspace = deps.workspaceManager.get(request.params.id)
+    if (!workspace) {
+      reply.code(404)
+      return { error: "Workspace not found" }
+    }
+
+    try {
+      const { getMcpManager } = await import("../../mcp/client")
+      const mcpManager = getMcpManager()
+
+      // Load and connect to all configured MCPs
+      await mcpManager.loadConfig(workspace.path)
+
+      // Get the tools to trigger connections
+      const tools = await mcpManager.getAllTools()
+      const status = mcpManager.getStatus()
+
+      return {
+        success: true,
+        servers: status,
+        toolCount: tools.length
+      }
+    } catch (error) {
+      request.log.error({ err: error }, "Failed to connect MCPs")
+      reply.code(500)
+      return { error: "Failed to connect MCPs" }
+    }
+  })
+
+  app.post<{
+    Params: { id: string }
+    Body: { name: string; description?: string; systemPrompt: string; mode?: string }
+  }>("/api/workspaces/:id/agents", async (request, reply) => {
+    const workspace = deps.workspaceManager.get(request.params.id)
+    if (!workspace) {
+      reply.code(404)
+      return { error: "Workspace not found" }
+    }
+
+    const { name, description, systemPrompt } = request.body
+    if (!name || !systemPrompt) {
+      reply.code(400)
+      return { error: "Name and systemPrompt are required" }
+    }
+
+    try {
+      const data = await deps.instanceStore.read(workspace.path)
+      const customAgents = data.customAgents || []
+
+      // Update existing or add new
+      const existingIndex = customAgents.findIndex(a => a.name === name)
+      const agentData = { name, description, prompt: systemPrompt }
+
+      if (existingIndex >= 0) {
+        customAgents[existingIndex] = agentData
+      } else {
+        customAgents.push(agentData)
+      }
+
+      await deps.instanceStore.write(workspace.path, {
+        ...data,
+        customAgents
+      })
+
+      return { success: true, agent: agentData }
+    } catch (error) {
+      request.log.error({ err: error }, "Failed to save custom agent")
+      reply.code(500)
+      return { error: "Failed to save custom agent" }
+    }
+  })
+
  app.post<{
    Body: { source: string; destination: string; includeConfig?: boolean }
  }>("/api/workspaces/import", async (request, reply) => {
@@ -308,6 +418,53 @@ export function registerWorkspaceRoutes(app: FastifyInstance, deps: RouteDeps) {

    return workspace
  })
+
+  // Serve static files from workspace for preview
+  app.get<{ Params: { id: string; "*": string } }>("/api/workspaces/:id/serve/*", async (request, reply) => {
+    const workspace = deps.workspaceManager.get(request.params.id)
+    if (!workspace) {
+      reply.code(404)
+      return { error: "Workspace not found" }
+    }
+
+    const relativePath = request.params["*"]
+    const filePath = path.join(workspace.path, relativePath)
+
+    // Security check: ensure file is within workspace.path
+    if (!filePath.startsWith(workspace.path)) {
+      reply.code(403)
+      return { error: "Access denied" }
+    }
+
+    if (!existsSync(filePath)) {
+      reply.code(404)
+      return { error: "File not found" }
+    }
+
+    const stat = await readFileStat(filePath)
+    if (!stat.isFile()) {
+      reply.code(400)
+      return { error: "Not a file" }
+    }
+
+    const ext = path.extname(filePath).toLowerCase()
+    const mimeTypes: Record<string, string> = {
+      ".html": "text/html",
+      ".htm": "text/html",
+      ".js": "application/javascript",
+      ".css": "text/css",
+      ".json": "application/json",
+      ".png": "image/png",
+      ".jpg": "image/jpeg",
+      ".jpeg": "image/jpeg",
+      ".gif": "image/gif",
+      ".svg": "image/svg+xml",
+      ".txt": "text/plain",
+    }
+
+    reply.type(mimeTypes[ext] || "application/octet-stream")
+    return await readFile(filePath)
+  })
 }


--- a/packages/server/src/server/routes/zai.ts
+++ b/packages/server/src/server/routes/zai.ts
@@ -1,9 +1,11 @@
 import { FastifyInstance } from "fastify"
-import { ZAIClient, ZAI_MODELS, type ZAIConfig, type ZAIChatRequest, ZAIChatRequestSchema } from "../../integrations/zai-api"
+import { ZAIClient, ZAI_MODELS, type ZAIConfig, type ZAIChatRequest, type ZAIMessage } from "../../integrations/zai-api"
 import { Logger } from "../../logger"
 import { existsSync, readFileSync, writeFileSync, mkdirSync } from "fs"
 import { join } from "path"
 import { getUserIntegrationsDir } from "../../user-data"
+import { CORE_TOOLS, executeTools, type ToolCall, type ToolResult } from "../../tools/executor"
+import { getMcpManager } from "../../mcp/client"

 interface ZAIRouteDeps {
    logger: Logger
@@ -12,6 +14,9 @@ interface ZAIRouteDeps {
 const CONFIG_DIR = getUserIntegrationsDir()
 const CONFIG_FILE = join(CONFIG_DIR, "zai-config.json")

+// Maximum number of tool execution loops to prevent infinite recursion
+const MAX_TOOL_LOOPS = 10
+
 export async function registerZAIRoutes(
    app: FastifyInstance,
    deps: ZAIRouteDeps
@@ -75,7 +80,7 @@ export async function registerZAIRoutes(
        }
    })

-    // Chat completion endpoint
+    // Chat completion endpoint WITH MCP TOOL SUPPORT
    app.post('/api/zai/chat', async (request, reply) => {
        try {
            const config = getZAIConfig()
@@ -84,9 +89,46 @@ export async function registerZAIRoutes(
            }

            const client = new ZAIClient(config)
-            const chatRequest = request.body as ZAIChatRequest
+            const chatRequest = request.body as ZAIChatRequest & {
+                workspacePath?: string
+                enableTools?: boolean
+            }

-            // Handle streaming
+            // Extract workspace path for tool execution
+            // IMPORTANT: workspacePath must be provided by frontend, otherwise tools write to server directory
+            const workspacePath = chatRequest.workspacePath || process.cwd()
+            const enableTools = chatRequest.enableTools !== false // Default to true
+
+            logger.info({
+                workspacePath,
+                receivedWorkspacePath: chatRequest.workspacePath,
+                enableTools
+            }, "Z.AI chat request received")
+
+            // Load MCP tools from workspace config
+            let allTools = [...CORE_TOOLS]
+            if (enableTools && workspacePath) {
+                try {
+                    const mcpManager = getMcpManager()
+                    await mcpManager.loadConfig(workspacePath)
+                    const mcpTools = await mcpManager.getToolsAsOpenAIFormat()
+                    allTools = [...CORE_TOOLS, ...mcpTools]
+                    if (mcpTools.length > 0) {
+                        logger.info({ mcpToolCount: mcpTools.length }, "Loaded MCP tools")
+                    }
+                } catch (mcpError) {
+                    logger.warn({ error: mcpError }, "Failed to load MCP tools, using core tools only")
+                }
+            }
+
+            // Inject tools into request if enabled
+            const requestWithTools: ZAIChatRequest = {
+                ...chatRequest,
+                tools: enableTools ? allTools : undefined,
+                tool_choice: enableTools ? "auto" : undefined
+            }
+
+            // Handle streaming with tool execution loop
            if (chatRequest.stream) {
                reply.raw.writeHead(200, {
                    'Content-Type': 'text/event-stream',
@@ -95,17 +137,14 @@ export async function registerZAIRoutes(
                })

                try {
-                    for await (const chunk of client.chatStream(chatRequest)) {
-                        reply.raw.write(`data: ${JSON.stringify(chunk)}\n\n`)
-
-                        // Check for finish_reason to end stream
-                        const finishReason = chunk.choices[0]?.finish_reason
-                        if (finishReason) {
-                            reply.raw.write('data: [DONE]\n\n')
-                            break
-                        }
-                    }
-
+                    await streamWithToolLoop(
+                        client,
+                        requestWithTools,
+                        workspacePath,
+                        enableTools,
+                        reply.raw,
+                        logger
+                    )
                    reply.raw.end()
                } catch (streamError) {
                    logger.error({ error: streamError }, "Z.AI streaming failed")
@@ -113,7 +152,14 @@ export async function registerZAIRoutes(
                    reply.raw.end()
                }
            } else {
-                const response = await client.chat(chatRequest)
+                // Non-streaming with tool loop
+                const response = await chatWithToolLoop(
+                    client,
+                    requestWithTools,
+                    workspacePath,
+                    enableTools,
+                    logger
+                )
                return response
            }
        } catch (error) {
@@ -122,7 +168,184 @@ export async function registerZAIRoutes(
        }
    })

-    logger.info("Z.AI routes registered")
+    logger.info("Z.AI routes registered with MCP tool support")
+}
+
+/**
+ * Streaming chat with tool execution loop
+ */
+async function streamWithToolLoop(
+    client: ZAIClient,
+    request: ZAIChatRequest,
+    workspacePath: string,
+    enableTools: boolean,
+    rawResponse: any,
+    logger: Logger
+): Promise<void> {
+    let messages = [...request.messages]
+    let loopCount = 0
+
+    while (loopCount < MAX_TOOL_LOOPS) {
+        loopCount++
+
+        // Accumulate tool calls from stream
+        let accumulatedToolCalls: { [index: number]: { id: string; name: string; arguments: string } } = {}
+        let hasToolCalls = false
+        let textContent = ""
+
+        // Stream response
+        for await (const chunk of client.chatStream({ ...request, messages })) {
+            // Write chunk to client
+            rawResponse.write(`data: ${JSON.stringify(chunk)}\n\n`)
+
+            const choice = chunk.choices[0]
+            if (!choice) continue
+
+            // Accumulate text content
+            if (choice.delta?.content) {
+                textContent += choice.delta.content
+            }
+
+            // Accumulate tool calls from delta
+            if (choice.delta?.tool_calls) {
+                hasToolCalls = true
+                for (const tc of choice.delta.tool_calls) {
+                    const idx = tc.index ?? 0
+                    if (!accumulatedToolCalls[idx]) {
+                        accumulatedToolCalls[idx] = { id: tc.id || "", name: "", arguments: "" }
+                    }
+                    if (tc.id) accumulatedToolCalls[idx].id = tc.id
+                    if (tc.function?.name) accumulatedToolCalls[idx].name += tc.function.name
+                    if (tc.function?.arguments) accumulatedToolCalls[idx].arguments += tc.function.arguments
+                }
+            }
+
+            // Check if we should stop
+            if (choice.finish_reason === "stop") {
+                rawResponse.write('data: [DONE]\n\n')
+                return
+            }
+        }
+
+        // If no tool calls, we're done
+        if (!hasToolCalls || !enableTools) {
+            rawResponse.write('data: [DONE]\n\n')
+            return
+        }
+
+        // Convert accumulated tool calls
+        const toolCalls: ToolCall[] = Object.values(accumulatedToolCalls).map(tc => ({
+            id: tc.id,
+            type: "function" as const,
+            function: {
+                name: tc.name,
+                arguments: tc.arguments
+            }
+        }))
+
+        if (toolCalls.length === 0) {
+            rawResponse.write('data: [DONE]\n\n')
+            return
+        }
+
+        logger.info({ toolCalls: toolCalls.map(tc => tc.function.name) }, "Executing tool calls")
+
+        // Add assistant message with tool calls
+        const assistantMessage: ZAIMessage = {
+            role: "assistant",
+            content: textContent || undefined,
+            tool_calls: toolCalls
+        }
+        messages.push(assistantMessage)
+
+        // Execute tools
+        const toolResults = await executeTools(workspacePath, toolCalls)
+
+        // Notify client about tool execution via special event
+        for (const result of toolResults) {
+            const toolEvent = {
+                type: "tool_result",
+                tool_call_id: result.tool_call_id,
+                content: result.content
+            }
+            rawResponse.write(`data: ${JSON.stringify(toolEvent)}\n\n`)
+        }
+
+        // Add tool results to messages
+        for (const result of toolResults) {
+            const toolMessage: ZAIMessage = {
+                role: "tool",
+                content: result.content,
+                tool_call_id: result.tool_call_id
+            }
+            messages.push(toolMessage)
+        }
+
+        logger.info({ loopCount, toolsExecuted: toolResults.length }, "Tool loop iteration complete")
+    }
+
+    logger.warn({ loopCount }, "Max tool loops reached")
+    rawResponse.write('data: [DONE]\n\n')
+}
+
+/**
+ * Non-streaming chat with tool execution loop
+ */
+async function chatWithToolLoop(
+    client: ZAIClient,
+    request: ZAIChatRequest,
+    workspacePath: string,
+    enableTools: boolean,
+    logger: Logger
+): Promise<any> {
+    let messages = [...request.messages]
+    let loopCount = 0
+    let lastResponse: any = null
+
+    while (loopCount < MAX_TOOL_LOOPS) {
+        loopCount++
+
+        const response = await client.chat({ ...request, messages, stream: false })
+        lastResponse = response
+
+        const choice = response.choices[0]
+        if (!choice) break
+
+        const toolCalls = choice.message?.tool_calls
+
+        // If no tool calls or finish_reason is "stop", return
+        if (!toolCalls || toolCalls.length === 0 || !enableTools) {
+            return response
+        }
+
+        logger.info({ toolCalls: toolCalls.map((tc: any) => tc.function.name) }, "Executing tool calls")
+
+        // Add assistant message
+        const assistantMessage: ZAIMessage = {
+            role: "assistant",
+            content: choice.message.content || undefined,
+            tool_calls: toolCalls
+        }
+        messages.push(assistantMessage)
+
+        // Execute tools
+        const toolResults = await executeTools(workspacePath, toolCalls)
+
+        // Add tool results
+        for (const result of toolResults) {
+            const toolMessage: ZAIMessage = {
+                role: "tool",
+                content: result.content,
+                tool_call_id: result.tool_call_id
+            }
+            messages.push(toolMessage)
+        }
+
+        logger.info({ loopCount, toolsExecuted: toolResults.length }, "Tool loop iteration complete")
+    }
+
+    logger.warn({ loopCount }, "Max tool loops reached")
+    return lastResponse
 }

 function getZAIConfig(): ZAIConfig {
@@ -131,9 +354,9 @@ function getZAIConfig(): ZAIConfig {
            const data = readFileSync(CONFIG_FILE, 'utf-8')
            return JSON.parse(data)
        }
-        return { enabled: false, endpoint: "https://api.z.ai/api/paas/v4", timeout: 300000 }
+        return { enabled: false, endpoint: "https://api.z.ai/api/coding/paas/v4", timeout: 300000 }
    } catch {
-        return { enabled: false, endpoint: "https://api.z.ai/api/paas/v4", timeout: 300000 }
+        return { enabled: false, endpoint: "https://api.z.ai/api/coding/paas/v4", timeout: 300000 }
    }
 }

--- a/packages/server/src/storage/session-store.ts
+++ b/packages/server/src/storage/session-store.ts
@@ -0,0 +1,284 @@
+/**
+ * Session Store - Native session management without OpenCode binary
+ * 
+ * This provides a complete replacement for OpenCode's session management,
+ * allowing NomadArch to work in "Binary-Free Mode".
+ */
+
+import { readFile, writeFile, mkdir } from "fs/promises"
+import { existsSync } from "fs"
+import path from "path"
+import { ulid } from "ulid"
+import { createLogger } from "../logger"
+
+const log = createLogger({ component: "session-store" })
+
+// Types matching OpenCode's schema for compatibility
+export interface SessionMessage {
+    id: string
+    sessionId: string
+    role: "user" | "assistant" | "system" | "tool"
+    content?: string
+    parts?: MessagePart[]
+    createdAt: number
+    updatedAt: number
+    toolCalls?: ToolCall[]
+    toolCallId?: string
+    status?: "pending" | "streaming" | "completed" | "error"
+}
+
+export interface MessagePart {
+    type: "text" | "tool_call" | "tool_result" | "thinking" | "code"
+    content?: string
+    toolCall?: ToolCall
+    toolResult?: ToolResult
+}
+
+export interface ToolCall {
+    id: string
+    type: "function"
+    function: {
+        name: string
+        arguments: string
+    }
+}
+
+export interface ToolResult {
+    toolCallId: string
+    content: string
+    isError?: boolean
+}
+
+export interface Session {
+    id: string
+    workspaceId: string
+    title?: string
+    parentId?: string | null
+    createdAt: number
+    updatedAt: number
+    messageIds: string[]
+    model?: {
+        providerId: string
+        modelId: string
+    }
+    agent?: string
+    revert?: {
+        messageID: string
+        reason?: string
+    } | null
+}
+
+export interface SessionStore {
+    sessions: Record<string, Session>
+    messages: Record<string, SessionMessage>
+}
+
+/**
+ * Native session management for Binary-Free Mode
+ */
+export class NativeSessionManager {
+    private stores = new Map<string, SessionStore>()
+    private dataDir: string
+
+    constructor(dataDir: string) {
+        this.dataDir = dataDir
+    }
+
+    private getStorePath(workspaceId: string): string {
+        return path.join(this.dataDir, workspaceId, "sessions.json")
+    }
+
+    private async ensureDir(workspaceId: string): Promise<void> {
+        const dir = path.join(this.dataDir, workspaceId)
+        if (!existsSync(dir)) {
+            await mkdir(dir, { recursive: true })
+        }
+    }
+
+    private async loadStore(workspaceId: string): Promise<SessionStore> {
+        if (this.stores.has(workspaceId)) {
+            return this.stores.get(workspaceId)!
+        }
+
+        const storePath = this.getStorePath(workspaceId)
+        let store: SessionStore = { sessions: {}, messages: {} }
+
+        if (existsSync(storePath)) {
+            try {
+                const data = await readFile(storePath, "utf-8")
+                store = JSON.parse(data)
+            } catch (error) {
+                log.error({ workspaceId, error }, "Failed to load session store")
+            }
+        }
+
+        this.stores.set(workspaceId, store)
+        return store
+    }
+
+    private async saveStore(workspaceId: string): Promise<void> {
+        const store = this.stores.get(workspaceId)
+        if (!store) return
+
+        await this.ensureDir(workspaceId)
+        const storePath = this.getStorePath(workspaceId)
+        await writeFile(storePath, JSON.stringify(store, null, 2), "utf-8")
+    }
+
+    // Session CRUD operations
+
+    async listSessions(workspaceId: string): Promise<Session[]> {
+        const store = await this.loadStore(workspaceId)
+        return Object.values(store.sessions).sort((a, b) => b.updatedAt - a.updatedAt)
+    }
+
+    async getSession(workspaceId: string, sessionId: string): Promise<Session | null> {
+        const store = await this.loadStore(workspaceId)
+        return store.sessions[sessionId] ?? null
+    }
+
+    async createSession(workspaceId: string, options?: {
+        title?: string
+        parentId?: string
+        model?: { providerId: string; modelId: string }
+        agent?: string
+    }): Promise<Session> {
+        const store = await this.loadStore(workspaceId)
+        const now = Date.now()
+
+        const session: Session = {
+            id: ulid(),
+            workspaceId,
+            title: options?.title ?? "New Session",
+            parentId: options?.parentId ?? null,
+            createdAt: now,
+            updatedAt: now,
+            messageIds: [],
+            model: options?.model,
+            agent: options?.agent,
+        }
+
+        store.sessions[session.id] = session
+        await this.saveStore(workspaceId)
+
+        log.info({ workspaceId, sessionId: session.id }, "Created new session")
+        return session
+    }
+
+    async updateSession(workspaceId: string, sessionId: string, updates: Partial<Session>): Promise<Session | null> {
+        const store = await this.loadStore(workspaceId)
+        const session = store.sessions[sessionId]
+        if (!session) return null
+
+        const updated = {
+            ...session,
+            ...updates,
+            id: session.id, // Prevent ID change
+            workspaceId: session.workspaceId, // Prevent workspace change
+            updatedAt: Date.now(),
+        }
+
+        store.sessions[sessionId] = updated
+        await this.saveStore(workspaceId)
+        return updated
+    }
+
+    async deleteSession(workspaceId: string, sessionId: string): Promise<boolean> {
+        const store = await this.loadStore(workspaceId)
+        const session = store.sessions[sessionId]
+        if (!session) return false
+
+        // Delete all messages in the session
+        for (const messageId of session.messageIds) {
+            delete store.messages[messageId]
+        }
+
+        delete store.sessions[sessionId]
+        await this.saveStore(workspaceId)
+
+        log.info({ workspaceId, sessionId }, "Deleted session")
+        return true
+    }
+
+    // Message operations
+
+    async getSessionMessages(workspaceId: string, sessionId: string): Promise<SessionMessage[]> {
+        const store = await this.loadStore(workspaceId)
+        const session = store.sessions[sessionId]
+        if (!session) return []
+
+        return session.messageIds
+            .map(id => store.messages[id])
+            .filter((msg): msg is SessionMessage => msg !== undefined)
+    }
+
+    async addMessage(workspaceId: string, sessionId: string, message: Omit<SessionMessage, "id" | "sessionId" | "createdAt" | "updatedAt">): Promise<SessionMessage> {
+        const store = await this.loadStore(workspaceId)
+        const session = store.sessions[sessionId]
+        if (!session) throw new Error(`Session not found: ${sessionId}`)
+
+        const now = Date.now()
+        const newMessage: SessionMessage = {
+            ...message,
+            id: ulid(),
+            sessionId,
+            createdAt: now,
+            updatedAt: now,
+        }
+
+        store.messages[newMessage.id] = newMessage
+        session.messageIds.push(newMessage.id)
+        session.updatedAt = now
+
+        await this.saveStore(workspaceId)
+        return newMessage
+    }
+
+    async updateMessage(workspaceId: string, messageId: string, updates: Partial<SessionMessage>): Promise<SessionMessage | null> {
+        const store = await this.loadStore(workspaceId)
+        const message = store.messages[messageId]
+        if (!message) return null
+
+        const updated = {
+            ...message,
+            ...updates,
+            id: message.id, // Prevent ID change
+            sessionId: message.sessionId, // Prevent session change
+            updatedAt: Date.now(),
+        }
+
+        store.messages[messageId] = updated
+        await this.saveStore(workspaceId)
+        return updated
+    }
+
+    // Utility
+
+    async clearWorkspace(workspaceId: string): Promise<void> {
+        this.stores.delete(workspaceId)
+        // Optionally delete file
+    }
+
+    getActiveSessionCount(workspaceId: string): number {
+        const store = this.stores.get(workspaceId)
+        return store ? Object.keys(store.sessions).length : 0
+    }
+}
+
+// Singleton instance
+let sessionManager: NativeSessionManager | null = null
+
+export function getSessionManager(dataDir?: string): NativeSessionManager {
+    if (!sessionManager) {
+        if (!dataDir) {
+            throw new Error("Session manager not initialized - provide dataDir")
+        }
+        sessionManager = new NativeSessionManager(dataDir)
+    }
+    return sessionManager
+}
+
+export function initSessionManager(dataDir: string): NativeSessionManager {
+    sessionManager = new NativeSessionManager(dataDir)
+    return sessionManager
+}
--- a/packages/server/src/tools/executor.ts
+++ b/packages/server/src/tools/executor.ts
@@ -0,0 +1,352 @@
+/**
+ * Tool Executor Service
+ * Provides MCP-compatible tool definitions and execution for all AI models.
+ * This enables Z.AI, Qwen, OpenCode Zen, etc. to write files, read files, and interact with the workspace.
+ */
+
+import fs from "fs"
+import path from "path"
+import { createLogger } from "../logger"
+import { getMcpManager } from "../mcp/client"
+
+const log = createLogger({ component: "tool-executor" })
+
+// OpenAI-compatible Tool Definition Schema
+export interface ToolDefinition {
+    type: "function"
+    function: {
+        name: string
+        description: string
+        parameters: {
+            type: "object"
+            properties: Record<string, { type: string; description?: string }>
+            required?: string[]
+        }
+    }
+}
+
+// Tool Call from LLM Response
+export interface ToolCall {
+    id: string
+    type: "function"
+    function: {
+        name: string
+        arguments: string // JSON string
+    }
+}
+
+// Tool Execution Result
+export interface ToolResult {
+    tool_call_id: string
+    role: "tool"
+    content: string
+}
+
+/**
+ * Core Tool Definitions for MCP
+ * These follow OpenAI's function calling schema (compatible with Z.AI GLM-4)
+ */
+export const CORE_TOOLS: ToolDefinition[] = [
+    {
+        type: "function",
+        function: {
+            name: "write_file",
+            description: "Write content to a file in the workspace. Creates the file if it doesn't exist, or overwrites if it does. Use this to generate code files, configuration, or any text content.",
+            parameters: {
+                type: "object",
+                properties: {
+                    path: {
+                        type: "string",
+                        description: "Relative path to the file within the workspace (e.g., 'src/components/Button.tsx')"
+                    },
+                    content: {
+                        type: "string",
+                        description: "The full content to write to the file"
+                    }
+                },
+                required: ["path", "content"]
+            }
+        }
+    },
+    {
+        type: "function",
+        function: {
+            name: "read_file",
+            description: "Read the contents of a file from the workspace.",
+            parameters: {
+                type: "object",
+                properties: {
+                    path: {
+                        type: "string",
+                        description: "Relative path to the file within the workspace"
+                    }
+                },
+                required: ["path"]
+            }
+        }
+    },
+    {
+        type: "function",
+        function: {
+            name: "list_files",
+            description: "List files and directories in a workspace directory.",
+            parameters: {
+                type: "object",
+                properties: {
+                    path: {
+                        type: "string",
+                        description: "Relative path to the directory (use '.' for root)"
+                    }
+                },
+                required: ["path"]
+            }
+        }
+    },
+    {
+        type: "function",
+        function: {
+            name: "create_directory",
+            description: "Create a directory in the workspace. Creates parent directories if needed.",
+            parameters: {
+                type: "object",
+                properties: {
+                    path: {
+                        type: "string",
+                        description: "Relative path to the directory to create"
+                    }
+                },
+                required: ["path"]
+            }
+        }
+    },
+    {
+        type: "function",
+        function: {
+            name: "delete_file",
+            description: "Delete a file from the workspace.",
+            parameters: {
+                type: "object",
+                properties: {
+                    path: {
+                        type: "string",
+                        description: "Relative path to the file to delete"
+                    }
+                },
+                required: ["path"]
+            }
+        }
+    }
+]
+
+/**
+ * Execute a tool call within a workspace context
+ */
+export async function executeTool(
+    workspacePath: string,
+    toolCall: ToolCall
+): Promise<ToolResult> {
+    const { id, function: fn } = toolCall
+    const name = fn.name
+    let args: Record<string, unknown>
+
+    try {
+        args = JSON.parse(fn.arguments)
+    } catch (e) {
+        return {
+            tool_call_id: id,
+            role: "tool",
+            content: `Error: Failed to parse tool arguments: ${fn.arguments}`
+        }
+    }
+
+    log.info({ tool: name, args, workspacePath }, "Executing tool")
+
+    try {
+        switch (name) {
+            case "write_file": {
+                const relativePath = String(args.path || "")
+                const content = String(args.content || "")
+                const fullPath = path.resolve(workspacePath, relativePath)
+
+                // Security check: ensure we're still within workspace
+                if (!fullPath.startsWith(path.resolve(workspacePath))) {
+                    return {
+                        tool_call_id: id,
+                        role: "tool",
+                        content: `Error: Path escapes workspace boundary: ${relativePath}`
+                    }
+                }
+
+                // Ensure parent directory exists
+                const dir = path.dirname(fullPath)
+                if (!fs.existsSync(dir)) {
+                    fs.mkdirSync(dir, { recursive: true })
+                }
+
+                fs.writeFileSync(fullPath, content, "utf-8")
+                log.info({ path: relativePath, bytes: content.length }, "File written successfully")
+                return {
+                    tool_call_id: id,
+                    role: "tool",
+                    content: `Successfully wrote ${content.length} bytes to ${relativePath}`
+                }
+            }
+
+            case "read_file": {
+                const relativePath = String(args.path || "")
+                const fullPath = path.resolve(workspacePath, relativePath)
+
+                if (!fullPath.startsWith(path.resolve(workspacePath))) {
+                    return {
+                        tool_call_id: id,
+                        role: "tool",
+                        content: `Error: Path escapes workspace boundary: ${relativePath}`
+                    }
+                }
+
+                if (!fs.existsSync(fullPath)) {
+                    return {
+                        tool_call_id: id,
+                        role: "tool",
+                        content: `Error: File not found: ${relativePath}`
+                    }
+                }
+
+                const content = fs.readFileSync(fullPath, "utf-8")
+                return {
+                    tool_call_id: id,
+                    role: "tool",
+                    content: content.slice(0, 50000) // Limit to prevent context overflow
+                }
+            }
+
+            case "list_files": {
+                const relativePath = String(args.path || ".")
+                const fullPath = path.resolve(workspacePath, relativePath)
+
+                if (!fullPath.startsWith(path.resolve(workspacePath))) {
+                    return {
+                        tool_call_id: id,
+                        role: "tool",
+                        content: `Error: Path escapes workspace boundary: ${relativePath}`
+                    }
+                }
+
+                if (!fs.existsSync(fullPath)) {
+                    return {
+                        tool_call_id: id,
+                        role: "tool",
+                        content: `Error: Directory not found: ${relativePath}`
+                    }
+                }
+
+                const entries = fs.readdirSync(fullPath, { withFileTypes: true })
+                const listing = entries.map(e =>
+                    e.isDirectory() ? `${e.name}/` : e.name
+                ).join("\n")
+
+                return {
+                    tool_call_id: id,
+                    role: "tool",
+                    content: listing || "(empty directory)"
+                }
+            }
+
+            case "create_directory": {
+                const relativePath = String(args.path || "")
+                const fullPath = path.resolve(workspacePath, relativePath)
+
+                if (!fullPath.startsWith(path.resolve(workspacePath))) {
+                    return {
+                        tool_call_id: id,
+                        role: "tool",
+                        content: `Error: Path escapes workspace boundary: ${relativePath}`
+                    }
+                }
+
+                fs.mkdirSync(fullPath, { recursive: true })
+                return {
+                    tool_call_id: id,
+                    role: "tool",
+                    content: `Successfully created directory: ${relativePath}`
+                }
+            }
+
+            case "delete_file": {
+                const relativePath = String(args.path || "")
+                const fullPath = path.resolve(workspacePath, relativePath)
+
+                if (!fullPath.startsWith(path.resolve(workspacePath))) {
+                    return {
+                        tool_call_id: id,
+                        role: "tool",
+                        content: `Error: Path escapes workspace boundary: ${relativePath}`
+                    }
+                }
+
+                if (!fs.existsSync(fullPath)) {
+                    return {
+                        tool_call_id: id,
+                        role: "tool",
+                        content: `Error: File not found: ${relativePath}`
+                    }
+                }
+
+                fs.unlinkSync(fullPath)
+                return {
+                    tool_call_id: id,
+                    role: "tool",
+                    content: `Successfully deleted: ${relativePath}`
+                }
+            }
+
+            default: {
+                // Check if this is an MCP tool (format: mcp_servername_toolname)
+                if (name.startsWith("mcp_")) {
+                    try {
+                        const mcpManager = getMcpManager()
+                        const result = await mcpManager.executeTool(name, args)
+                        return {
+                            tool_call_id: id,
+                            role: "tool",
+                            content: result
+                        }
+                    } catch (mcpError) {
+                        const message = mcpError instanceof Error ? mcpError.message : String(mcpError)
+                        return {
+                            tool_call_id: id,
+                            role: "tool",
+                            content: `MCP tool error: ${message}`
+                        }
+                    }
+                }
+
+                return {
+                    tool_call_id: id,
+                    role: "tool",
+                    content: `Error: Unknown tool: ${name}`
+                }
+            }
+        }
+    } catch (error) {
+        const message = error instanceof Error ? error.message : String(error)
+        log.error({ tool: name, error: message }, "Tool execution failed")
+        return {
+            tool_call_id: id,
+            role: "tool",
+            content: `Error executing ${name}: ${message}`
+        }
+    }
+}
+
+/**
+ * Execute multiple tool calls in parallel
+ */
+export async function executeTools(
+    workspacePath: string,
+    toolCalls: ToolCall[]
+): Promise<ToolResult[]> {
+    return Promise.all(
+        toolCalls.map(tc => executeTool(workspacePath, tc))
+    )
+}
--- a/packages/server/src/tools/index.ts
+++ b/packages/server/src/tools/index.ts
@@ -0,0 +1,13 @@
+/**
+ * Tools Module Index
+ * Exports MCP-compatible tool definitions and executor for AI agent integration.
+ */
+
+export {
+    CORE_TOOLS,
+    executeTool,
+    executeTools,
+    type ToolDefinition,
+    type ToolCall,
+    type ToolResult
+} from "./executor"
--- a/packages/server/src/workspaces/runtime.ts
+++ b/packages/server/src/workspaces/runtime.ts
@@ -28,7 +28,7 @@ interface ManagedProcess {
 export class WorkspaceRuntime {
  private processes = new Map<string, ManagedProcess>()

-  constructor(private readonly eventBus: EventBus, private readonly logger: Logger) {}
+  constructor(private readonly eventBus: EventBus, private readonly logger: Logger) { }

  async launch(options: LaunchOptions): Promise<{ pid: number; port: number; exitPromise: Promise<ProcessExitInfo>; getLastOutput: () => string }> {
    this.validateFolder(options.folder)
@@ -58,7 +58,23 @@ export class WorkspaceRuntime {
    const exitPromise = new Promise<ProcessExitInfo>((resolveExit) => {
      exitResolve = resolveExit
    })
-    let lastOutput = ""
+
+    // Store recent output for debugging - keep last 20 lines from each stream
+    const MAX_OUTPUT_LINES = 20
+    const recentStdout: string[] = []
+    const recentStderr: string[] = []
+    const getLastOutput = () => {
+      const combined: string[] = []
+      if (recentStderr.length > 0) {
+        combined.push("=== STDERR ===")
+        combined.push(...recentStderr.slice(-10))
+      }
+      if (recentStdout.length > 0) {
+        combined.push("=== STDOUT ===")
+        combined.push(...recentStdout.slice(-10))
+      }
+      return combined.join("\n")
+    }

    return new Promise((resolve, reject) => {
      this.logger.info(
@@ -149,23 +165,28 @@ export class WorkspaceRuntime {
        for (const line of lines) {
          const trimmed = line.trim()
          if (!trimmed) continue
-          lastOutput = trimmed
+
+          // Store in recent buffer for debugging
+          recentStdout.push(trimmed)
+          if (recentStdout.length > MAX_OUTPUT_LINES) {
+            recentStdout.shift()
+          }
+
          this.emitLog(options.workspaceId, "info", line)

          if (!portFound) {
            this.logger.debug({ workspaceId: options.workspaceId, line: trimmed }, "OpenCode output line")
            // Try multiple patterns for port detection
            const portMatch = line.match(/opencode server listening on http:\/\/.+:(\d+)/i) ||
-                             line.match(/server listening on http:\/\/.+:(\d+)/i) ||
-                             line.match(/listening on http:\/\/.+:(\d+)/i) ||
-                             line.match(/:(\d+)/i)
+              line.match(/server listening on http:\/\/.+:(\d+)/i) ||
+              line.match(/listening on http:\/\/.+:(\d+)/i) ||
+              line.match(/:(\d+)/i)

            if (portMatch) {
              portFound = true
              child.removeListener("error", handleError)
              const port = parseInt(portMatch[1], 10)
              this.logger.info({ workspaceId: options.workspaceId, port, matchedLine: trimmed }, "Workspace runtime allocated port - PORT DETECTED")
-              const getLastOutput = () => lastOutput.trim()
              resolve({ pid: child.pid!, port, exitPromise, getLastOutput })
            } else {
              this.logger.debug({ workspaceId: options.workspaceId, line: trimmed }, "Port detection - no match in this line")
@@ -183,7 +204,13 @@ export class WorkspaceRuntime {
        for (const line of lines) {
          const trimmed = line.trim()
          if (!trimmed) continue
-          lastOutput = `[stderr] ${trimmed}`
+
+          // Store in recent buffer for debugging
+          recentStderr.push(trimmed)
+          if (recentStderr.length > MAX_OUTPUT_LINES) {
+            recentStderr.shift()
+          }
+
          this.emitLog(options.workspaceId, "error", line)
        }
      })
--- a/packages/ui/src/components/chat/minimal-chat.tsx
+++ b/packages/ui/src/components/chat/minimal-chat.tsx
@@ -0,0 +1,320 @@
+/**
+ * MINIMAL CHAT BYPASS
+ * 
+ * This is a stripped-down chat component that:
+ * - Uses minimal store access (just for model/session info)
+ * - Makes direct fetch calls
+ * - Has NO complex effects/memos
+ * - Renders messages as a simple list
+ * 
+ * Purpose: Test if the UI responsiveness issue is in the
+ * reactivity system or something else entirely.
+ */
+
+import { createSignal, For, Show, onMount } from "solid-js"
+import { sessions } from "@/stores/session-state"
+
+interface Message {
+    id: string
+    role: "user" | "assistant"
+    content: string
+    timestamp: number
+    status: "sending" | "streaming" | "complete" | "error"
+}
+
+interface MinimalChatProps {
+    instanceId: string
+    sessionId: string
+}
+
+export function MinimalChat(props: MinimalChatProps) {
+    const [messages, setMessages] = createSignal<Message[]>([])
+    const [inputText, setInputText] = createSignal("")
+    const [isLoading, setIsLoading] = createSignal(false)
+    const [error, setError] = createSignal<string | null>(null)
+    const [currentModel, setCurrentModel] = createSignal("minimax-m1")
+
+    let scrollContainer: HTMLDivElement | undefined
+    let inputRef: HTMLTextAreaElement | undefined
+
+    function generateId() {
+        return `msg_${Date.now()}_${Math.random().toString(36).slice(2, 9)}`
+    }
+
+    function scrollToBottom() {
+        if (scrollContainer) {
+            scrollContainer.scrollTop = scrollContainer.scrollHeight
+        }
+    }
+
+    // Get model from session on mount (one-time read, no reactive dependency)
+    onMount(() => {
+        try {
+            const instanceSessions = sessions().get(props.instanceId)
+            const session = instanceSessions?.get(props.sessionId)
+            if (session?.model?.modelId) {
+                setCurrentModel(session.model.modelId)
+            }
+        } catch (e) {
+            console.warn("Could not get session model, using default", e)
+        }
+        inputRef?.focus()
+    })
+
+    async function sendMessage() {
+        const text = inputText().trim()
+        if (!text || isLoading()) return
+
+        setError(null)
+        setInputText("")
+        setIsLoading(true)
+
+        const userMessage: Message = {
+            id: generateId(),
+            role: "user",
+            content: text,
+            timestamp: Date.now(),
+            status: "complete"
+        }
+
+        const assistantMessage: Message = {
+            id: generateId(),
+            role: "assistant",
+            content: "",
+            timestamp: Date.now(),
+            status: "streaming"
+        }
+
+        // Add messages to state
+        setMessages(prev => [...prev, userMessage, assistantMessage])
+        scrollToBottom()
+
+        try {
+            // Direct fetch with streaming
+            const response = await fetch("/api/ollama/chat", {
+                method: "POST",
+                headers: { "Content-Type": "application/json" },
+                body: JSON.stringify({
+                    model: currentModel(),
+                    messages: [
+                        ...messages().filter(m => m.status === "complete").map(m => ({ role: m.role, content: m.content })),
+                        { role: "user", content: text }
+                    ],
+                    stream: true
+                })
+            })
+
+            if (!response.ok) {
+                throw new Error(`Request failed: ${response.status}`)
+            }
+
+            const reader = response.body?.getReader()
+            if (!reader) throw new Error("No response body")
+
+            const decoder = new TextDecoder()
+            let fullContent = ""
+            let buffer = ""
+
+            while (true) {
+                const { done, value } = await reader.read()
+                if (done) break
+
+                buffer += decoder.decode(value, { stream: true })
+                const lines = buffer.split("\n")
+                buffer = lines.pop() || ""
+
+                for (const line of lines) {
+                    const trimmed = line.trim()
+                    if (!trimmed.startsWith("data:")) continue
+                    const data = trimmed.slice(5).trim()
+                    if (!data || data === "[DONE]") continue
+
+                    try {
+                        const chunk = JSON.parse(data)
+                        const delta = chunk?.message?.content
+                        if (typeof delta === "string" && delta.length > 0) {
+                            fullContent += delta
+
+                            // Update assistant message content (simple state update)
+                            setMessages(prev =>
+                                prev.map(m =>
+                                    m.id === assistantMessage.id
+                                        ? { ...m, content: fullContent }
+                                        : m
+                                )
+                            )
+                            scrollToBottom()
+                        }
+                    } catch {
+                        // Ignore parse errors
+                    }
+                }
+            }
+
+            // Mark as complete
+            setMessages(prev =>
+                prev.map(m =>
+                    m.id === assistantMessage.id
+                        ? { ...m, status: "complete" }
+                        : m
+                )
+            )
+        } catch (e) {
+            const errorMsg = e instanceof Error ? e.message : "Unknown error"
+            setError(errorMsg)
+
+            // Mark as error
+            setMessages(prev =>
+                prev.map(m =>
+                    m.id === assistantMessage.id
+                        ? { ...m, status: "error", content: `Error: ${errorMsg}` }
+                        : m
+                )
+            )
+        } finally {
+            setIsLoading(false)
+            scrollToBottom()
+        }
+    }
+
+    function handleKeyDown(e: KeyboardEvent) {
+        if (e.key === "Enter" && !e.shiftKey) {
+            e.preventDefault()
+            sendMessage()
+        }
+    }
+
+    return (
+        <div style={{
+            display: "flex",
+            "flex-direction": "column",
+            height: "100%",
+            background: "#0a0a0b",
+            color: "#e4e4e7"
+        }}>
+            {/* Header */}
+            <div style={{
+                padding: "16px",
+                "border-bottom": "1px solid #27272a",
+                background: "#18181b"
+            }}>
+                <h2 style={{ margin: 0, "font-size": "16px" }}>
+                    🧪 Minimal Chat (Bypass Mode)
+                </h2>
+                <p style={{ margin: "4px 0 0", "font-size": "12px", color: "#71717a" }}>
+                    Model: {currentModel()} | Testing UI responsiveness
+                </p>
+            </div>
+
+            {/* Messages */}
+            <div
+                ref={scrollContainer}
+                style={{
+                    flex: 1,
+                    overflow: "auto",
+                    padding: "16px"
+                }}
+            >
+                <Show when={messages().length === 0}>
+                    <div style={{
+                        "text-align": "center",
+                        color: "#71717a",
+                        padding: "48px"
+                    }}>
+                        Send a message to test UI responsiveness
+                    </div>
+                </Show>
+
+                <For each={messages()}>
+                    {(message) => (
+                        <div style={{
+                            "margin-bottom": "16px",
+                            padding: "12px",
+                            background: message.role === "user" ? "#27272a" : "#18181b",
+                            "border-radius": "8px",
+                            "border-left": message.role === "assistant" ? "3px solid #6366f1" : "none"
+                        }}>
+                            <div style={{
+                                "font-size": "11px",
+                                color: "#71717a",
+                                "margin-bottom": "8px"
+                            }}>
+                                {message.role === "user" ? "You" : "Assistant"}
+                                {message.status === "streaming" && " (streaming...)"}
+                                {message.status === "error" && " (error)"}
+                            </div>
+                            <div style={{
+                                "white-space": "pre-wrap",
+                                "word-break": "break-word",
+                                "font-size": "14px",
+                                "line-height": "1.6"
+                            }}>
+                                {message.content || (message.status === "streaming" ? "▋" : "")}
+                            </div>
+                        </div>
+                    )}
+                </For>
+            </div>
+
+            {/* Error display */}
+            <Show when={error()}>
+                <div style={{
+                    padding: "8px 16px",
+                    background: "#7f1d1d",
+                    color: "#fecaca",
+                    "font-size": "12px"
+                }}>
+                    Error: {error()}
+                </div>
+            </Show>
+
+            {/* Input area */}
+            <div style={{
+                padding: "16px",
+                "border-top": "1px solid #27272a",
+                background: "#18181b"
+            }}>
+                <div style={{ display: "flex", gap: "8px" }}>
+                    <textarea
+                        ref={inputRef}
+                        value={inputText()}
+                        onInput={(e) => setInputText(e.currentTarget.value)}
+                        onKeyDown={handleKeyDown}
+                        placeholder="Type a message... (Enter to send)"
+                        disabled={isLoading()}
+                        style={{
+                            flex: 1,
+                            padding: "12px",
+                            background: "#27272a",
+                            border: "1px solid #3f3f46",
+                            "border-radius": "8px",
+                            color: "#e4e4e7",
+                            resize: "none",
+                            "font-size": "14px",
+                            "min-height": "48px",
+                            "max-height": "150px"
+                        }}
+                        rows={1}
+                    />
+                    <button
+                        onClick={sendMessage}
+                        disabled={isLoading() || !inputText().trim()}
+                        style={{
+                            padding: "12px 24px",
+                            background: isLoading() ? "#3f3f46" : "#6366f1",
+                            color: "white",
+                            border: "none",
+                            "border-radius": "8px",
+                            cursor: isLoading() ? "wait" : "pointer",
+                            "font-weight": "600"
+                        }}
+                    >
+                        {isLoading() ? "..." : "Send"}
+                    </button>
+                </div>
+            </div>
+        </div>
+    )
+}
+
+export default MinimalChat
--- a/packages/ui/src/components/chat/multi-task-chat.tsx
+++ b/packages/ui/src/components/chat/multi-task-chat.tsx
@@ -1,7 +1,7 @@
-import { createSignal, Show, onMount, For, createMemo, createEffect, onCleanup } from "solid-js";
+import { createSignal, Show, onMount, For, createMemo, createEffect, onCleanup, untrack } from "solid-js";
 import { sessions, withSession, setActiveSession } from "@/stores/session-state";
 import { instances } from "@/stores/instances";
-import { sendMessage, compactSession, updateSessionAgent, updateSessionModelForSession } from "@/stores/session-actions";
+import { sendMessage, compactSession, updateSessionAgent, updateSessionModelForSession, forceReset } from "@/stores/session-actions";
 import { addTask, setActiveTask, archiveTask } from "@/stores/task-actions";
 import { messageStoreBus } from "@/stores/message-v2/bus";
 import MessageBlockList, { getMessageAnchorId } from "@/components/message-block-list";
@@ -42,6 +42,7 @@ import {
 } from "lucide-solid";
 import ModelSelector from "@/components/model-selector";
 import AgentSelector from "@/components/agent-selector";
+import { DebugOverlay, setForceResetFn } from "@/components/debug-overlay";
 import AttachmentChip from "@/components/attachment-chip";
 import { createFileAttachment } from "@/types/attachment";
 import type { InstanceMessageStore } from "@/stores/message-v2/instance-store";
@@ -145,26 +146,22 @@ export default function MultiTaskChat(props: MultiTaskChatProps) {
  });

  const tokenStats = createMemo(() => {
-    const usage = sessionUsage();
-    return {
-      used: usage?.actualUsageTokens ?? 0,
-      total: usage?.totalCost ?? 0,
-      // input: usage?.inputTokens ?? 0,
-      // output: usage?.outputTokens ?? 0,
-      // reasoning: usage?.reasoningTokens ?? 0,
-      // cacheRead: usage?.cacheReadTokens ?? 0,
-      // cacheWrite: usage?.cacheWriteTokens ?? 0,
-      cost: usage?.totalCost ?? 0,
-    };
-  });
-
-  // Get current model from active task session
-  const currentModel = createMemo(() => {
-    const instanceSessions = sessions().get(props.instanceId);
-    const session = instanceSessions?.get(activeTaskSessionId());
-    return session?.model?.modelId || "unknown";
+    return untrack(() => {
+      const usage = sessionUsage();
+      return {
+        used: usage?.actualUsageTokens ?? 0,
+        total: usage?.totalCost ?? 0,
+        // input: usage?.inputTokens ?? 0,
+        // output: usage?.outputTokens ?? 0,
+        // reasoning: usage?.reasoningTokens ?? 0,
+        // cacheRead: usage?.cacheReadTokens ?? 0,
+        // cacheWrite: usage?.cacheWriteTokens ?? 0,
+        cost: usage?.totalCost ?? 0,
+      };
+    });
  });

+  // Get active task session ID (must be defined before memos that use it)
  const activeTaskSessionId = createMemo(() => {
    const task = selectedTask();
    return task?.taskSessionId || props.sessionId;
@@ -175,6 +172,13 @@ export default function MultiTaskChat(props: MultiTaskChatProps) {
    return instanceSessions?.get(activeTaskSessionId());
  });

+  // Get current model from active task session
+  const currentModel = createMemo(() => {
+    const instanceSessions = sessions().get(props.instanceId);
+    const session = instanceSessions?.get(activeTaskSessionId());
+    return session?.model?.modelId || "unknown";
+  });
+
  const currentTaskAgent = createMemo(() => activeTaskSession()?.agent || "");
  const currentTaskModel = createMemo(() => activeTaskSession()?.model || { providerId: "", modelId: "" });

@@ -207,8 +211,6 @@ export default function MultiTaskChat(props: MultiTaskChatProps) {
    // Show thinking while we're actively sending
    if (isSending()) return true;

-    // Only check the last message instead of iterating all messages
-    // This prevents O(n) reactive subscriptions during streaming
    const ids = filteredMessageIds();
    if (ids.length === 0) return false;

@@ -217,22 +219,34 @@ export default function MultiTaskChat(props: MultiTaskChatProps) {
    return lastMsg?.role === "assistant" && (lastMsg.status === "streaming" || lastMsg.status === "sending");
  });

-  // Auto-scroll during streaming - must be after isAgentThinking is defined
-  createEffect(() => {
-    const streaming = isAgentThinking();
-    if (!streaming) return;
-
-    // During streaming, scroll periodically to keep up with content (unless user is scrolling)
-    const interval = setInterval(() => {
-      if (!userScrolling()) {
-        scrollToBottom();
-      }
-    }, 300);
-    return () => clearInterval(interval);
-  });
+  // Auto-scroll during streaming - DISABLED for performance testing
+  // createEffect(() => {
+  //   const streaming = isAgentThinking();
+  //   if (!streaming) return;
+  //
+  //   let lastScrollTime = 0;
+  //   const scrollThrottled = () => {
+  //     const now = Date.now();
+  //     if (now - lastScrollTime > 500) {
+  //       lastScrollTime = now;
+  //       if (!userScrolling()) {
+  //         scrollToBottom();
+  //       }
+  //     }
+  //   };
+  //
+  //   const interval = setInterval(() => {
+  //     if (!userScrolling()) {
+  //       requestAnimationFrame(scrollToBottom);
+  //     }
+  //   }, 200);
+  //   return () => clearInterval(interval);
+  // });

  // Auto-scroll when new messages arrive (throttled to count changes only)
  let lastScrolledCount = 0;
+  let scrollTimeoutId: ReturnType<typeof setTimeout> | undefined;
+
  createEffect(() => {
    const ids = filteredMessageIds();
    const count = ids.length;
@@ -242,14 +256,27 @@ export default function MultiTaskChat(props: MultiTaskChatProps) {
    // Note: Streaming scrolling is handled by the interval in the isAgentThinking effect above
    if (count !== lastScrolledCount && count > 0 && !userScrolling()) {
      lastScrolledCount = count;
+
+      // Clear any existing timeout to prevent timer accumulation
+      if (scrollTimeoutId) {
+        clearTimeout(scrollTimeoutId);
+      }
+
+      // Use requestAnimationFrame for smoother scrolling without locking specific frames
      requestAnimationFrame(() => {
-        setTimeout(scrollToBottom, 50);
+        scrollToBottom();
      });
    }
  });

  // Scroll event listener to detect user scrolling
  onMount(() => {
+    // Wire up debug overlay reset function (must be inside onMount to avoid SolidJS errors)
+    setForceResetFn(() => {
+      forceReset();
+      setIsSending(false);
+    });
+
    const handleScroll = () => {
      if (scrollContainer) {
        const isScrollingUp = scrollContainer.scrollTop < lastScrollTop();
@@ -270,10 +297,24 @@ export default function MultiTaskChat(props: MultiTaskChatProps) {
    container?.addEventListener('scroll', handleScroll, { passive: true });

    return () => {
-      container?.removeEventListener('scroll', handleScroll);
+      // Enhanced cleanup: remove scroll listener and clear any pending timeouts
+      if (container) {
+        container.removeEventListener('scroll', handleScroll);
+      }
+      // Clear any pending scroll timeout
+      if (scrollTimeoutId) {
+        clearTimeout(scrollTimeoutId);
+      }
    };
  });

+  // Additional cleanup on component unmount
+  onCleanup(() => {
+    if (scrollTimeoutId) {
+      clearTimeout(scrollTimeoutId);
+    }
+  });
+
  const handleSendMessage = async () => {
    const message = chatInput().trim();
    if (!message || isSending()) return;
@@ -371,7 +412,15 @@ export default function MultiTaskChat(props: MultiTaskChatProps) {
  };

  // Stop/cancel the current agent operation
-  const handleStopAgent = async () => {
+  const handleStopAgent = async (e?: MouseEvent) => {
+    // Check for Force Reset (Shift + Click)
+    if (e?.shiftKey) {
+      log.warn("Shift+Click detected on Stop Agent - Triggering Force Reset");
+      forceReset();
+      setIsSending(false);
+      return;
+    }
+
    const task = selectedTask();
    if (!task) return;

@@ -447,6 +496,7 @@ export default function MultiTaskChat(props: MultiTaskChatProps) {

  return (
    <main class="absolute inset-0 flex flex-col bg-[#0a0a0b] text-zinc-300 font-sans selection:bg-indigo-500/30 overflow-hidden">
+      <DebugOverlay />
      {/* Header */}
      <header class="h-14 px-4 flex items-center justify-between bg-zinc-900/60 backdrop-blur-xl border-b border-white/5 relative z-30 shrink-0">
        <div class="flex items-center space-x-3">
@@ -875,7 +925,7 @@ export default function MultiTaskChat(props: MultiTaskChatProps) {
                    <button
                      onClick={handleStopAgent}
                      class="px-3 py-1.5 bg-rose-500/20 hover:bg-rose-500/30 text-rose-300 rounded-lg text-[10px] font-bold uppercase tracking-wide transition-all border border-rose-500/30"
-                      title="Stop response"
+                      title="Stop response (Shift+Click to Force Reset UI)"
                    >
                      <StopCircle size={12} class="inline-block mr-1" />
                      Stop
--- a/packages/ui/src/components/chat/multix-chat-v2.tsx
+++ b/packages/ui/src/components/chat/multix-chat-v2.tsx
--- a/packages/ui/src/components/chat/multix-v2/core/SimpleMessageBlock.tsx
+++ b/packages/ui/src/components/chat/multix-v2/core/SimpleMessageBlock.tsx
@@ -0,0 +1,101 @@
+/**
+ * SimpleMessageBlock - Polling-based message renderer
+ * 
+ * Updates content via interval, not reactive cascade.
+ * This prevents the freeze during streaming.
+ */
+
+import { createSignal, Show, onMount, onCleanup } from "solid-js";
+import type { InstanceMessageStore } from "@/stores/message-v2/instance-store";
+
+interface SimpleMessageBlockProps {
+    messageId: string;
+    store: () => InstanceMessageStore;
+}
+
+export function SimpleMessageBlock(props: SimpleMessageBlockProps) {
+    const [content, setContent] = createSignal("");
+    const [isStreaming, setIsStreaming] = createSignal(false);
+    const [isUser, setIsUser] = createSignal(false);
+    const [timestamp, setTimestamp] = createSignal("");
+    const [tokenCount, setTokenCount] = createSignal(0);
+
+    function updateFromStore() {
+        const message = props.store().getMessage(props.messageId);
+        if (!message) return;
+
+        setIsUser(message.role === "user");
+        setIsStreaming(message.status === "streaming" || message.status === "sending");
+
+        // Extract text content from parts
+        const parts = message.parts || {};
+        let text = "";
+        for (const partId of Object.keys(parts)) {
+            const partRecord = parts[partId];
+            if (partRecord?.data?.type === "text") {
+                text = (partRecord.data as any).text || "";
+                break;
+            }
+        }
+
+        // Fallback to direct content
+        if (!text && (message as any).content) {
+            text = (message as any).content;
+        }
+
+        setContent(text);
+        setTokenCount(Math.ceil(text.length / 4));
+
+        // Note: MessageRecord doesn't have time property, skip timestamp
+    }
+
+    onMount(() => {
+        updateFromStore();
+
+        // Poll for updates during streaming (every 100ms)
+        const interval = setInterval(() => {
+            const msg = props.store().getMessage(props.messageId);
+            if (msg?.status === "streaming" || msg?.status === "sending" || isStreaming()) {
+                updateFromStore();
+            }
+        }, 100);
+
+        onCleanup(() => clearInterval(interval));
+    });
+
+    return (
+        <div
+            id={`message-anchor-${props.messageId}`}
+            class={`rounded-xl p-4 transition-all min-w-0 overflow-hidden ${isUser()
+                ? "bg-zinc-800/50 border border-zinc-700/50"
+                : "bg-zinc-900/50 border border-indigo-500/20"
+                }`}
+        >
+            <div class="flex items-center justify-between mb-2">
+                <div class="flex items-center gap-2">
+                    <div class={`text-[10px] font-bold uppercase tracking-wide ${isUser() ? "text-indigo-400" : "text-emerald-400"}`}>
+                        {isUser() ? "You" : "Assistant"}
+                    </div>
+                    <Show when={isStreaming()}>
+                        <div class="flex items-center gap-2">
+                            <div class="flex items-center gap-1 text-[9px] text-violet-400">
+                                <div class="w-1.5 h-1.5 bg-violet-400 rounded-full animate-pulse" />
+                                <span>Thinking...</span>
+                            </div>
+                            <span class="text-[9px] font-mono text-zinc-500 bg-zinc-800/50 px-1 rounded">
+                                {tokenCount()} tks
+                            </span>
+                        </div>
+                    </Show>
+                </div>
+                <div class="text-[9px] text-zinc-600">{timestamp()}</div>
+            </div>
+            <div
+                class="text-sm text-zinc-100 leading-relaxed whitespace-pre-wrap break-words overflow-hidden"
+                style={{ "word-break": "break-word", "overflow-wrap": "anywhere" }}
+            >
+                {content() || (isStreaming() ? "▋" : "")}
+            </div>
+        </div>
+    );
+}
--- a/packages/ui/src/components/chat/multix-v2/exports.ts
+++ b/packages/ui/src/components/chat/multix-v2/exports.ts
@@ -0,0 +1,8 @@
+// Re-export all MultiX v2 components
+export { default as MultiXV2 } from "./index";
+export { SimpleMessageBlock } from "./core/SimpleMessageBlock";
+export { PipelineView } from "./features/PipelineView";
+export { MessageNavSidebar } from "./features/MessageNavSidebar";
+export { LiteAgentSelector } from "./features/LiteAgentSelector";
+export { LiteModelSelector } from "./features/LiteModelSelector";
+export { enhancePrompt, getQuickTips } from "./features/PromptEnhancer";
--- a/packages/ui/src/components/chat/multix-v2/features/LiteAgentSelector.tsx
+++ b/packages/ui/src/components/chat/multix-v2/features/LiteAgentSelector.tsx
@@ -0,0 +1,637 @@
+/**
+ * LiteAgentSelector - Non-reactive agent selector for MultiX v2
+ * 
+ * Uses polling instead of reactive subscriptions to prevent cascading updates.
+ * Includes AI Agent Generator feature.
+ */
+
+import { createSignal, For, onMount, onCleanup, Show } from "solid-js";
+import { agents, setAgents, providers } from "@/stores/session-state";
+import { fetchAgents } from "@/stores/session-api";
+import { updateInstanceConfig } from "@/stores/instance-config";
+import { toast } from "solid-toast";
+import { ChevronDown, Bot, Plus, Sparkles, Loader2, Save, X, RefreshCw } from "lucide-solid";
+import { serverApi } from "@/lib/api-client";
+
+interface LiteAgentSelectorProps {
+    instanceId: string;
+    sessionId: string;
+    currentAgent: string;
+    onAgentChange: (agent: string) => void;
+}
+
+interface AgentInfo {
+    name: string;
+    description?: string;
+    systemPrompt?: string;
+}
+
+export function LiteAgentSelector(props: LiteAgentSelectorProps) {
+    const [isOpen, setIsOpen] = createSignal(false);
+    const [agentList, setAgentList] = createSignal<AgentInfo[]>([]);
+    const [isGenerating, setIsGenerating] = createSignal(false);
+    const [showGenerator, setShowGenerator] = createSignal(false);
+    const [generatorInput, setGeneratorInput] = createSignal("");
+    const [generatedAgent, setGeneratedAgent] = createSignal<AgentInfo | null>(null);
+    const [isSaving, setIsSaving] = createSignal(false);
+    const [selectedModel, setSelectedModel] = createSignal("glm-4");
+    const [availableModels, setAvailableModels] = createSignal<{ id: string, name: string, provider: string }[]>([]);
+
+    // Load agents once on mount, then poll
+    function loadAgents() {
+        try {
+            const instanceAgents = agents().get(props.instanceId) || [];
+            const nonSubagents = instanceAgents.filter((a: any) => a.mode !== "subagent");
+            setAgentList(nonSubagents.map((a: any) => ({
+                name: a.name,
+                description: a.description,
+                systemPrompt: a.systemPrompt
+            })));
+        } catch (e) {
+            console.warn("Failed to load agents", e);
+        }
+    }
+
+    onMount(() => {
+        loadAgents();
+
+        // Populate available models
+        const allProviders = providers().get(props.instanceId) || [];
+        const models: { id: string, name: string, provider: string }[] = [];
+        allProviders.forEach(p => {
+            p.models.forEach(m => {
+                models.push({ id: m.id, name: m.name || m.id, provider: p.id });
+            });
+        });
+
+        // Add defaults if none found
+        if (models.length === 0) {
+            models.push({ id: "glm-4", name: "GLM-4 (Z.AI)", provider: "zai" });
+            models.push({ id: "qwen-coder-plus-latest", name: "Qwen Coder Plus (Zen)", provider: "opencode-zen" });
+            models.push({ id: "minimax-m1", name: "MiniMax M1 (Ollama)", provider: "ollama" });
+        }
+        setAvailableModels(models);
+
+        // Poll every 5 seconds (agents don't change often)
+        const interval = setInterval(loadAgents, 5000);
+        onCleanup(() => clearInterval(interval));
+    });
+
+    const handleSelect = (agentName: string) => {
+        props.onAgentChange(agentName);
+        setIsOpen(false);
+    };
+
+    const handleGenerateAgent = async () => {
+        const input = generatorInput().trim();
+        if (!input || isGenerating()) return;
+
+        setIsGenerating(true);
+        const modelInfo = availableModels().find(m => m.id === selectedModel());
+
+        // Normalize provider ID - handle variants like "ollama-cloud" -> "ollama"
+        let provider = modelInfo?.provider || "zai";
+        if (provider.includes("ollama")) provider = "ollama";
+        if (provider.includes("zen")) provider = "opencode-zen";
+
+        console.log(`[AgentGenerator] Using provider: ${provider}, model: ${selectedModel()}`);
+
+        // AI generation prompt - focused on unique, creative output
+        const generationPrompt = `Create a unique AI coding assistant agent based on: "${input}"
+
+RULES:
+1. NAME: Create a catchy, memorable 1-3 word name (e.g., "Neon Architect", "Logic Ghost", "Cortex", "Syntax Specter"). BE CREATIVE!
+2. DESCRIPTION: One powerful sentence about their unique paradigm or specialty.
+3. SYSTEM PROMPT: Write a 400+ word deep-dive into their psyche, expertise, and operational style. 
+   - DO NOT be generic. 
+   - Give them a clear VOICE and philosophy.
+   - Professional, yet distinct.
+   - Mention specific methodologies they favor.
+   - Explain how they view the relationship between code and problem-solving.
+
+IMPORTANT: Return ONLY valid JSON in this format:
+{"name": "...", "description": "...", "systemPrompt": "..."}`;
+
+        const endpoints: Record<string, string> = {
+            "zai": "/api/zai/chat",
+            "opencode-zen": "/api/opencode-zen/chat",
+            "ollama": "/api/ollama/chat"
+        };
+
+        // Timeout wrapper for fetch with 60 second limit
+        const fetchWithTimeout = async (url: string, options: RequestInit, timeoutMs: number = 60000) => {
+            const controller = new AbortController();
+            const timeoutId = setTimeout(() => controller.abort(), timeoutMs);
+            try {
+                const response = await fetch(url, { ...options, signal: controller.signal });
+                clearTimeout(timeoutId);
+                return response;
+            } catch (e) {
+                clearTimeout(timeoutId);
+                throw e;
+            }
+        };
+
+        const tryEndpoint = async (prov: string, model: string) => {
+            try {
+                console.log(`[AgentGenerator] Attempting generation with ${prov}/${model}...`);
+
+                // Use absolute URL from serverApi to avoid port issues
+                const baseUrl = serverApi.getApiBase();
+                const endpoint = `${baseUrl}${endpoints[prov]}`;
+
+                if (!endpoints[prov]) {
+                    console.warn(`[AgentGenerator] No endpoint configured for provider: ${prov}`);
+                    return null;
+                }
+
+                const response = await fetchWithTimeout(endpoint, {
+                    method: "POST",
+                    headers: { "Content-Type": "application/json" },
+                    body: JSON.stringify({
+                        model: model,
+                        messages: [{ role: "user", content: generationPrompt }],
+                        stream: false
+                    })
+                }, 60000); // 60 second timeout
+
+                if (response.ok) {
+                    const data = await response.json();
+                    const content = prov === "zai" || prov === "opencode-zen"
+                        ? (data?.choices?.[0]?.message?.content || data?.message?.content || "")
+                        : (data?.message?.content || "");
+
+                    console.log(`[AgentGenerator] Received content from ${prov}:`, content.substring(0, 100) + "...");
+
+                    const result = tryParseAgentJson(content, input);
+                    if (result) return result;
+
+                    console.warn(`[AgentGenerator] Failed to parse JSON from ${prov} response`);
+                } else {
+                    const errText = await response.text();
+                    console.error(`[AgentGenerator] Endpoint ${prov} returned ${response.status}:`, errText);
+                }
+            } catch (e: any) {
+                if (e.name === 'AbortError') {
+                    console.warn(`[AgentGenerator] Request to ${prov} timed out after 60s`);
+                    toast.error(`Generation timed out. Try a faster model.`, { duration: 5000 });
+                } else {
+                    console.warn(`[AgentGenerator] Endpoint ${prov} failed:`, e);
+                }
+            }
+            return null;
+        };
+
+        // 1. Try selected model
+        let parsed = await tryEndpoint(provider, selectedModel());
+
+        // 2. Fallbacks if selected fails - try faster models
+        if (!parsed) {
+            console.log("[AgentGenerator] Selected model failed, trying fallbacks...");
+            const fallbacks = [
+                { prov: "ollama", model: "qwen3:8b" },
+                { prov: "opencode-zen", model: "qwen-coder-plus-latest" },
+                { prov: "zai", model: "glm-4" },
+            ].filter(f => f.model !== selectedModel());
+
+            for (const f of fallbacks) {
+                parsed = await tryEndpoint(f.prov, f.model);
+                if (parsed) break;
+            }
+        }
+
+        if (parsed) {
+            setGeneratedAgent(parsed);
+            toast.success("Agent generated!", { icon: "🎉", duration: 3000 });
+        } else {
+            console.warn("[AgentGenerator] All AI endpoints failed, using smart fallback");
+            setGeneratedAgent(generateSmartFallback(input));
+            toast.success("Agent created (local fallback)", { duration: 3000 });
+        }
+
+        setIsGenerating(false);
+    };
+
+    // Try to parse JSON from AI response
+    const tryParseAgentJson = (content: string, input: string): { name: string; description: string; systemPrompt: string } | null => {
+        try {
+            const jsonMatch = content.match(/\{[\s\S]*\}/);
+            if (jsonMatch) {
+                const parsed = JSON.parse(jsonMatch[0]);
+                if (parsed.name && parsed.systemPrompt && parsed.systemPrompt.length > 100) {
+                    return {
+                        name: parsed.name,
+                        description: parsed.description || input,
+                        systemPrompt: parsed.systemPrompt
+                    };
+                }
+            }
+        } catch (e) {
+            console.error("JSON parse error:", e);
+        }
+        return null;
+    };
+
+    // Generate a smart fallback that actually feels unique
+    const generateSmartFallback = (input: string): { name: string; description: string; systemPrompt: string } => {
+        const name = generateFallbackName(input);
+        const timestamp = Date.now();
+
+        // Create unique content based on input analysis
+        const inputLower = input.toLowerCase();
+        const isFrontend = /react|vue|angular|css|html|ui|frontend|web/.test(inputLower);
+        const isBackend = /api|server|node|python|database|backend/.test(inputLower);
+        const isFullStack = /full.?stack|complete|everything/.test(inputLower);
+        const isAI = /ai|ml|machine|learning|neural|gpt|claude|llm/.test(inputLower);
+        const isDevOps = /devops|docker|kubernetes|ci|cd|deploy/.test(inputLower);
+
+        let specialty = "general software development";
+        let techStack = "JavaScript, TypeScript, Python";
+        let uniqueTrait = "methodical approach to problem-solving";
+
+        if (isFrontend) {
+            specialty = "frontend architecture and user experience";
+            techStack = "React, Vue, TypeScript, CSS, Tailwind";
+            uniqueTrait = "pixel-perfect attention to detail and smooth animations";
+        } else if (isBackend) {
+            specialty = "backend systems and API design";
+            techStack = "Node.js, Python, PostgreSQL, Redis, GraphQL";
+            uniqueTrait = "building scalable, fault-tolerant services";
+        } else if (isFullStack) {
+            specialty = "end-to-end application development";
+            techStack = "React, Node.js, PostgreSQL, Docker, AWS";
+            uniqueTrait = "seamless integration between frontend and backend";
+        } else if (isAI) {
+            specialty = "AI/ML integration and prompt engineering";
+            techStack = "Python, LangChain, OpenAI, HuggingFace, Vector DBs";
+            uniqueTrait = "crafting intelligent, context-aware AI solutions";
+        } else if (isDevOps) {
+            specialty = "infrastructure and deployment automation";
+            techStack = "Docker, Kubernetes, Terraform, GitHub Actions, AWS";
+            uniqueTrait = "zero-downtime deployments and infrastructure as code";
+        }
+
+        return {
+            name,
+            description: `Expert in ${specialty} with ${uniqueTrait}`,
+            systemPrompt: `You are ${name}, a senior software engineer with 10+ years of expertise in ${specialty}.
+
+## Your Personality
+You are confident but humble, always explaining your reasoning clearly. You prefer elegant, maintainable solutions over clever hacks. When you don't know something, you say so honestly and suggest ways to find the answer.
+
+## Technical Expertise
+Your primary stack: ${techStack}
+Your specialty: ${specialty}
+Your unique strength: ${uniqueTrait}
+
+## How You Work
+1. **Understand First**: Before writing code, you analyze the existing codebase structure, patterns, and conventions
+2. **Plan Carefully**: You outline your approach before implementing, considering edge cases and potential issues
+3. **Code Quality**: Every line you write follows best practices - clean naming, proper error handling, comprehensive types
+4. **Test Thinking**: You consider how code will be tested, even if tests aren't explicitly requested
+5. **Documentation**: You add meaningful comments for complex logic, not obvious operations
+
+## Code Standards You Follow
+- Use descriptive variable and function names that reveal intent
+- Keep functions small and focused (single responsibility)
+- Handle errors gracefully with informative messages
+- Prefer composition over inheritance
+- Write self-documenting code, supplement with comments only where needed
+- Always consider performance implications
+
+## Communication Style
+- Be direct and actionable in your responses
+- When suggesting changes, explain WHY not just WHAT
+- If multiple approaches exist, briefly mention pros/cons
+- Celebrate good code when you see it
+- Provide constructive feedback on improvements
+
+## Tool Usage
+- Use read_file to understand existing code before modifying
+- Use list_files to understand project structure
+- Use write_file to create or update files with complete, working code
+- Always verify syntax correctness before submitting
+
+Built for: ${input}
+Session ID: ${timestamp}`
+        };
+    };
+
+    // Generate a professional fallback name from user input
+    const generateFallbackName = (input: string): string => {
+        // Extract key words and create a professional sounding name
+        const words = input.toLowerCase().split(/\s+/).filter(w => w.length > 2);
+
+        // Common tech keywords to look for
+        const keywords: Record<string, string> = {
+            'typescript': 'TypeScript Pro',
+            'javascript': 'JS Expert',
+            'react': 'React Master',
+            'python': 'Python Guru',
+            'api': 'API Architect',
+            'code': 'Code Expert',
+            'full': 'Full Stack Pro',
+            'frontend': 'Frontend Master',
+            'backend': 'Backend Pro',
+            'mcp': 'MCP Specialist',
+            'agent': 'Smart Agent',
+            'thinking': 'Deep Thinker',
+            'claude': 'AI Assistant',
+            'smart': 'Smart Coder',
+            'fix': 'Bug Hunter',
+            'test': 'Test Master',
+            'debug': 'Debug Pro',
+            'architect': 'Code Architect',
+            'review': 'Code Reviewer'
+        };
+
+        // Try to find a matching keyword
+        for (const word of words) {
+            for (const [key, name] of Object.entries(keywords)) {
+                if (word.includes(key)) {
+                    return name;
+                }
+            }
+        }
+
+        // Default: Create from first few words
+        const titleWords = words.slice(0, 2).map(w =>
+            w.charAt(0).toUpperCase() + w.slice(1)
+        );
+        return titleWords.length > 0 ? titleWords.join(' ') + ' Pro' : 'Custom Agent';
+    }
+
+    // Generate a sophisticated fallback prompt when API fails
+    const generateFallbackPrompt = (description: string): string => {
+        return `# ${description}
+
+## IDENTITY & CORE MISSION
+You are a world-class AI coding assistant specialized in: ${description}. You combine deep technical expertise with exceptional problem-solving abilities to deliver production-ready code that exceeds professional standards.
+
+## CODEBASE AWARENESS PROTOCOL
+Before writing any code, you MUST:
+1. **Analyze Context**: Understand the existing project structure, patterns, and conventions
+2. **Identify Dependencies**: Check package.json, imports, and installed libraries
+3. **Match Style**: Adapt your output to the existing code style in the project
+4. **Verify Compatibility**: Ensure new code integrates seamlessly with existing modules
+
+## TECHNICAL EXPERTISE
+- **Languages**: JavaScript, TypeScript, Python, and relevant frameworks
+- **Patterns**: SOLID principles, DRY, KISS, Clean Architecture
+- **Testing**: TDD approach, comprehensive test coverage
+- **Documentation**: Clear comments, JSDoc/TSDoc, README updates
+
+## CODING STANDARDS
+1. **Naming**: Use descriptive, intention-revealing names
+2. **Functions**: Single responsibility, max 20-30 lines per function
+3. **Error Handling**: Always handle errors gracefully with informative messages
+4. **Types**: Prefer strict typing, avoid \`any\` type
+5. **Comments**: Explain WHY, not WHAT (the code explains what)
+
+## ARCHITECTURAL PRINCIPLES
+- Favor composition over inheritance
+- Implement proper separation of concerns
+- Design for extensibility and maintainability
+- Consider performance implications of design choices
+- Apply appropriate design patterns (Factory, Strategy, Observer, etc.)
+
+## COMMUNICATION STYLE
+- Be concise but thorough in explanations
+- Provide rationale for technical decisions
+- Offer alternatives when relevant
+- Acknowledge limitations and edge cases
+- Use code examples to illustrate concepts
+
+## TOOL USAGE
+When modifying the codebase:
+1. Use \`read_file\` to understand existing code before making changes
+2. Use \`list_files\` to understand project structure
+3. Use \`write_file\` to create or update files with complete, working code
+4. Always verify your changes are syntactically correct
+5. Consider impact on other files that may need updates
+
+## OUTPUT QUALITY STANDARDS
+Every piece of code you generate must be:
+- ✅ Syntactically correct and immediately runnable
+- ✅ Following existing project conventions
+- ✅ Properly typed (if TypeScript)
+- ✅ Including necessary imports
+- ✅ Handling edge cases and errors
+- ✅ Well-documented where appropriate
+
+You are committed to excellence and take pride in delivering code that professionals would admire.`
+    }
+
+    const handleSaveAgent = async () => {
+        const agent = generatedAgent();
+        if (!agent || isSaving()) return;
+
+        setIsSaving(true);
+        const toastId = toast.loading("Saving agent...");
+
+        try {
+            // Save to backend
+            const response = await fetch(`/api/workspaces/${props.instanceId}/agents`, {
+                method: "POST",
+                headers: { "Content-Type": "application/json" },
+                body: JSON.stringify({
+                    name: agent.name,
+                    description: agent.description,
+                    systemPrompt: agent.systemPrompt,
+                    mode: "agent"
+                })
+            });
+
+            if (response.ok) {
+                // CRITICAL: Update local instance config to keep it in sync with backend
+                // This is the source of truth that fetchAgents() reads from
+                await updateInstanceConfig(props.instanceId, (draft) => {
+                    if (!draft.customAgents) {
+                        draft.customAgents = [];
+                    }
+                    const existingIndex = draft.customAgents.findIndex(a => a.name === agent.name);
+                    const agentData = {
+                        name: agent.name,
+                        description: agent.description || "",
+                        prompt: agent.systemPrompt || ""
+                    };
+                    if (existingIndex >= 0) {
+                        draft.customAgents[existingIndex] = agentData;
+                    } else {
+                        draft.customAgents.push(agentData);
+                    }
+                });
+
+                // Fetch fresh agents from backend to update global signals
+                await fetchAgents(props.instanceId);
+
+                // Refresh local agent list
+                loadAgents();
+
+                // Manual update to ensure immediate feedback (fix for list lag)
+                setAgentList(prev => {
+                    if (prev.some(a => a.name === agent.name)) return prev;
+                    return [...prev, { name: agent.name, description: agent.description, systemPrompt: agent.systemPrompt }];
+                });
+
+                // Select the new agent
+                props.onAgentChange(agent.name);
+
+                toast.success(`Agent "${agent.name}" saved and activated!`, { id: toastId });
+
+                // Close generator
+                setShowGenerator(false);
+                setGeneratedAgent(null);
+                setGeneratorInput("");
+                setIsOpen(false);
+            } else {
+                const errorData = await response.json().catch(() => ({}));
+                console.error("Failed to save agent:", response.status, errorData);
+                toast.error(`Failed to save agent: ${errorData.error || response.statusText}`, { id: toastId });
+            }
+        } catch (error) {
+            console.error("Failed to save agent:", error);
+            toast.error("Network error while saving agent", { id: toastId });
+        } finally {
+            setIsSaving(false);
+        }
+    };
+
+    return (
+        <div class="relative">
+            <button
+                onClick={() => setIsOpen(!isOpen())}
+                class="flex items-center justify-between w-full px-3 py-2 bg-zinc-900/60 border border-white/10 rounded-lg text-left hover:border-indigo-500/30 transition-all"
+            >
+                <div class="flex items-center gap-2">
+                    <Bot size={14} class="text-indigo-400" />
+                    <span class="text-[11px] font-bold text-zinc-200 truncate">
+                        {props.currentAgent || "Select Agent"}
+                    </span>
+                </div>
+                <ChevronDown size={12} class={`text-zinc-500 transition-transform ${isOpen() ? "rotate-180" : ""}`} />
+            </button>
+
+            <Show when={isOpen()}>
+                <div class="absolute top-full left-0 right-0 mt-1 bg-zinc-900 border border-white/10 rounded-lg shadow-xl z-50 max-h-[80vh] overflow-y-auto">
+                    {/* Agent Generator Toggle */}
+                    <button
+                        onClick={() => setShowGenerator(!showGenerator())}
+                        class="w-full px-3 py-2 text-left hover:bg-indigo-500/10 transition-colors flex items-center gap-2 border-b border-white/5 text-indigo-400"
+                    >
+                        <Sparkles size={12} />
+                        <span class="text-[11px] font-bold">AI Agent Generator</span>
+                        <Plus size={12} class="ml-auto" />
+                    </button>
+
+                    {/* Generator Panel */}
+                    <Show when={showGenerator()}>
+                        <div class="p-3 border-b border-white/10 bg-zinc-950/50 space-y-3">
+                            <div class="space-y-1">
+                                <div class="text-[10px] text-zinc-400 font-medium">Generation Model:</div>
+                                <select
+                                    value={selectedModel()}
+                                    onChange={(e) => setSelectedModel(e.currentTarget.value)}
+                                    class="w-full bg-zinc-800 border border-white/10 rounded px-2 py-1.5 text-[10px] text-zinc-200 outline-none focus:border-indigo-500/50"
+                                >
+                                    <For each={availableModels()}>
+                                        {(model) => (
+                                            <option value={model.id}>{model.name}</option>
+                                        )}
+                                    </For>
+                                </select>
+                            </div>
+
+                            <div class="text-[10px] text-zinc-400 font-medium">
+                                Describe the agent you want to create:
+                            </div>
+                            <textarea
+                                value={generatorInput()}
+                                onInput={(e) => setGeneratorInput(e.currentTarget.value)}
+                                placeholder="e.g., A TypeScript expert who focuses on clean code and best practices..."
+                                class="w-full bg-zinc-800 border border-white/10 rounded-lg px-3 py-2 text-[11px] text-zinc-200 placeholder-zinc-600 resize-none outline-none focus:border-indigo-500/50"
+                                rows={3}
+                            />
+                            <div class="flex items-center gap-2">
+                                <button
+                                    onClick={handleGenerateAgent}
+                                    disabled={!generatorInput().trim() || isGenerating()}
+                                    class="flex-1 px-3 py-1.5 bg-indigo-500/20 border border-indigo-500/40 rounded-lg text-[10px] font-bold text-indigo-300 hover:bg-indigo-500/30 disabled:opacity-50 disabled:cursor-not-allowed flex items-center justify-center gap-2"
+                                >
+                                    <Show when={isGenerating()} fallback={<Sparkles size={12} />}>
+                                        <Loader2 size={12} class="animate-spin" />
+                                    </Show>
+                                    {isGenerating() ? "Generating..." : "Generate Agent"}
+                                </button>
+                            </div>
+
+                            {/* Generated Agent Preview */}
+                            <Show when={generatedAgent()}>
+                                <div class="bg-zinc-800/50 rounded-lg p-3 border border-emerald-500/30 space-y-2">
+                                    <div class="flex items-center justify-between">
+                                        <span class="text-[10px] font-bold text-emerald-400">Generated Agent</span>
+                                        <button
+                                            onClick={() => setGeneratedAgent(null)}
+                                            class="text-zinc-500 hover:text-zinc-300"
+                                        >
+                                            <X size={12} />
+                                        </button>
+                                    </div>
+                                    <div class="text-[12px] font-bold text-zinc-100">{generatedAgent()?.name}</div>
+                                    <div class="text-[10px] text-zinc-400">{generatedAgent()?.description}</div>
+                                    <div class="text-[9px] text-zinc-400 max-h-60 overflow-y-auto whitespace-pre-wrap font-mono bg-black/20 p-2 rounded border border-white/5">
+                                        {generatedAgent()?.systemPrompt}
+                                    </div>
+                                    <button
+                                        onClick={handleSaveAgent}
+                                        disabled={isSaving()}
+                                        class="w-full flex items-center justify-center gap-2 py-2 bg-emerald-600 hover:bg-emerald-500 disabled:opacity-50 disabled:cursor-not-allowed text-white rounded-md text-[11px] font-bold transition-all shadow-lg active:scale-95"
+                                    >
+                                        <Show when={isSaving()} fallback={<Save size={14} />}>
+                                            <Loader2 size={14} class="animate-spin" />
+                                        </Show>
+                                        {isSaving() ? "Saving..." : "Save & Use Agent"}
+                                    </button>
+                                </div>
+                            </Show>
+                        </div>
+                    </Show>
+
+                    {/* Agent List */}
+                    <div class="px-3 py-1.5 flex items-center justify-between border-t border-white/5 bg-zinc-950/30">
+                        <span class="text-[9px] font-bold text-zinc-500 uppercase tracking-widest">Saved Agents</span>
+                        <button
+                            onClick={(e) => { e.stopPropagation(); loadAgents(); fetchAgents(); }}
+                            class="p-1 hover:bg-white/5 rounded text-zinc-500 hover:text-zinc-300 transition-colors"
+                            title="Refresh agents"
+                        >
+                            <RefreshCw size={10} />
+                        </button>
+                    </div>
+                    <div class="max-h-48 overflow-y-auto custom-scrollbar">
+                        <For each={agentList()}>
+                            {(agent) => (
+                                <button
+                                    onClick={() => handleSelect(agent.name)}
+                                    class={`w-full px-3 py-2 text-left hover:bg-white/5 transition-colors flex items-center gap-2 ${props.currentAgent === agent.name ? "bg-indigo-500/10 text-indigo-300" : "text-zinc-300"
+                                        }`}
+                                >
+                                    <Bot size={12} class="text-zinc-500" />
+                                    <div class="min-w-0">
+                                        <div class="text-[11px] font-bold truncate">{agent.name}</div>
+                                        {agent.description && (
+                                            <div class="text-[9px] text-zinc-500 truncate">{agent.description}</div>
+                                        )}
+                                    </div>
+                                </button>
+                            )}
+                        </For>
+                        <Show when={agentList().length === 0}>
+                            <div class="px-3 py-2 text-[10px] text-zinc-600">No agents available</div>
+                        </Show>
+                    </div>
+                </div>
+            </Show>
+        </div>
+    );
+}
--- a/packages/ui/src/components/chat/multix-v2/features/LiteModelSelector.tsx
+++ b/packages/ui/src/components/chat/multix-v2/features/LiteModelSelector.tsx
@@ -0,0 +1,121 @@
+/**
+ * LiteModelSelector - Non-reactive model selector for MultiX v2
+ * 
+ * Uses polling instead of reactive subscriptions to prevent cascading updates.
+ */
+
+import { createSignal, For, onMount, onCleanup, Show } from "solid-js";
+import { providers } from "@/stores/session-state";
+import { ChevronDown, Cpu } from "lucide-solid";
+
+interface Model {
+    id: string;
+    name: string;
+    providerId: string;
+}
+
+interface Provider {
+    id: string;
+    name: string;
+    models: Model[];
+}
+
+interface LiteModelSelectorProps {
+    instanceId: string;
+    sessionId: string;
+    currentModel: { providerId: string; modelId: string };
+    onModelChange: (model: { providerId: string; modelId: string }) => void;
+}
+
+export function LiteModelSelector(props: LiteModelSelectorProps) {
+    const [isOpen, setIsOpen] = createSignal(false);
+    const [providerList, setProviderList] = createSignal<Provider[]>([]);
+
+    // Load providers once on mount, then poll
+    function loadProviders() {
+        try {
+            const instanceProviders = providers().get(props.instanceId) || [];
+            setProviderList(instanceProviders.map((p: any) => ({
+                id: p.id,
+                name: p.name,
+                models: (p.models || []).map((m: any) => ({
+                    id: m.id,
+                    name: m.name,
+                    providerId: p.id,
+                })),
+            })));
+        } catch (e) {
+            console.warn("Failed to load providers", e);
+        }
+    }
+
+    onMount(() => {
+        loadProviders();
+        // Poll every 10 seconds (providers don't change often)
+        const interval = setInterval(loadProviders, 10000);
+        onCleanup(() => clearInterval(interval));
+    });
+
+    const handleSelect = (providerId: string, modelId: string) => {
+        props.onModelChange({ providerId, modelId });
+        setIsOpen(false);
+    };
+
+    const getCurrentModelName = () => {
+        if (!props.currentModel.modelId) return "Select Model";
+        for (const provider of providerList()) {
+            for (const model of provider.models) {
+                if (model.id === props.currentModel.modelId) {
+                    return model.name;
+                }
+            }
+        }
+        return props.currentModel.modelId;
+    };
+
+    return (
+        <div class="relative">
+            <button
+                onClick={() => setIsOpen(!isOpen())}
+                class="flex items-center justify-between w-full px-3 py-2 bg-zinc-900/60 border border-white/10 rounded-lg text-left hover:border-indigo-500/30 transition-all"
+            >
+                <div class="flex items-center gap-2">
+                    <Cpu size={14} class="text-emerald-400" />
+                    <span class="text-[11px] font-bold text-zinc-200 truncate">
+                        {getCurrentModelName()}
+                    </span>
+                </div>
+                <ChevronDown size={12} class={`text-zinc-500 transition-transform ${isOpen() ? "rotate-180" : ""}`} />
+            </button>
+
+            <Show when={isOpen()}>
+                <div class="absolute top-full left-0 right-0 mt-1 bg-zinc-900 border border-white/10 rounded-lg shadow-xl z-50 max-h-64 overflow-y-auto">
+                    <For each={providerList()}>
+                        {(provider) => (
+                            <div>
+                                <div class="px-3 py-1.5 text-[9px] font-bold text-zinc-500 uppercase tracking-wide bg-zinc-950/50 sticky top-0">
+                                    {provider.name}
+                                </div>
+                                <For each={provider.models}>
+                                    {(model) => (
+                                        <button
+                                            onClick={() => handleSelect(provider.id, model.id)}
+                                            class={`w-full px-3 py-2 text-left hover:bg-white/5 transition-colors flex items-center gap-2 ${props.currentModel.modelId === model.id ? "bg-emerald-500/10 text-emerald-300" : "text-zinc-300"
+                                                }`}
+                                        >
+                                            <Cpu size={12} class="text-zinc-500" />
+                                            <span class="text-[11px] font-medium truncate">{model.name}</span>
+                                        </button>
+                                    )}
+                                </For>
+                            </div>
+                        )}
+                    </For>
+                    <Show when={providerList().length === 0}>
+                        <div class="px-3 py-2 text-[10px] text-zinc-600">No models available</div>
+                    </Show>
+                </div>
+            </Show>
+        </div>
+    );
+}
--- a/packages/ui/src/components/chat/multix-v2/features/LiteSkillsSelector.tsx
+++ b/packages/ui/src/components/chat/multix-v2/features/LiteSkillsSelector.tsx
@@ -0,0 +1,230 @@
+/**
+ * LiteSkillsSelector - Non-reactive skills selector for MultiX v2
+ * 
+ * Uses polling instead of reactive subscriptions to prevent cascading updates.
+ * Displays selected skills as chips with ability to add/remove.
+ */
+
+import { createSignal, For, onMount, onCleanup, Show } from "solid-js";
+import { catalog, catalogLoading, loadCatalog } from "@/stores/skills";
+import { getSessionSkills, setSessionSkills } from "@/stores/session-state";
+import { ChevronDown, Sparkles, X, Check, Loader2 } from "lucide-solid";
+import type { SkillSelection } from "@/types/session";
+
+interface LiteSkillsSelectorProps {
+    instanceId: string;
+    sessionId: string;
+}
+
+interface SkillInfo {
+    id: string;
+    name: string;
+    description?: string;
+}
+
+export function LiteSkillsSelector(props: LiteSkillsSelectorProps) {
+    const [isOpen, setIsOpen] = createSignal(false);
+    const [skillList, setSkillList] = createSignal<SkillInfo[]>([]);
+    const [selectedSkills, setSelectedSkills] = createSignal<SkillSelection[]>([]);
+    const [isLoading, setIsLoading] = createSignal(false);
+    const [filterText, setFilterText] = createSignal("");
+
+    // Load skills once on mount, then poll
+    function loadSkills() {
+        try {
+            const skills = catalog();
+            setSkillList(skills.map((s) => ({
+                id: s.id,
+                name: s.name || s.id,
+                description: s.description
+            })));
+        } catch (e) {
+            console.warn("Failed to load skills", e);
+        }
+    }
+
+    function loadSelected() {
+        try {
+            const skills = getSessionSkills(props.instanceId, props.sessionId);
+            setSelectedSkills(skills);
+        } catch (e) {
+            console.warn("Failed to load selected skills", e);
+        }
+    }
+
+    onMount(async () => {
+        // Load catalog if not already loaded
+        if (catalog().length === 0) {
+            setIsLoading(true);
+            await loadCatalog();
+            setIsLoading(false);
+        }
+        loadSkills();
+        loadSelected();
+
+        // Poll every 2 seconds
+        const interval = setInterval(() => {
+            loadSkills();
+            loadSelected();
+        }, 2000);
+        onCleanup(() => clearInterval(interval));
+    });
+
+    const toggleSkill = (skill: SkillInfo) => {
+        const current = selectedSkills();
+        const isSelected = current.some(s => s.id === skill.id);
+
+        let next: SkillSelection[];
+        if (isSelected) {
+            next = current.filter(s => s.id !== skill.id);
+        } else {
+            next = [...current, { id: skill.id, name: skill.name, description: skill.description }];
+        }
+
+        setSelectedSkills(next);
+        setSessionSkills(props.instanceId, props.sessionId, next);
+    };
+
+    const removeSkill = (id: string) => {
+        const next = selectedSkills().filter(s => s.id !== id);
+        setSelectedSkills(next);
+        setSessionSkills(props.instanceId, props.sessionId, next);
+    };
+
+    const filteredSkills = () => {
+        const term = filterText().toLowerCase().trim();
+        if (!term) return skillList();
+        return skillList().filter(s =>
+            s.name.toLowerCase().includes(term) ||
+            s.id.toLowerCase().includes(term) ||
+            (s.description?.toLowerCase().includes(term) ?? false)
+        );
+    };
+
+    const isSkillSelected = (id: string) => selectedSkills().some(s => s.id === id);
+
+    return (
+        <div class="relative w-full">
+            {/* Main Button */}
+            <button
+                onClick={() => setIsOpen(!isOpen())}
+                class="flex items-center justify-between w-full px-3 py-2 bg-zinc-900/60 border border-white/10 rounded-lg text-left hover:border-purple-500/30 transition-all"
+            >
+                <div class="flex items-center gap-2 min-w-0 flex-1">
+                    <Sparkles size={14} class="text-purple-400 shrink-0" />
+                    <Show
+                        when={selectedSkills().length > 0}
+                        fallback={<span class="text-[11px] text-zinc-500">No skills</span>}
+                    >
+                        <div class="flex items-center gap-1 overflow-hidden">
+                            <span class="text-[11px] font-bold text-purple-300">
+                                {selectedSkills().length} skill{selectedSkills().length !== 1 ? 's' : ''}
+                            </span>
+                            <For each={selectedSkills().slice(0, 2)}>
+                                {(skill) => (
+                                    <span class="text-[10px] px-1.5 py-0.5 bg-purple-500/20 text-purple-300 rounded truncate max-w-[80px]">
+                                        {skill.name}
+                                    </span>
+                                )}
+                            </For>
+                            <Show when={selectedSkills().length > 2}>
+                                <span class="text-[10px] text-zinc-500">+{selectedSkills().length - 2}</span>
+                            </Show>
+                        </div>
+                    </Show>
+                </div>
+                <ChevronDown size={12} class={`text-zinc-500 transition-transform shrink-0 ${isOpen() ? "rotate-180" : ""}`} />
+            </button>
+
+            {/* Dropdown */}
+            <Show when={isOpen()}>
+                <div class="absolute top-full left-0 right-0 mt-1 bg-zinc-900 border border-white/10 rounded-lg shadow-xl z-50 max-h-80 overflow-hidden flex flex-col">
+                    {/* Selected Skills Chips */}
+                    <Show when={selectedSkills().length > 0}>
+                        <div class="px-3 py-2 border-b border-white/5 flex flex-wrap gap-1">
+                            <For each={selectedSkills()}>
+                                {(skill) => (
+                                    <span class="inline-flex items-center gap-1 px-2 py-0.5 bg-purple-500/20 text-purple-300 rounded-full text-[10px]">
+                                        {skill.name}
+                                        <button
+                                            onClick={(e) => {
+                                                e.stopPropagation();
+                                                removeSkill(skill.id);
+                                            }}
+                                            class="hover:text-red-400"
+                                        >
+                                            <X size={10} />
+                                        </button>
+                                    </span>
+                                )}
+                            </For>
+                        </div>
+                    </Show>
+
+                    {/* Filter Input */}
+                    <div class="px-3 py-2 border-b border-white/5">
+                        <input
+                            type="text"
+                            placeholder="Filter skills..."
+                            value={filterText()}
+                            onInput={(e) => setFilterText(e.currentTarget.value)}
+                            class="w-full bg-white/5 border border-white/10 rounded px-2 py-1 text-xs text-zinc-200 outline-none focus:border-purple-500/40"
+                        />
+                    </div>
+
+                    {/* Skills List */}
+                    <div class="overflow-y-auto flex-1 max-h-48">
+                        <Show
+                            when={!isLoading() && !catalogLoading()}
+                            fallback={
+                                <div class="px-3 py-4 text-center text-[11px] text-zinc-500 flex items-center justify-center gap-2">
+                                    <Loader2 size={12} class="animate-spin" />
+                                    Loading skills...
+                                </div>
+                            }
+                        >
+                            <Show
+                                when={filteredSkills().length > 0}
+                                fallback={
+                                    <div class="px-3 py-4 text-center text-[11px] text-zinc-500">
+                                        No skills found
+                                    </div>
+                                }
+                            >
+                                <For each={filteredSkills()}>
+                                    {(skill) => (
+                                        <button
+                                            onClick={() => toggleSkill(skill)}
+                                            class={`w-full px-3 py-2 text-left hover:bg-white/5 transition-colors flex items-center gap-2 ${isSkillSelected(skill.id) ? "bg-purple-500/10" : ""
+                                                }`}
+                                        >
+                                            <div class={`w-4 h-4 rounded border flex items-center justify-center shrink-0 ${isSkillSelected(skill.id)
+                                                    ? "bg-purple-500 border-purple-500"
+                                                    : "border-white/20"
+                                                }`}>
+                                                <Show when={isSkillSelected(skill.id)}>
+                                                    <Check size={10} class="text-white" />
+                                                </Show>
+                                            </div>
+                                            <div class="flex-1 min-w-0">
+                                                <div class={`text-[11px] font-medium truncate ${isSkillSelected(skill.id) ? "text-purple-300" : "text-zinc-300"
+                                                    }`}>
+                                                    {skill.name}
+                                                </div>
+                                                <Show when={skill.description}>
+                                                    <div class="text-[10px] text-zinc-500 truncate">
+                                                        {skill.description}
+                                                    </div>
+                                                </Show>
+                                            </div>
+                                        </button>
+                                    )}
+                                </For>
+                            </Show>
+                        </Show>
+                    </div>
+                </div>
+            </Show>
+        </div>
+    );
+}
--- a/packages/ui/src/components/chat/multix-v2/features/MessageNavSidebar.tsx
+++ b/packages/ui/src/components/chat/multix-v2/features/MessageNavSidebar.tsx
@@ -0,0 +1,87 @@
+/**
+ * MessageNavSidebar - Quick navigation for messages
+ * 
+ * Shows YOU/ASST labels with hover preview.
+ */
+
+import { For, Show, createSignal, type Accessor } from "solid-js";
+import type { InstanceMessageStore } from "@/stores/message-v2/instance-store";
+
+interface MessageNavSidebarProps {
+    messageIds: Accessor<string[]>;
+    store: () => InstanceMessageStore;
+    scrollContainer: HTMLDivElement | undefined;
+    onTabClick: (messageId: string) => void;
+}
+
+export function MessageNavSidebar(props: MessageNavSidebarProps) {
+    return (
+        <div class="w-14 shrink-0 bg-zinc-900/40 border-l border-white/5 overflow-hidden py-2 px-1.5 flex flex-col items-center gap-1">
+            <For each={props.messageIds()}>
+                {(messageId, index) => {
+                    const [showPreview, setShowPreview] = createSignal(false);
+
+                    const msg = () => props.store().getMessage(messageId);
+                    const isUser = () => msg()?.role === "user";
+
+                    // Get message preview text (first 150 chars)
+                    const previewText = () => {
+                        const message = msg();
+                        if (!message) return "";
+
+                        // Try to get text from parts
+                        const parts = message.parts || {};
+                        let text = "";
+                        for (const partId of Object.keys(parts)) {
+                            const partRecord = parts[partId];
+                            if (partRecord?.data?.type === "text") {
+                                text = (partRecord.data as any).text || "";
+                                break;
+                            }
+                        }
+
+                        // Fallback to direct content
+                        if (!text && (message as any).content) {
+                            text = (message as any).content;
+                        }
+
+                        return text.length > 150 ? text.substring(0, 150) + "..." : text;
+                    };
+
+                    return (
+                        <div class="relative group">
+                            <button
+                                onClick={() => props.onTabClick(messageId)}
+                                onMouseEnter={() => setShowPreview(true)}
+                                onMouseLeave={() => setShowPreview(false)}
+                                class={`w-10 py-1.5 rounded text-[8px] font-black uppercase transition-all cursor-pointer ${isUser()
+                                    ? "bg-indigo-500/20 border border-indigo-500/40 text-indigo-400 hover:bg-indigo-500/40 hover:scale-105"
+                                    : "bg-emerald-500/20 border border-emerald-500/40 text-emerald-400 hover:bg-emerald-500/40 hover:scale-105"
+                                    }`}
+                            >
+                                {isUser() ? "YOU" : "ASST"}
+                            </button>
+
+                            {/* Hover Preview Tooltip */}
+                            <Show when={showPreview()}>
+                                <div class="absolute right-full mr-2 top-0 w-72 max-h-40 overflow-y-auto bg-zinc-900 border border-white/10 rounded-lg shadow-xl p-3 z-50 animate-in fade-in slide-in-from-right-2 duration-150 custom-scrollbar">
+                                    <div class="flex items-center justify-between mb-2">
+                                        <div class={`text-[9px] font-bold uppercase ${isUser() ? "text-indigo-400" : "text-emerald-400"}`}>
+                                            {isUser() ? "You" : "Assistant"} • Msg {index() + 1}
+                                        </div>
+                                        <div class="text-[8px] text-zinc-600">
+                                            {msg()?.status === "streaming" ? "• Streaming" : ""}
+                                        </div>
+                                    </div>
+                                    <p class="text-[10px] text-zinc-300 leading-relaxed whitespace-pre-wrap">
+                                        {previewText()}
+                                    </p>
+                                </div>
+                            </Show>
+                        </div>
+                    );
+                }}
+            </For>
+        </div>
+    );
+}
--- a/packages/ui/src/components/chat/multix-v2/features/PipelineView.tsx
+++ b/packages/ui/src/components/chat/multix-v2/features/PipelineView.tsx
@@ -0,0 +1,89 @@
+/**
+ * PipelineView - Task Dashboard
+ * 
+ * Shows all active tasks as cards when no task is selected.
+ */
+
+import { For, Show, type Accessor } from "solid-js";
+import { Plus, ChevronRight, X } from "lucide-solid";
+import type { Task } from "@/types/session";
+
+interface PipelineViewProps {
+    visibleTasks: Accessor<Task[]>;
+    onTaskClick: (taskId: string) => void;
+    onArchiveTask: (taskId: string) => void;
+}
+
+export function PipelineView(props: PipelineViewProps) {
+    return (
+        <div class="p-4 space-y-8 animate-in fade-in slide-in-from-bottom-4 duration-500">
+            <div class="space-y-2">
+                <h2 class="text-2xl font-black text-white tracking-tight leading-none">Pipeline</h2>
+                <p class="text-xs font-medium text-zinc-500 uppercase tracking-[0.2em]">Agentic Orchestration</p>
+            </div>
+
+            <div class="space-y-4">
+                <div class="flex items-center justify-between">
+                    <span class="text-[10px] font-bold text-zinc-600 uppercase tracking-widest">Active Threads</span>
+                    <div class="h-px flex-1 bg-white/5 mx-4" />
+                    <span class="text-[10px] font-black text-indigo-400 bg-indigo-500/10 px-2 py-0.5 rounded border border-indigo-500/20">
+                        {props.visibleTasks().length}
+                    </span>
+                </div>
+
+                <div class="grid gap-3">
+                    <Show when={props.visibleTasks().length === 0}>
+                        <div class="group relative p-8 rounded-3xl border border-dashed border-white/5 bg-zinc-900/20 flex flex-col items-center justify-center text-center space-y-4 transition-all hover:bg-zinc-900/40 hover:border-white/10">
+                            <div class="w-12 h-12 rounded-2xl bg-white/5 flex items-center justify-center text-zinc-600 group-hover:text-indigo-400 group-hover:scale-110 transition-all duration-500">
+                                <Plus size={24} strokeWidth={1.5} />
+                            </div>
+                            <div class="space-y-1">
+                                <p class="text-sm font-bold text-zinc-400">No active tasks</p>
+                                <p class="text-[11px] text-zinc-600">Send a message below to start a new thread</p>
+                            </div>
+                        </div>
+                    </Show>
+
+                    <For each={props.visibleTasks()}>
+                        {(task) => (
+                            <button
+                                onClick={() => props.onTaskClick(task.id)}
+                                class={`group relative p-4 rounded-2xl border border-white/5 bg-zinc-900/40 hover:bg-zinc-800/60 hover:border-indigo-500/30 transition-all duration-300 text-left flex items-start space-x-4 active:scale-[0.98] ${task.title.toLowerCase().includes("smart fix") ? "smart-fix-highlight" : ""}`}
+                            >
+                                <div class={`mt-1 w-2 h-2 rounded-full shadow-[0_0_10px_rgba(var(--color),0.5)] ${task.status === "completed" ? "bg-emerald-500 shadow-emerald-500/40" :
+                                    task.status === "in-progress" ? "bg-indigo-500 shadow-indigo-500/40 animate-pulse" :
+                                        "bg-zinc-600 shadow-zinc-600/20"
+                                    }`} />
+                                <div class="flex-1 min-w-0 space-y-1">
+                                    <p class="text-sm font-bold text-zinc-100 truncate group-hover:text-white transition-colors">
+                                        {task.title}
+                                    </p>
+                                    <div class="flex items-center space-x-3 text-[10px] font-bold text-zinc-500 uppercase tracking-tight">
+                                        <span>{new Date(task.timestamp).toLocaleTimeString([], { hour: '2-digit', minute: '2-digit' })}</span>
+                                        <span class="w-1 h-1 rounded-full bg-zinc-800" />
+                                        <span>{task.messageIds?.length || 0} messages</span>
+                                    </div>
+                                </div>
+                                <div class="flex items-center space-x-2">
+                                    <span
+                                        role="button"
+                                        tabindex={0}
+                                        onClick={(event) => {
+                                            event.stopPropagation();
+                                            props.onArchiveTask(task.id);
+                                        }}
+                                        class="text-zinc-600 hover:text-zinc-200 transition-colors"
+                                        title="Archive task"
+                                    >
+                                        <X size={14} />
+                                    </span>
+                                    <ChevronRight size={16} class="text-zinc-700 group-hover:text-indigo-400 group-hover:translate-x-1 transition-all" />
+                                </div>
+                            </button>
+                        )}
+                    </For>
+                </div>
+            </div>
+        </div>
+    );
+}
--- a/packages/ui/src/components/chat/multix-v2/features/PromptEnhancer.ts
+++ b/packages/ui/src/components/chat/multix-v2/features/PromptEnhancer.ts
@@ -0,0 +1,155 @@
+/**
+ * PromptEnhancer - Clavix-inspired prompt optimization
+ * 
+ * Source: https://github.com/ClavixDev/Clavix.git
+ * 
+ * Takes a user's raw input and refines it into a precise,
+ * context-aware, actionable prompt using the session's configured model.
+ */
+
+import { getLogger } from "@/lib/logger";
+import { sessions } from "@/stores/session-state";
+
+const log = getLogger("prompt-enhancer");
+
+// The meta-prompt based on Clavix CLEAR framework
+const ENHANCEMENT_PROMPT = `You are an ELITE Software Architect and Prompt Engineer, powered by the "ThoughtBox" reasoning engine.
+
+YOUR MISSION:
+Transform the user's raw input into a "God-Tier" System Prompt—a comprehensive, execution-ready technical specification that a senior engineer could implement without further questions.
+
+TARGET OUTPUT:
+- Detailed, file-level architectural blueprint
+- Explicit coding standards (TypeScript/solid-js/tailwindcss context implied)
+- Comprehensive error handling and edge case strategy
+- Step-by-step implementation plan
+
+METHODOLOGY (ThoughtBox):
+1. **Decode Intent**: What is the root problem? What is the *value*?
+2. **Context Inference**: Assume a high-performance TypeScript/React/Electron environment. Infer necessary imports, stores, and services.
+3. **Architectural Strategy**: Define the component hierarchy, state management (signals/stores), and side effects.
+4. **Specification Generation**: Write the actual prompt.
+
+OUTPUT FORMAT:
+Return ONLY the enhanced prompt string, formatted as follows:
+
+# 🎯 OBJECTIVE
+[Concise, high-level goal]
+
+# 🏗️ ARCHITECTURE & DESIGN
+- **Files**: List exact file paths to touch/create.
+- **Components**: Define props, state, and interfaces.
+- **Data Flow**: Explain signal/store interactions.
+
+# 🛡️ RESTRICTIONS & STANDARDS
+- **Tech Stack**: TypeScript, SolidJS, TailwindCSS, Lucide Icons.
+- **Rules**: NO placeholders, NO "todo", Strict Types, Accessibility-first.
+- **Context**: [Infer from input, e.g., "Use ContextEngine for retrieval"]
+
+# 📝 IMPLEMENTATION PLAN
+1. [Step 1: Description]
+2. [Step 2: Description]
+...
+
+# 💡 ORIGINAL REQUEST
+"""
+{INPUT}
+"""
+`;
+
+/**
+ * Get the model configured for a session
+ */
+function getSessionModel(instanceId: string, sessionId: string): string {
+    try {
+        const instanceSessions = sessions().get(instanceId);
+        const session = instanceSessions?.get(sessionId);
+        if (session?.model?.modelId) {
+            return session.model.modelId;
+        }
+    } catch (e) {
+        log.warn("Could not get session model", e);
+    }
+    return "minimax-m1"; // Fallback
+}
+
+/**
+ * Enhance a user's prompt using the session's AI model
+ */
+export async function enhancePrompt(
+    userInput: string,
+    instanceId: string,
+    sessionId?: string
+): Promise<string> {
+    if (!userInput.trim()) {
+        return userInput;
+    }
+
+    // Get the model from the session
+    const model = sessionId ? getSessionModel(instanceId, sessionId) : "minimax-m1";
+
+    log.info("Enhancing prompt...", { length: userInput.length, model });
+
+    try {
+        // Call the Ollama API for enhancement using the session's model
+        const response = await fetch("/api/ollama/chat", {
+            method: "POST",
+            headers: { "Content-Type": "application/json" },
+            body: JSON.stringify({
+                model,
+                messages: [
+                    {
+                        role: "user",
+                        content: ENHANCEMENT_PROMPT.replace("{INPUT}", userInput)
+                    }
+                ],
+                stream: false
+            })
+        });
+
+        if (!response.ok) {
+            log.warn("Enhancement API failed, returning original", { status: response.status });
+            return userInput;
+        }
+
+        const data = await response.json();
+        const enhanced = data?.message?.content || data?.choices?.[0]?.message?.content;
+
+        if (!enhanced || enhanced.trim().length === 0) {
+            log.warn("Enhancement returned empty, using original");
+            return userInput;
+        }
+
+        log.info("Prompt enhanced successfully", {
+            originalLength: userInput.length,
+            enhancedLength: enhanced.length,
+            model
+        });
+
+        return enhanced.trim();
+    } catch (error) {
+        log.error("Prompt enhancement failed", error);
+        return userInput;
+    }
+}
+
+/**
+ * Get a quick suggestion for improving a prompt (synchronous hint)
+ */
+export function getQuickTips(userInput: string): string[] {
+    const tips: string[] = [];
+
+    if (userInput.length < 20) {
+        tips.push("Add more context for better results");
+    }
+
+    if (!userInput.includes("file") && !userInput.includes("function") && !userInput.includes("component")) {
+        tips.push("Mention specific files or functions if applicable");
+    }
+
+    if (!userInput.match(/\b(create|fix|update|add|remove|refactor)\b/i)) {
+        tips.push("Start with an action verb: create, fix, update, etc.");
+    }
+
+    return tips;
+}
--- a/packages/ui/src/components/chat/multix-v2/index.tsx
+++ b/packages/ui/src/components/chat/multix-v2/index.tsx
@@ -0,0 +1,849 @@
+/**
+ * MultiX v2 - Main Entry Point
+ * 
+ * A complete rebuild of the MultiTaskChat component with:
+ * 1. Local signals + polling (no reactive cascade = no freeze)
+ * 2. 100% feature parity with original
+ * 3. New features: Context-Engine, Compaction, Prompt Enhancement
+ */
+
+import { createSignal, Show, onMount, For, onCleanup, batch } from "solid-js";
+import toast from "solid-toast";
+import { sessions, activeSessionId, setActiveSession } from "@/stores/session-state";
+import { sendMessage, compactSession, updateSessionAgent, updateSessionModelForSession, forceReset, abortSession } from "@/stores/session-actions";
+import { addTask, setActiveTask, archiveTask } from "@/stores/task-actions";
+import { messageStoreBus } from "@/stores/message-v2/bus";
+import { formatTokenTotal } from "@/lib/formatters";
+import { addToTaskQueue, getSoloState, setActiveTaskId, toggleAutonomous, toggleAutoApproval, toggleApex } from "@/stores/solo-store";
+import { getLogger } from "@/lib/logger";
+import { clearCompactionSuggestion, getCompactionSuggestion } from "@/stores/session-compaction";
+import { emitSessionSidebarRequest } from "@/lib/session-sidebar-events";
+import {
+    Command, Plus, PanelRight, ListTodo, AtSign, Hash, Mic, ArrowUp,
+    ChevronRight, Loader2, X, Zap, Layers, Sparkles, StopCircle, Key,
+    FileArchive, Paperclip, Wand2, Shield,
+} from "lucide-solid";
+// Using Lite versions to avoid reactive cascade
+// import ModelSelector from "@/components/model-selector";
+// import AgentSelector from "@/components/agent-selector";
+import { DebugOverlay, setForceResetFn } from "@/components/debug-overlay";
+import AttachmentChip from "@/components/attachment-chip";
+import { createFileAttachment } from "@/types/attachment";
+import type { InstanceMessageStore } from "@/stores/message-v2/instance-store";
+import type { Task, Session } from "@/types/session";
+
+// Sub-components
+import { SimpleMessageBlock } from "./core/SimpleMessageBlock";
+import { PipelineView } from "./features/PipelineView";
+import { MessageNavSidebar } from "./features/MessageNavSidebar";
+import { enhancePrompt } from "./features/PromptEnhancer";
+import { LiteAgentSelector } from "./features/LiteAgentSelector";
+import { LiteModelSelector } from "./features/LiteModelSelector";
+import { LiteSkillsSelector } from "./features/LiteSkillsSelector";
+import MessageBlockList from "@/components/message-block-list";
+
+const OPEN_ADVANCED_SETTINGS_EVENT = "open-advanced-settings";
+const log = getLogger("multix-v2");
+
+interface MultiXV2Props {
+    instanceId: string;
+    sessionId: string;
+}
+
+export default function MultiXV2(props: MultiXV2Props) {
+    // ============================================================================
+    // LOCAL STATE (No reactive memos on stores - polling instead)
+    // ============================================================================
+    // Per-task sending state (Map of taskId -> boolean)
+    const [sendingTasks, setSendingTasks] = createSignal<Set<string>>(new Set());
+    const [chatInput, setChatInput] = createSignal("");
+    const [isCompacting, setIsCompacting] = createSignal(false);
+    const [attachments, setAttachments] = createSignal<ReturnType<typeof createFileAttachment>[]>([]);
+    const [userScrolling, setUserScrolling] = createSignal(false);
+    const [isEnhancing, setIsEnhancing] = createSignal(false);
+
+    // Cached store values - updated via polling
+    const [tasks, setTasks] = createSignal<Task[]>([]);
+    const [visibleTasks, setVisibleTasks] = createSignal<Task[]>([]);
+    const [selectedTaskId, setSelectedTaskIdLocal] = createSignal<string | null>(null);
+    const [messageIds, setMessageIds] = createSignal<string[]>([]);
+    const [cachedModelId, setCachedModelId] = createSignal("unknown");
+    const [cachedAgent, setCachedAgent] = createSignal("");
+    const [cachedTokensUsed, setCachedTokensUsed] = createSignal(0);
+    const [cachedCost, setCachedCost] = createSignal(0);
+    const [isAgentThinking, setIsAgentThinking] = createSignal(false);
+    const [compactionSuggestion, setCompactionSuggestion] = createSignal<{ reason: string } | null>(null);
+    const [soloState, setSoloState] = createSignal({ isApex: false, isAutonomous: false, autoApproval: false, activeTaskId: null as string | null });
+    const [lastAssistantIndex, setLastAssistantIndex] = createSignal(-1);
+    const [bottomSentinel, setBottomSentinel] = createSignal<HTMLDivElement | null>(null);
+
+    // Helper to check if CURRENT task is sending
+    const isSending = () => {
+        const taskId = selectedTaskId();
+        if (!taskId) return sendingTasks().size > 0; // If no task selected, check if any is sending
+        return sendingTasks().has(taskId);
+    };
+
+    // Helper to set sending state for a task
+    const setTaskSending = (taskId: string, sending: boolean) => {
+        setSendingTasks(prev => {
+            const next = new Set(prev);
+            if (sending) {
+                next.add(taskId);
+            } else {
+                next.delete(taskId);
+            }
+            return next;
+        });
+    };
+
+    let scrollContainer: HTMLDivElement | undefined;
+    let fileInputRef: HTMLInputElement | undefined;
+
+    // ============================================================================
+    // STORE ACCESS HELPERS (Non-reactive reads)
+    // ============================================================================
+    function getSession(): Session | undefined {
+        const instanceSessions = sessions().get(props.instanceId);
+        return instanceSessions?.get(props.sessionId);
+    }
+
+    function getMessageStore(): InstanceMessageStore {
+        return messageStoreBus.getOrCreate(props.instanceId);
+    }
+
+    function getSelectedTask(): Task | undefined {
+        return visibleTasks().find(t => t.id === selectedTaskId());
+    }
+
+    function getActiveTaskSessionId(): string {
+        const task = getSelectedTask();
+        return task?.taskSessionId || props.sessionId;
+    }
+
+    function getActiveTaskSession(): Session | undefined {
+        const sessionId = getActiveTaskSessionId();
+        const instanceSessions = sessions().get(props.instanceId);
+        return instanceSessions?.get(sessionId);
+    }
+
+    // ============================================================================
+    // POLLING-BASED SYNC (Updates local state from stores every 150ms)
+    // ============================================================================
+    function syncFromStore() {
+        try {
+            const session = getSession();
+            if (session) {
+                const allTasks = session.tasks || [];
+                setTasks(allTasks);
+                setVisibleTasks(allTasks.filter(t => !t.archived));
+                // NOTE: Don't overwrite selectedTaskId from store - local state is authoritative
+                // This prevents the reactive cascade when the store updates
+            }
+
+            // Get message IDs for currently selected task
+            const currentTaskId = selectedTaskId();
+            if (currentTaskId) {
+                const task = visibleTasks().find(t => t.id === currentTaskId);
+                if (task) {
+                    const store = getMessageStore();
+                    if (task.taskSessionId) {
+                        setMessageIds(store.getSessionMessageIds(task.taskSessionId));
+                    } else {
+                        setMessageIds(task.messageIds || []);
+                    }
+                } else {
+                    setMessageIds([]);
+                }
+            } else {
+                setMessageIds([]);
+            }
+
+            const taskSession = getActiveTaskSession();
+            if (taskSession?.model?.modelId) {
+                setCachedModelId(taskSession.model.modelId);
+            }
+            if (taskSession?.agent) {
+                setCachedAgent(taskSession.agent);
+            }
+
+            const store = getMessageStore();
+            const usage = store.getSessionUsage(props.sessionId);
+            if (usage) {
+                setCachedTokensUsed(usage.actualUsageTokens ?? 0);
+                setCachedCost(usage.totalCost ?? 0);
+            }
+
+            const ids = messageIds();
+            if (ids.length > 0) {
+                const lastMsg = store.getMessage(ids[ids.length - 1]);
+                setIsAgentThinking(
+                    lastMsg?.role === "assistant" &&
+                    (lastMsg.status === "streaming" || lastMsg.status === "sending")
+                );
+
+                // Calculate lastAssistantIndex
+                let lastIdx = -1;
+                for (let i = ids.length - 1; i >= 0; i--) {
+                    const msg = store.getMessage(ids[i]);
+                    if (msg?.role === "assistant") {
+                        lastIdx = i;
+                        break;
+                    }
+                }
+                setLastAssistantIndex(lastIdx);
+            } else {
+                setIsAgentThinking(false);
+                setLastAssistantIndex(-1);
+            }
+
+            const suggestion = getCompactionSuggestion(props.instanceId, getActiveTaskSessionId());
+            setCompactionSuggestion(suggestion);
+
+            setSoloState(getSoloState(props.instanceId));
+        } catch (e) {
+            log.error("syncFromStore error", e);
+        }
+    }
+
+    // ============================================================================
+    // LIFECYCLE
+    // ============================================================================
+    onMount(() => {
+        setForceResetFn(() => {
+            forceReset();
+            // Clear all sending states on force reset
+            setSendingTasks(new Set<string>());
+        });
+
+        syncFromStore();
+
+        const interval = setInterval(syncFromStore, 150);
+
+        const handleScroll = () => {
+            if (!scrollContainer) return;
+            const isAtBottom = scrollContainer.scrollHeight - scrollContainer.scrollTop - scrollContainer.clientHeight < 50;
+            setUserScrolling(!isAtBottom);
+        };
+        scrollContainer?.addEventListener('scroll', handleScroll, { passive: true });
+
+        onCleanup(() => {
+            clearInterval(interval);
+            scrollContainer?.removeEventListener('scroll', handleScroll);
+        });
+    });
+
+    // ============================================================================
+    // ACTIONS
+    // ============================================================================
+    const scrollToBottom = () => {
+        if (scrollContainer && !userScrolling()) {
+            scrollContainer.scrollTop = scrollContainer.scrollHeight;
+        }
+    };
+
+    const setSelectedTaskId = (id: string | null) => {
+        // Update local state immediately (fast)
+        setSelectedTaskIdLocal(id);
+
+        // Immediately sync to load the new task's agent/model
+        syncFromStore();
+
+        // Defer the global store update using idle callback (non-blocking)
+        if (typeof requestIdleCallback !== 'undefined') {
+            requestIdleCallback(() => {
+                setActiveTask(props.instanceId, props.sessionId, id || undefined);
+            }, { timeout: 500 });
+        } else {
+            // Fallback: use setTimeout with longer delay
+            setTimeout(() => {
+                setActiveTask(props.instanceId, props.sessionId, id || undefined);
+            }, 50);
+        }
+    };
+
+    const handleSendMessage = async () => {
+        const message = chatInput().trim();
+        if (!message) return;
+
+        // Check if THIS specific task is already sending
+        const currentTaskId = selectedTaskId();
+        if (currentTaskId && sendingTasks().has(currentTaskId)) return;
+
+        const currentMessage = message;
+        const currentAttachments = attachments();
+
+        batch(() => {
+            setChatInput("");
+            setAttachments([]);
+        });
+
+        // Track which task we're sending for (might be created below)
+        let taskIdForSending: string | null = null;
+
+        try {
+            let taskId = currentTaskId;
+            let targetSessionId = props.sessionId;
+
+            if (!taskId) {
+                // Create new task
+                const title = currentMessage.length > 30 ? currentMessage.substring(0, 27) + "..." : currentMessage;
+                log.info("[MultiX] Creating task...", { title });
+                const result = await addTask(props.instanceId, props.sessionId, title);
+                taskId = result.id;
+                targetSessionId = result.taskSessionId || props.sessionId;
+                log.info("[MultiX] Task created", { taskId, targetSessionId, hasTaskSession: !!result.taskSessionId });
+
+                // Immediately sync to get the new task in our local state
+                syncFromStore();
+
+                // Set the selected task
+                setSelectedTaskIdLocal(taskId);
+
+                const s = soloState();
+                if (s.isAutonomous) {
+                    if (!s.activeTaskId) {
+                        setActiveTaskId(props.instanceId, taskId);
+                    } else {
+                        addToTaskQueue(props.instanceId, taskId);
+                    }
+                }
+            } else {
+                // Existing task - get up-to-date task info
+                syncFromStore();
+                const task = visibleTasks().find(t => t.id === taskId);
+                targetSessionId = task?.taskSessionId || props.sessionId;
+                log.info("[MultiX] Existing task", { taskId, targetSessionId });
+            }
+
+            // Mark THIS task as sending
+            taskIdForSending = taskId;
+            setTaskSending(taskId, true);
+
+            log.info("[MultiX] Sending message", { instanceId: props.instanceId, targetSessionId, messageLength: currentMessage.length, taskId });
+
+            // Send the message (this is async and will stream)
+            await sendMessage(props.instanceId, targetSessionId, currentMessage, currentAttachments, taskId || undefined);
+            log.info("[MultiX] Message sent successfully");
+
+            // Force sync after message is sent to pick up the new messages
+            setTimeout(() => syncFromStore(), 100);
+            setTimeout(() => syncFromStore(), 500);
+            setTimeout(() => syncFromStore(), 1000);
+
+            setTimeout(scrollToBottom, 150);
+        } catch (error) {
+            log.error("Send failed:", error);
+            console.error("[MultiX] Send failed:", error);
+        } finally {
+            // Clear sending state for this specific task
+            if (taskIdForSending) {
+                setTaskSending(taskIdForSending, false);
+            }
+        }
+    };
+
+    const handleCreateTask = () => {
+        // Allow creating new tasks even when other tasks are processing
+        const nextIndex = tasks().length + 1;
+        const title = `Task ${nextIndex} `;
+
+        setTimeout(async () => {
+            try {
+                const result = await addTask(props.instanceId, props.sessionId, title);
+                setSelectedTaskIdLocal(result.id);
+                setTimeout(() => syncFromStore(), 50);
+            } catch (error) {
+                log.error("handleCreateTask failed", error);
+            }
+        }, 0);
+    };
+
+    const handleKeyDown = (e: KeyboardEvent) => {
+        if (e.key === "Enter" && !e.shiftKey) {
+            e.preventDefault();
+            handleSendMessage();
+        }
+    };
+
+    const handleStopAgent = async (e?: MouseEvent) => {
+        if (e?.shiftKey) {
+            forceReset();
+            // Clear all sending states on force reset
+            setSendingTasks(new Set<string>());
+            return;
+        }
+
+        const task = getSelectedTask();
+        // If no task selected, we might be in global pipeline, use sessionId
+        const targetSessionId = task?.taskSessionId || props.sessionId;
+        const taskId = task?.id || selectedTaskId();
+
+        try {
+            await abortSession(props.instanceId, targetSessionId);
+
+            // Manually force UI update
+            if (taskId) {
+                setTaskSending(taskId, false);
+            }
+            setIsAgentThinking(false);
+
+            setTimeout(() => syncFromStore(), 50);
+        } catch (error) {
+            log.error("Failed to stop agent", error);
+        }
+    };
+
+    const handleCompact = async () => {
+        const targetSessionId = getActiveTaskSessionId();
+        if (isCompacting()) return;
+
+        // Get message count to verify we have messages to compact
+        const store = getMessageStore();
+        const msgIds = store.getSessionMessageIds(targetSessionId);
+
+        log.info("[MultiX] Starting compaction", {
+            instanceId: props.instanceId,
+            sessionId: targetSessionId,
+            messageCount: msgIds.length
+        });
+
+        if (msgIds.length < 3) {
+            log.info("[MultiX] Session too small to compact", { count: msgIds.length });
+            toast.success("Session is already concise. No compaction needed.", {
+                icon: <Zap size={14} class="text-amber-400" />
+            });
+            return;
+        }
+
+        setIsCompacting(true);
+        const toastId = toast.loading("Compacting session history...");
+        try {
+            clearCompactionSuggestion(props.instanceId, targetSessionId);
+            const result = await compactSession(props.instanceId, targetSessionId);
+
+            // CRITICAL: Restore the parent session as active to prevent navigation away from MultiX
+            const currentActive = activeSessionId().get(props.instanceId);
+            if (currentActive !== props.sessionId) {
+                setActiveSession(props.instanceId, props.sessionId);
+            }
+            log.info("[MultiX] Compaction complete", {
+                success: result.success,
+                tokenBefore: result.token_before,
+                tokenAfter: result.token_after,
+                reduction: result.token_reduction_pct
+            });
+
+            toast.success(`Compacted! Reduced by ${result.token_reduction_pct}% (${result.token_after} tokens)`, {
+                id: toastId,
+                duration: 4000
+            });
+
+            // Sync to update UI after compaction
+            syncFromStore();
+        } catch (error) {
+            log.error("Failed to compact session", error);
+            toast.error("Compaction failed. Please try again.", { id: toastId });
+        } finally {
+            setIsCompacting(false);
+        }
+    };
+
+    const handleOpenAdvancedSettings = () => {
+        window.dispatchEvent(new CustomEvent(OPEN_ADVANCED_SETTINGS_EVENT, {
+            detail: { instanceId: props.instanceId, sessionId: props.sessionId }
+        }));
+    };
+
+    const handleEnhancePrompt = async () => {
+        const input = chatInput().trim();
+        if (!input || isEnhancing()) return;
+
+        setIsEnhancing(true);
+        try {
+            // Pass sessionId so it uses the task's configured model
+            const taskSessionId = getActiveTaskSessionId();
+            const enhanced = await enhancePrompt(input, props.instanceId, taskSessionId);
+            setChatInput(enhanced);
+        } catch (error) {
+            log.error("Prompt enhancement failed", error);
+        } finally {
+            setIsEnhancing(false);
+        }
+    };
+
+    const toggleApexPro = () => {
+        const s = soloState();
+        const currentState = s.isAutonomous && s.autoApproval;
+        if (currentState) {
+            if (s.isAutonomous) toggleAutonomous(props.instanceId);
+            if (s.autoApproval) toggleAutoApproval(props.instanceId);
+        } else {
+            if (!s.isAutonomous) toggleAutonomous(props.instanceId);
+            if (!s.autoApproval) toggleAutoApproval(props.instanceId);
+        }
+    };
+
+    const isApexPro = () => {
+        const s = soloState();
+        return s.isAutonomous && s.autoApproval;
+    };
+
+    const handleArchiveTask = (taskId: string) => {
+        archiveTask(props.instanceId, props.sessionId, taskId);
+    };
+
+    const addAttachment = (attachment: ReturnType<typeof createFileAttachment>) => {
+        setAttachments((prev) => [...prev, attachment]);
+    };
+
+    const removeAttachment = (attachmentId: string) => {
+        setAttachments((prev) => prev.filter((item) => item.id !== attachmentId));
+    };
+
+    const handleFileSelect = (event: Event) => {
+        const input = event.currentTarget as HTMLInputElement;
+        if (!input.files || input.files.length === 0) return;
+
+        Array.from(input.files).forEach((file) => {
+            const reader = new FileReader();
+            reader.onload = () => {
+                const buffer = reader.result instanceof ArrayBuffer ? reader.result : null;
+                const data = buffer ? new Uint8Array(buffer) : undefined;
+                const attachment = createFileAttachment(file.name, file.name, file.type || "application/octet-stream", data);
+                if (file.type.startsWith("image/") && typeof reader.result === "string") {
+                    attachment.url = reader.result;
+                }
+                addAttachment(attachment);
+            };
+            reader.readAsArrayBuffer(file);
+        });
+
+        input.value = "";
+    };
+
+    const handleTabClick = (messageId: string) => {
+        const anchorId = `message-anchor-${messageId}`;
+        const element = scrollContainer?.querySelector(`#${anchorId}`);
+        if (element) {
+            element.scrollIntoView({ behavior: "smooth", block: "center" });
+            element.classList.add("message-highlight");
+            setTimeout(() => element.classList.remove("message-highlight"), 2000);
+        }
+    };
+
+    // ============================================================================
+    // RENDER (Gemini 3 Pro)
+    // ============================================================================
+    return (
+        <div class="absolute inset-0 flex flex-col bg-[#0a0a0b] text-zinc-300 font-sans selection:bg-indigo-500/30 overflow-hidden">
+            <DebugOverlay />
+
+            {/* ===== GEMINI 3 PRO HEADER ===== */}
+            <header class="h-12 px-2 flex items-center justify-between bg-[#0a0a0b]/90 backdrop-blur-xl border-b border-white/5 relative z-30 shrink-0 select-none">
+                <div class="flex items-center gap-2 overflow-hidden flex-1">
+                    {/* Brand / Mode Indicator */}
+                    <div class="flex items-center gap-2 px-2 py-1 rounded-md text-zinc-400">
+                        <Layers size={14} class="text-indigo-500" />
+                        <span class="text-[11px] font-bold tracking-wider text-zinc-300">MULTIX</span>
+                    </div>
+
+                    <div class="h-4 w-px bg-white/5 shrink-0" />
+
+                    {/* Pipeline / Task Switcher */}
+                    <div class="flex items-center gap-1 overflow-x-auto no-scrollbar mask-linear-fade">
+                        {/* Pipeline Tab */}
+                        <button
+                            onClick={() => setSelectedTaskId(null)}
+                            class={`flex items-center gap-2 px-3 py-1.5 rounded-lg text-[10px] font-bold uppercase tracking-wider transition-all border ${!selectedTaskId()
+                                ? "bg-indigo-500/10 text-indigo-400 border-indigo-500/20 shadow-[0_0_10px_rgba(99,102,241,0.1)]"
+                                : "text-zinc-500 border-transparent hover:text-zinc-300 hover:bg-white/5"
+                                }`}
+                        >
+                            <span class="font-mono">PIPELINE</span>
+                        </button>
+
+                        {/* Active Tasks */}
+                        <For each={visibleTasks()}>
+                            {(task) => (
+                                <button
+                                    onClick={() => setSelectedTaskId(task.id)}
+                                    class={`group flex items-center gap-2 px-3 py-1.5 rounded-lg text-[10px] font-bold transition-all border max-w-[140px] ${selectedTaskId() === task.id
+                                        ? "bg-zinc-800 text-zinc-100 border-zinc-700 shadow-lg"
+                                        : "text-zinc-500 border-transparent hover:text-zinc-300 hover:bg-white/5"
+                                        } ${task.title.toLowerCase().includes("smart fix") ? "smart-fix-highlight" : ""}`}
+                                >
+                                    <div class={`w-1.5 h-1.5 rounded-full ${task.status === "completed" ? "bg-emerald-500" :
+                                        task.status === "interrupted" ? "bg-rose-500" :
+                                            "bg-indigo-500 animate-pulse"
+                                        }`} />
+                                    <span class="truncate">{task.title}</span>
+                                    <span
+                                        onClick={(e) => { e.stopPropagation(); handleArchiveTask(task.id); }}
+                                        class="opacity-0 group-hover:opacity-100 hover:text-red-400 transition-opacity"
+                                    >
+                                        <X size={10} />
+                                    </span>
+                                </button>
+                            )}
+                        </For>
+
+                        {/* New Task */}
+                        <button
+                            onClick={handleCreateTask}
+                            class="w-6 h-6 flex items-center justify-center rounded-md text-zinc-600 hover:text-zinc-200 hover:bg-white/5 transition-colors"
+                        >
+                            <Plus size={14} />
+                        </button>
+                    </div>
+                </div>
+
+                {/* Right Actions */}
+                <div class="flex items-center gap-2 shrink-0 pl-4">
+                    {/* Stream Status */}
+                    <Show when={isAgentThinking()}>
+                        <div class="flex items-center gap-2 px-2 py-1 rounded-full bg-violet-500/10 border border-violet-500/20">
+                            <Loader2 size={10} class="animate-spin text-violet-400" />
+                            <span class="text-[9px] font-mono text-violet-300">{formatTokenTotal(cachedTokensUsed())}</span>
+                        </div>
+                    </Show>
+
+                    <div class="h-4 w-px bg-white/5" />
+
+                    {/* Tools */}
+                    <button
+                        onClick={handleCompact}
+                        disabled={!selectedTaskId()}
+                        class="p-1.5 text-zinc-500 hover:text-zinc-200 hover:bg-white/5 rounded-md transition-colors disabled:opacity-30"
+                        title="Compact Context"
+                    >
+                        <FileArchive size={14} />
+                    </button>
+
+                    <button
+                        onClick={() => emitSessionSidebarRequest({ instanceId: props.instanceId, action: "show-skills" })}
+                        class="p-1.5 text-zinc-500 hover:text-indigo-300 hover:bg-indigo-500/10 rounded-md transition-colors"
+                        title="Skills"
+                    >
+                        <Sparkles size={14} />
+                    </button>
+                </div>
+            </header>
+
+            {/* ===== AGENT/MODEL SELECTORS (LITE VERSIONS - PER TASK) ===== */}
+            <Show when={getSelectedTask()}>
+                <div class="px-4 py-3 border-b border-white/5 bg-[#0a0a0b]">
+                    <div class="grid grid-cols-1 md:grid-cols-2 gap-2">
+                        <LiteAgentSelector
+                            instanceId={props.instanceId}
+                            sessionId={getActiveTaskSessionId()}
+                            currentAgent={cachedAgent()}
+                            onAgentChange={(agent) => {
+                                // Update the TASK's session, not a global cache
+                                const taskSessionId = getActiveTaskSessionId();
+                                log.info("[MultiX] Changing agent for task session", { taskSessionId, agent });
+                                updateSessionAgent(props.instanceId, taskSessionId, agent);
+                                // Force immediate sync to reflect the change
+                                setTimeout(() => syncFromStore(), 50);
+                            }}
+                        />
+                        <LiteModelSelector
+                            instanceId={props.instanceId}
+                            sessionId={getActiveTaskSessionId()}
+                            currentModel={{ providerId: "", modelId: cachedModelId() }}
+                            onModelChange={(model) => {
+                                // Update the TASK's session, not a global cache
+                                const taskSessionId = getActiveTaskSessionId();
+                                log.info("[MultiX] Changing model for task session", { taskSessionId, model });
+                                updateSessionModelForSession(props.instanceId, taskSessionId, model);
+                                // Force immediate sync to reflect the change
+                                setTimeout(() => syncFromStore(), 50);
+                            }}
+                        />
+                        <LiteSkillsSelector
+                            instanceId={props.instanceId}
+                            sessionId={getActiveTaskSessionId()}
+                        />
+                    </div>
+                </div>
+            </Show>
+
+            {/* ===== MAIN CONTENT AREA (Row Layout) ===== */}
+            <div class="flex-1 flex flex-row min-h-0 relative bg-[#050505] overflow-hidden w-full h-full">
+
+                {/* Chat Column */}
+                <div class="flex-1 min-h-0 flex flex-col overflow-hidden relative">
+                    <div ref={scrollContainer} class="flex-1 min-h-0 overflow-y-auto overflow-x-hidden custom-scrollbar">
+                        {/* Compaction Suggestion Banner */}
+                        <Show when={compactionSuggestion()}>
+                            <div class="mx-3 mt-3 mb-1 rounded-xl border border-emerald-500/30 bg-emerald-500/10 px-3 py-2 text-[11px] text-emerald-200 flex items-center justify-between gap-3">
+                                <span class="font-semibold">Compact suggested: {compactionSuggestion()?.reason}</span>
+                                <button
+                                    type="button"
+                                    class="px-2.5 py-1 rounded-lg text-[10px] font-bold uppercase tracking-wide bg-emerald-500/20 border border-emerald-500/40 text-emerald-200 hover:bg-emerald-500/30 transition-colors"
+                                    onClick={handleCompact}
+                                >
+                                    Compact now
+                                </button>
+                            </div>
+                        </Show>
+
+                        <Show when={!selectedTaskId()} fallback={
+                            /* Message List - Using full MessageBlockList for proper streaming */
+                            <div class="min-h-full pb-4">
+                                <MessageBlockList
+                                    instanceId={props.instanceId}
+                                    sessionId={getActiveTaskSessionId()}
+                                    store={getMessageStore}
+                                    messageIds={() => messageIds()}
+                                    lastAssistantIndex={() => lastAssistantIndex()}
+                                    showThinking={() => true}
+                                    thinkingDefaultExpanded={() => true}
+                                    showUsageMetrics={() => true}
+                                    scrollContainer={() => scrollContainer}
+                                    setBottomSentinel={setBottomSentinel}
+                                />
+                                {/* Bottom anchor */}
+                                <div id="bottom-anchor" class="h-10 w-full" />
+                            </div>
+                        }>
+                            {/* Pipeline View */}
+                            <PipelineView
+                                visibleTasks={visibleTasks}
+                                onTaskClick={setSelectedTaskId}
+                                onArchiveTask={handleArchiveTask}
+                            />
+                        </Show>
+                    </div>
+
+                    {/* ===== INPUT AREA ===== */}
+                    <div class="p-4 bg-[#0a0a0b] border-t border-white/5 shrink-0 z-20">
+                        {/* Input Container */}
+                        <div class="w-full bg-zinc-900/50 border border-white/10 rounded-2xl shadow-sm overflow-hidden focus-within:border-indigo-500/30 transition-all">
+                            {/* Input Header Row */}
+                            <div class="flex items-center justify-between px-3 pt-2 pb-1">
+                                <div class="flex items-center space-x-2">
+                                    <div class="flex flex-col">
+                                        <span class="text-[10px] font-bold text-zinc-400 uppercase tracking-wide">
+                                            {selectedTaskId() ? "Task Context" : "Global Pipeline"}
+                                        </span>
+                                    </div>
+                                </div>
+                                <div class="flex items-center space-x-1">
+                                    {/* APEX / Shield Toggles */}
+                                    <button
+                                        onClick={() => toggleApex(props.instanceId)}
+                                        title="Apex"
+                                        class={`p-1 rounded transition-colors ${soloState().isApex ? "text-rose-400 bg-rose-500/10" : "text-zinc-600 hover:text-zinc-400"}`}
+                                    >
+                                        <Zap size={10} />
+                                    </button>
+                                    <button
+                                        onClick={() => toggleAutoApproval(props.instanceId)}
+                                        title="Shield"
+                                        class={`p-1 rounded transition-colors ${soloState().autoApproval ? "text-emerald-400 bg-emerald-500/10" : "text-zinc-600 hover:text-zinc-400"}`}
+                                    >
+                                        <Shield size={10} />
+                                    </button>
+                                </div>
+                            </div>
+
+                            {/* Attachments */}
+                            <Show when={attachments().length > 0}>
+                                <div class="flex flex-wrap gap-1.5 px-3 py-1">
+                                    <For each={attachments()}>
+                                        {(attachment) => (
+                                            <AttachmentChip
+                                                attachment={attachment}
+                                                onRemove={() => removeAttachment(attachment.id)}
+                                            />
+                                        )}
+                                    </For>
+                                </div>
+                            </Show>
+
+                            {/* Text Input */}
+                            <textarea
+                                value={chatInput()}
+                                onInput={(e) => {
+                                    setChatInput(e.currentTarget.value);
+                                    e.currentTarget.style.height = "auto";
+                                    e.currentTarget.style.height = e.currentTarget.scrollHeight + "px";
+                                }}
+                                onKeyDown={handleKeyDown}
+                                placeholder={selectedTaskId() ? "Message agent..." : "Start a new task..."}
+                                class="w-full bg-transparent text-zinc-200 placeholder-zinc-500 text-sm p-3 outline-none resize-none max-h-[300px] min-h-[44px]"
+                                rows={1}
+                                disabled={isSending()}
+                            />
+
+                            {/* Toolbar */}
+                            <div class="flex items-center justify-between px-2 pb-2 mt-1 border-t border-white/5 pt-2 bg-zinc-900/30">
+                                <div class="flex items-center space-x-1">
+                                    <input
+                                        ref={fileInputRef}
+                                        type="file"
+                                        multiple
+                                        class="hidden"
+                                        onChange={handleFileSelect}
+                                    />
+                                    <button
+                                        onClick={() => fileInputRef?.click()}
+                                        class="p-1.5 text-zinc-500 hover:text-zinc-300 rounded hover:bg-white/5 transition-colors"
+                                    >
+                                        <Paperclip size={14} />
+                                    </button>
+                                    <button
+                                        onClick={handleEnhancePrompt}
+                                        disabled={!chatInput().trim() || isEnhancing()}
+                                        class={`p-1.5 rounded hover:bg-white/5 transition-colors ${isEnhancing() ? "text-amber-400 animate-pulse" : "text-zinc-500 hover:text-amber-300"}`}
+                                    >
+                                        <Wand2 size={14} class={isEnhancing() ? "animate-spin" : ""} />
+                                    </button>
+                                </div>
+
+                                <div class="flex items-center space-x-2">
+                                    <div class="text-[9px] text-zinc-600 font-mono hidden md:block">
+                                        {cachedModelId()}
+                                    </div>
+
+                                    {/* Stop Button (visible when agent is thinking) */}
+                                    <Show when={isAgentThinking() || isSending()}>
+                                        <button
+                                            onClick={handleStopAgent}
+                                            class="p-1.5 bg-rose-500/20 hover:bg-rose-500/30 text-rose-400 border border-rose-500/30 rounded-lg transition-all shadow-sm"
+                                            title="Stop Agent (Shift+Click = Force Reset)"
+                                        >
+                                            <StopCircle size={14} strokeWidth={2.5} />
+                                        </button>
+                                    </Show>
+
+                                    {/* Send Button */}
+                                    <button
+                                        onClick={handleSendMessage}
+                                        disabled={(!chatInput().trim() && attachments().length === 0) || isSending()}
+                                        class="p-1.5 bg-zinc-100 hover:bg-white text-black rounded-lg disabled:opacity-50 disabled:cursor-not-allowed transition-all shadow-sm"
+                                    >
+                                        <Show when={isSending()} fallback={<ArrowUp size={14} strokeWidth={3} />}>
+                                            <Loader2 size={14} class="animate-spin" />
+                                        </Show>
+                                    </button>
+                                </div>
+                            </div>
+                        </div>
+                    </div>
+                </div>
+
+                {/* Sidebar (Right) */}
+                <Show when={selectedTaskId() && messageIds().length > 0}>
+                    <MessageNavSidebar
+                        messageIds={messageIds}
+                        store={getMessageStore}
+                        scrollContainer={scrollContainer}
+                        onTabClick={handleTabClick}
+                    />
+                </Show>
+            </div>
+        </div>
+    );
+}
--- a/packages/ui/src/components/debug-overlay.tsx
+++ b/packages/ui/src/components/debug-overlay.tsx
@@ -0,0 +1,100 @@
+import { createSignal, onMount, onCleanup, Show } from "solid-js"
+
+// Simple debug log storage (no reactive overhead)
+export function addDebugLog(message: string, level: "info" | "warn" | "error" = "info") {
+    // Disabled - no-op for performance
+}
+
+// HARD STOP function - forces page reload
+function hardStop() {
+    console.warn("HARD STOP triggered - reloading page")
+    window.location.reload()
+}
+
+// Force reset function import placeholder
+let forceResetFn: (() => void) | null = null
+
+export function setForceResetFn(fn: () => void) {
+    forceResetFn = fn
+}
+
+export function DebugOverlay() {
+    const [visible, setVisible] = createSignal(false)
+
+    // Toggle with Ctrl+Shift+D
+    onMount(() => {
+        const handleKeyDown = (e: KeyboardEvent) => {
+            if (e.ctrlKey && e.shiftKey && e.key === "D") {
+                setVisible((v) => !v)
+            }
+        }
+        window.addEventListener("keydown", handleKeyDown)
+        onCleanup(() => window.removeEventListener("keydown", handleKeyDown))
+    })
+
+    return (
+        <Show when={visible()}>
+            <div
+                style={{
+                    position: "fixed",
+                    top: "10px",
+                    right: "10px",
+                    "z-index": "99999",
+                    background: "rgba(0,0,0,0.9)",
+                    color: "#fff",
+                    padding: "12px",
+                    "border-radius": "8px",
+                    "font-family": "monospace",
+                    "font-size": "11px",
+                    "min-width": "200px",
+                    border: "1px solid #333",
+                    "pointer-events": "auto",
+                }}
+            >
+                <div style={{ "margin-bottom": "8px", "font-weight": "bold" }}>
+                    DEBUG PANEL (Ctrl+Shift+D to toggle)
+                </div>
+
+                <div style={{ display: "flex", gap: "8px" }}>
+                    <button
+                        onClick={() => {
+                            if (forceResetFn) forceResetFn()
+                        }}
+                        style={{
+                            background: "#f59e0b",
+                            color: "#000",
+                            border: "none",
+                            padding: "6px 12px",
+                            "border-radius": "4px",
+                            cursor: "pointer",
+                            "font-weight": "bold",
+                            "font-size": "10px",
+                        }}
+                    >
+                        RESET UI
+                    </button>
+
+                    <button
+                        onClick={hardStop}
+                        style={{
+                            background: "#ef4444",
+                            color: "#fff",
+                            border: "none",
+                            padding: "6px 12px",
+                            "border-radius": "4px",
+                            cursor: "pointer",
+                            "font-weight": "bold",
+                            "font-size": "10px",
+                        }}
+                    >
+                        HARD RELOAD
+                    </button>
+                </div>
+
+                <div style={{ "margin-top": "8px", "font-size": "9px", color: "#888" }}>
+                    If stuck: Click HARD RELOAD or press F5
+                </div>
+            </div>
+        </Show>
+    )
+}
--- a/packages/ui/src/components/folder-selection-view.tsx
+++ b/packages/ui/src/components/folder-selection-view.tsx
@@ -295,7 +295,7 @@ const FolderSelectionView: Component<FolderSelectionViewProps> = (props) => {
              <img src={nomadArchLogo} alt="NomadArch logo" class="h-32 w-auto sm:h-48" loading="lazy" />
            </div>
            <h1 class="mb-2 text-3xl font-semibold text-primary">NomadArch</h1>
-            <p class="text-xs text-muted mb-1">Forked from OpenCode</p>
+            <p class="text-xs text-muted mb-1">An enhanced fork of CodeNomad</p>
            <Show when={activeUser()}>
              {(user) => (
                <p class="text-xs text-muted mb-1">
--- a/packages/ui/src/components/instance/instance-shell2.tsx
+++ b/packages/ui/src/components/instance/instance-shell2.tsx
@@ -10,6 +10,7 @@ import {
  type Accessor,
  type Component,
 } from "solid-js"
+import toast from "solid-toast"
 import type { ToolState } from "@opencode-ai/sdk"
 import { Accordion } from "@kobalte/core"
 import { ChevronDown } from "lucide-solid"
@@ -36,8 +37,11 @@ import {
  sessions,
  setActiveSession,
  executeCustomCommand,
+  sendMessage,
  runShellCommand,
 } from "../../stores/sessions"
+import { compactSession } from "../../stores/session-actions";
+import { addTask, setActiveTask } from "../../stores/task-actions"
 import { keyboardRegistry, type KeyboardShortcut } from "../../lib/keyboard-registry"
 import { messageStoreBus } from "../../stores/message-v2/bus"
 import { clearSessionRenderCache } from "../message-block"
@@ -54,14 +58,15 @@ import ModelSelector from "../model-selector"
 import ModelStatusSelector from "../model-status-selector"
 import CommandPalette from "../command-palette"
 import Kbd from "../kbd"
-import MultiTaskChat from "../chat/multi-task-chat"
+// Using rebuilt MultiX v2 with polling architecture (no freeze)
+import MultiTaskChat from "../chat/multix-v2"
 import { TodoListView } from "../tool-call/renderers/todo"
 import ContextUsagePanel from "../session/context-usage-panel"
 import SessionView from "../session/session-view"
 import { Sidebar, type FileNode } from "./sidebar"
 import { Editor } from "./editor"
 import { serverApi } from "../../lib/api-client"
-import { Sparkles, Layout as LayoutIcon, Terminal as TerminalIcon, Search, Loader2, Zap, Shield, Settings } from "lucide-solid"
+import { Sparkles, Layout as LayoutIcon, Terminal as TerminalIcon, Search, Loader2, Zap, Shield, Settings, FileArchive } from "lucide-solid"
 import { formatTokenTotal } from "../../lib/formatters"
 import { sseManager } from "../../lib/sse-manager"
 import { getLogger } from "../../lib/logger"
@@ -159,18 +164,32 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {
  const [selectedBinary, setSelectedBinary] = createSignal("opencode")

  // Handler to load file content when selected
+  createEffect(() => {
+    if (typeof window !== "undefined") {
+      (window as any).ACTIVE_INSTANCE_ID = props.instance.id;
+    }
+  });
+
  const handleFileSelect = async (file: FileNode) => {
    try {
      const response = await serverApi.readWorkspaceFile(props.instance.id, file.path)
      const language = file.name.split('.').pop() || 'text'
-      setCurrentFile({
+      const updatedFile = {
        ...file,
        content: response.contents,
        language,
-      })
+      }
+      setCurrentFile(updatedFile)
+
+      // If it's a previewable file, update the preview URL
+      if (file.name.endsWith('.html') || file.name.endsWith('.htm')) {
+        const origin = typeof window !== "undefined" ? window.location.origin : "http://localhost:3000"
+        const apiOrigin = origin.replace(":3000", ":9898")
+        const url = `${apiOrigin}/api/workspaces/${props.instance.id}/serve/${file.path}`
+        setPreviewUrl(url)
+      }
    } catch (error) {
      log.error('Failed to read file content', error)
-      // Still show the file but without content
      setCurrentFile(file)
    }
  }
@@ -292,21 +311,55 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {
    if (typeof window === "undefined") return
    const handler = async (event: Event) => {
      const detail = (event as CustomEvent<{ url?: string; instanceId?: string }>).detail
+      console.log(`[InstanceShell2] Received BUILD_PREVIEW_EVENT`, {
+        detail,
+        currentInstanceId: props.instance.id,
+        match: detail?.instanceId === props.instance.id
+      });
      if (!detail || detail.instanceId !== props.instance.id || !detail.url) return
+
      setPreviewUrl(detail.url)
-      const confirmed = await showConfirmDialog(`Preview available at ${detail.url}. Open now?`, {
-        title: "Preview ready",
-        confirmLabel: "Open preview",
-        cancelLabel: "Later",
+
+      // Auto-switch to preview mode for new AI content
+      setCenterTab("preview")
+      toast.success("Preview updated", {
+        icon: '🚀',
+        duration: 3000,
+        position: 'bottom-center'
      })
-      if (confirmed) {
-        setCenterTab("preview")
-      }
    }
    window.addEventListener(BUILD_PREVIEW_EVENT, handler)
    onCleanup(() => window.removeEventListener(BUILD_PREVIEW_EVENT, handler))
  })

+  onMount(() => {
+    if (typeof window === "undefined") return
+    const handler = async (event: Event) => {
+      const detail = (event as CustomEvent<{ code: string; fileName: string | null; instanceId: string }>).detail
+      if (!detail || detail.instanceId !== props.instance.id) return
+
+      if (detail.fileName) {
+        const origin = window.location.origin
+        const apiOrigin = origin.includes(":3000") ? origin.replace(":3000", ":9898") : origin
+        const url = `${apiOrigin}/api/workspaces/${props.instance.id}/serve/${detail.fileName}`
+        setPreviewUrl(url)
+      } else {
+        const blob = new Blob([detail.code], { type: 'text/html' })
+        const url = URL.createObjectURL(blob)
+        setPreviewUrl(url)
+      }
+
+      setCenterTab("preview")
+      toast.success("Previewing code block", {
+        icon: '🔍',
+        duration: 2000,
+        position: 'bottom-center'
+      })
+    }
+    window.addEventListener("MANUAL_PREVIEW_EVENT", handler)
+    onCleanup(() => window.removeEventListener("MANUAL_PREVIEW_EVENT", handler))
+  })
+
  createEffect(() => {
    if (typeof window === "undefined") return
    window.localStorage.setItem(LEFT_DRAWER_STORAGE_KEY, sessionSidebarWidth().toString())
@@ -402,23 +455,90 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {
    showCommandPalette(props.instance.id)
  }

+  /* Compact Logic */
+  const [isCompacting, setIsCompacting] = createSignal(false);
+  const handleCompact = async () => {
+    const sessionId = activeSessionIdForInstance();
+    if (!sessionId || sessionId === "info" || isCompacting()) return;
+
+    setIsCompacting(true);
+    const toastId = toast.loading("Compacting...", { icon: <FileArchive class="animate-pulse text-indigo-400" /> });
+    try {
+      await compactSession(props.instance.id, sessionId);
+      toast.success("Session compacted!", { id: toastId });
+    } catch (e) {
+      toast.error("Failed to compact", { id: toastId });
+    } finally {
+      setIsCompacting(false);
+    }
+  }
+
  const [isFixing, setIsFixing] = createSignal(false)
  const [isBuilding, setIsBuilding] = createSignal(false)

  const handleSmartFix = async () => {
-    const sessionId = activeSessionIdForInstance()
-    if (!sessionId || sessionId === "info" || isFixing()) {
+    const parentSessionId = activeSessionIdForInstance()
+    if (!parentSessionId || parentSessionId === "info" || isFixing()) {
      return
    }

    setIsFixing(true)
+    const toastId = toast.loading("Smart Fix: Creating analysis task...", {
+      icon: <Sparkles class="text-indigo-400 animate-spin" />
+    });
    try {
-      // Smart Fix targets the active task if available, otherwise general fix
-      const session = activeSessionForInstance()
-      const activeTaskId = session?.activeTaskId
-      const args = activeTaskId ? `task:${activeTaskId}` : ""
+      // ALWAYS create a dedicated "Smart Fix" task in the MultiX pipeline
+      // This ensures the analysis and fixes appear in their own tab
+      const timestamp = new Date().toLocaleTimeString('en-US', { hour: '2-digit', minute: '2-digit' })
+      const taskResult = await addTask(
+        props.instance.id,
+        parentSessionId,
+        `🔧 Smart Fix ${timestamp}`
+      )

-      await executeCustomCommand(props.instance.id, sessionId, "fix", args)
+      const targetSessionId = taskResult.taskSessionId || parentSessionId
+      const taskId = taskResult.id
+
+      // Set this as the active task so the user sees it immediately
+      setActiveTask(props.instance.id, parentSessionId, taskId)
+
+      toast.loading("Analyzing project...", { id: toastId });
+
+      // Use sendMessage to force visible feedback in the chat stream
+      // Prompt enforces: Report → Plan → Approval → Execute workflow
+      const smartFixPrompt = `**Smart Fix Analysis Request**
+
+Please analyze this project for errors, bugs, warnings, or potential improvements.
+
+**Your response MUST follow this exact format:**
+
+1. **ANALYSIS RESULTS:**
+   - If NO errors/issues found: Clearly state "✅ No errors or issues detected in the project."
+   - If errors/issues ARE found: List each issue with file path and line number if applicable.
+
+2. **FIX PLAN (only if issues found):**
+   For each issue, outline:
+   - What the problem is
+   - How you will fix it
+   - Which files will be modified
+
+3. **AWAIT APPROVAL:**
+   After presenting the plan, explicitly ask: "Do you approve this fix plan? Reply 'yes' to proceed, or provide feedback for adjustments."
+
+4. **EXECUTION (only after I say 'yes'):**
+   Only apply fixes after receiving explicit approval. Use write_file tool to make changes.
+
+Now analyze the project and report your findings.`
+
+      await sendMessage(
+        props.instance.id,
+        targetSessionId,
+        smartFixPrompt,
+        [],
+        taskId
+      )
+
+      toast.success("Smart Fix task created. Check the pipeline.", { id: toastId, duration: 3000 });

      // Auto-open right panel to show agent progress if it's not open
      if (!rightOpen()) {
@@ -427,6 +547,7 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {
      }
    } catch (error) {
      log.error("Failed to run Smart Fix command", error)
+      toast.error("Smart Fix failed to start", { id: toastId });
    } finally {
      setTimeout(() => setIsFixing(false), 2000) // Reset after delay
    }
@@ -1180,7 +1301,7 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {

  const sessionLayout = (
    <div
-      class="session-shell-panels flex flex-col flex-1 min-h-0 overflow-x-hidden relative bg-[#050505]"
+      class="session-shell-panels flex flex-col flex-1 min-h-0 w-full overflow-hidden relative bg-[#050505]"
      ref={(element) => {
        setDrawerHost(element)
        measureDrawerHost()
@@ -1190,8 +1311,8 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {
      <div class="absolute top-[-10%] left-[-10%] w-[40%] h-[40%] bg-blue-600/10 blur-[120px] rounded-full pointer-events-none z-0" />
      <div class="absolute bottom-[-10%] right-[-10%] w-[30%] h-[30%] bg-purple-600/5 blur-[100px] rounded-full pointer-events-none z-0" />

-      <AppBar position="sticky" color="default" elevation={0} class="border-b border-white/5 bg-[#050505]/80 backdrop-blur-md z-20">
-        <Toolbar variant="dense" class="session-toolbar flex flex-wrap items-center justify-between gap-2 py-0 min-h-[40px]">
+      <AppBar position="sticky" color="default" elevation={0} class="border-b border-white/5 bg-[#050505]/80 backdrop-blur-md z-20 shrink-0">
+        <Toolbar variant="dense" class="session-toolbar flex items-center justify-between gap-2 py-0 min-h-[48px]">
          <div class="flex items-center space-x-4">
            <IconButton
              ref={setLeftToggleButtonEl}
@@ -1221,6 +1342,19 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {
          <div class="flex items-center space-x-4">
            <Show when={activeSessionIdForInstance() && activeSessionIdForInstance() !== "info"}>
              <div class="flex items-center space-x-2">
+                {/* Compact Button */}
+                <button
+                  onClick={handleCompact}
+                  disabled={isCompacting()}
+                  class="flex items-center gap-1.5 px-2.5 py-1 text-[11px] font-semibold text-cyan-400 bg-cyan-500/10 border border-cyan-500/20 hover:bg-cyan-500/20 hover:border-cyan-500/40 transition-all rounded-full"
+                  title="Compact Context: Summarize conversation to save tokens"
+                >
+                  <Show when={isCompacting()} fallback={<FileArchive size={14} strokeWidth={2} />}>
+                    <Loader2 size={14} class="animate-spin" />
+                  </Show>
+                  <span>Compact</span>
+                </button>
+
                <ModelStatusSelector
                  instanceId={props.instance.id}
                  sessionId={activeSessionIdForInstance()!}
@@ -1246,14 +1380,10 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {
                onClick={handleSmartFix}
                disabled={isFixing()}
                title="Smart Fix: Automatically detect and fix issues in your code"
-                class={`transition-all flex items-center space-x-1.5 px-2 py-1 rounded-full hover:bg-white/10 ${isFixing() ? "text-blue-500" : "text-zinc-400 hover:text-white"}`}
+                class={`transition-all flex items-center space-x-1.5 px-3 py-1 rounded-full text-[10px] font-bold uppercase tracking-tight ${isFixing() ? "text-blue-500 smart-fix-highlight bg-blue-500/10" : "text-zinc-400 hover:text-white hover:bg-white/5"}`}
              >
-                <Show when={isFixing()} fallback={<Sparkles size={14} class="text-blue-400" />}>
-                  <Loader2 size={14} class="animate-spin text-blue-400" />
-                </Show>
-                <span class="text-[10px] font-bold uppercase tracking-tight">
-                  {isFixing() ? "FIXING..." : "SMART FIX"}
-                </span>
+                <Zap size={12} class={isFixing() ? "animate-bounce" : ""} />
+                <span>Fix</span>
              </button>
              <div class="w-px h-3 bg-white/10" />
              <button
@@ -1303,11 +1433,11 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {
                setRightOpen(newState)
                setIsSoloOpen(newState)
              }}
-              class={`flex items-center space-x-1.5 px-3 py-1 rounded-full text-[11px] font-bold transition-all ${(rightOpen() && isSoloOpen()) ? 'bg-blue-600/20 text-blue-400 border border-blue-500/30' : 'bg-white/5 text-zinc-400 border border-white/5'
+              class={`flex items-center space-x-1.5 px-3 py-1 rounded-full text-[10px] font-bold uppercase tracking-tight transition-all ${(rightOpen() && isSoloOpen()) ? 'bg-blue-600/20 text-blue-400 border border-blue-500/30' : 'bg-white/5 text-zinc-400 border border-white/5'
                }`}
            >
-              <span class={`w-1.5 h-1.5 bg-current rounded-full ${(rightOpen() && isSoloOpen()) ? 'animate-pulse' : ''}`} />
-              <span>MULTIX MODE</span>
+              <LayoutIcon size={12} />
+              <span>MultiX</span>
            </button>
            <IconButton
              ref={setRightToggleButtonEl}
@@ -1323,146 +1453,67 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {
        </Toolbar>
      </AppBar>

-      <Box sx={{ display: "flex", flex: 1, minHeight: 0, overflowX: "hidden", position: "relative", zIndex: 10 }}>
+      <Box sx={{ display: "flex", flex: 1, minHeight: 0, width: "100%", overflow: "hidden", position: "relative", zIndex: 10 }}>
        {renderLeftPanel()}

        <Box
-          component="main"
-          sx={{ flexGrow: 1, minHeight: 0, display: "flex", flexDirection: "column", overflowX: "hidden" }}
-          class="content-area relative"
+          component="div"
+          sx={{ flexGrow: 1, minHeight: 0, display: "flex", flexDirection: "column", overflow: "hidden" }}
+          class="content-area relative bg-[#050505]"
        >
-          <div class="flex-1 flex overflow-hidden min-h-0">
-            <Show when={!isPhoneLayout()}>
-              <div class="flex-1 flex flex-col min-h-0 bg-[#0d0d0d]">
-                <div class="h-10 glass border-b border-white/5 flex items-center justify-between px-4 shrink-0">
-                  <div class="flex items-center gap-2">
-                    <button
-                      type="button"
-                      class={`px-2.5 py-1 rounded-md text-[11px] font-semibold uppercase tracking-wide border ${centerTab() === "code"
-                          ? "bg-white/10 border-white/20 text-white"
-                          : "border-transparent text-zinc-400 hover:text-zinc-200 hover:bg-white/5"
-                        }`}
-                      onClick={() => setCenterTab("code")}
-                    >
-                      Code
-                    </button>
-                    <button
-                      type="button"
-                      class={`px-2.5 py-1 rounded-md text-[11px] font-semibold uppercase tracking-wide border ${centerTab() === "preview"
-                          ? "bg-white/10 border-white/20 text-white"
-                          : "border-transparent text-zinc-400 hover:text-zinc-200 hover:bg-white/5"
-                        }`}
-                      onClick={() => setCenterTab("preview")}
-                      disabled={!previewUrl()}
-                      title={previewUrl() || "Run build to enable preview"}
-                    >
-                      Preview
-                    </button>
-                  </div>
-                  <Show when={previewUrl()}>
-                    {(url) => (
-                      <div class="text-[10px] text-zinc-500 truncate max-w-[50%]" title={url()}>
-                        {url()}
-                      </div>
-                    )}
-                  </Show>
-                </div>
-                <Show when={centerTab() === "preview"} fallback={<Editor file={currentFile()} />}>
-                  <Show
-                    when={previewUrl()}
-                    fallback={
-                      <div class="flex-1 flex items-center justify-center text-zinc-500">
-                        <div class="text-center">
-                          <p>No preview available yet.</p>
-                          <p class="text-sm mt-2 opacity-60">Run build to detect a preview URL.</p>
-                        </div>
-                      </div>
-                    }
+          {/* Main workspace area */}
+          <div class="flex-1 flex flex-row min-h-0 w-full overflow-hidden">
+            {/* Center Area (Editor/Preview) */}
+            <div class="flex-1 flex flex-col min-h-0 bg-[#0d0d0d] overflow-hidden">
+              <div class="flex items-center justify-between px-4 py-2 border-b border-white/5 bg-[#111112]">
+                <div class="flex items-center space-x-4">
+                  <button
+                    onClick={() => setCenterTab("code")}
+                    class={`px-2.5 py-1 rounded-md text-[11px] font-semibold uppercase tracking-wide border ${centerTab() === "code"
+                      ? "bg-white/10 border-white/20 text-white"
+                      : "border-transparent text-zinc-400 hover:text-zinc-200 hover:bg-white/5"
+                      }`}
                  >
-                    {(url) => (
-                      <iframe
-                        class="flex-1 w-full h-full border-none bg-black"
-                        src={url()}
-                        title="App Preview"
-                        sandbox="allow-scripts allow-same-origin allow-forms allow-pointer-lock allow-popups"
-                      />
-                    )}
-                  </Show>
+                    Code
+                  </button>
+                  <button
+                    onClick={() => setCenterTab("preview")}
+                    class={`px-2.5 py-1 rounded-md text-[11px] font-semibold uppercase tracking-wide border ${centerTab() === "preview"
+                      ? "bg-white/10 border-white/20 text-white"
+                      : "border-transparent text-zinc-400 hover:text-zinc-200 hover:bg-white/5"
+                      }`}
+                  >
+                    Preview
+                  </button>
+                </div>
+                <Show when={previewUrl()}>
+                  {(url) => (
+                    <div class="text-[10px] text-zinc-500 truncate max-w-[50%]" title={url()}>
+                      {url()}
+                    </div>
+                  )}
                </Show>
              </div>
-            </Show>

-            <div
-              class="flex flex-col relative border-l border-white/5 min-h-0 overflow-hidden min-w-0"
-              style={{
-                width: isPhoneLayout() ? "100%" : `${chatPanelWidth()}px`,
-                "flex-shrink": isPhoneLayout() ? 1 : 0,
-              }}
-            >
-              <div
-                class="absolute -left-1 top-0 bottom-0 w-2 cursor-col-resize z-20 hover:bg-white/5 active:bg-white/10 transition-colors"
-                onMouseDown={handleResizeMouseDown("chat")}
-              />
-              <Show when={isSoloOpen()}>
-                <div class="flex-1 flex flex-col min-h-0 relative">
-                  <MultiTaskChat instanceId={props.instance.id} sessionId={activeSessionIdForInstance() || ""} />
+              <Show when={centerTab() === "preview"} fallback={<Editor file={currentFile()} />}>
+                <div class="flex-1 min-h-0 bg-white">
+                  <iframe
+                    src={previewUrl() || "about:blank"}
+                    class="w-full h-full border-none"
+                    title="Preview"
+                  />
                </div>
              </Show>
-
-              <div class="flex-1 flex flex-col relative min-h-0"
-                style={{ display: isSoloOpen() ? "none" : "flex" }}>
-                <Show
-                  when={showingInfoView()}
-                  fallback={
-                    <Show
-                      when={cachedSessionIds().length > 0 && activeSessionIdForInstance()}
-                      fallback={
-                        <div class="flex items-center justify-center h-full">
-                          <div class="text-center text-zinc-500">
-                            <p class="mb-2">No session selected</p>
-                            <p class="text-sm">Select a session to view messages</p>
-                          </div>
-                        </div>
-                      }
-                    >
-                      <For each={cachedSessionIds()}>
-                        {(sessionId) => {
-                          const isActive = () => activeSessionIdForInstance() === sessionId
-                          return (
-                            <div
-                              class="session-cache-pane flex flex-col flex-1 min-h-0"
-                              style={{ display: isActive() ? "flex" : "none" }}
-                              data-session-id={sessionId}
-                              aria-hidden={!isActive()}
-                            >
-                              <SessionView
-                                sessionId={sessionId}
-                                activeSessions={activeSessions()}
-                                instanceId={props.instance.id}
-                                instanceFolder={props.instance.folder}
-                                escapeInDebounce={props.escapeInDebounce}
-                                showSidebarToggle={showEmbeddedSidebarToggle()}
-                                onSidebarToggle={() => setLeftOpen(true)}
-                                forceCompactStatusLayout={showEmbeddedSidebarToggle()}
-                                isActive={isActive()}
-                              />
-                            </div>
-                          )
-                        }}
-                      </For>
-                    </Show>
-                  }
-                >
-                  <div class="info-view-pane flex flex-col flex-1 min-h-0 overflow-y-auto">
-                    <InfoView instanceId={props.instance.id} />
-                  </div>
-                </Show>
-              </div>
            </div>
+
+            {/* Right Panel (MultiX Chat) */}
+            <Show when={rightOpen() && isSoloOpen()}>
+              <div class="flex flex-col relative border-l border-white/5 min-h-0 overflow-hidden" style={{ width: `${chatPanelWidth()}px`, "flex-shrink": 0 }}>
+                <MultiTaskChat instanceId={props.instance.id} sessionId={activeSessionIdForInstance()!} />
+              </div>
+            </Show>
          </div>

-          {/* Bottom Toolbar/Terminal Area */}
-          {/* Bottom Toolbar/Terminal Area */}
          <div
            class="flex flex-col border-t border-white/5 relative bg-[#09090b] z-10 shrink-0 overflow-hidden"
            style={{
@@ -1502,23 +1553,12 @@ const InstanceShell2: Component<InstanceShellProps> = (props) => {
                  <span class="w-1.5 h-1.5 rounded-full bg-green-500 shadow-[0_0_5px_rgba(34,197,94,0.5)]" />
                  <span>Sync Active</span>
                </div>
-                <Show when={activeSessionForInstance()}>
-                  {(session) => (
-                    <>
-                      <span class="hover:text-zinc-300 cursor-pointer">{session().model.modelId}</span>
-                      <span class="hover:text-zinc-300 cursor-pointer">{session().agent}</span>
-                    </>
-                  )}
-                </Show>
              </div>
            </footer>
          </div>
        </Box>
-
        {renderRightPanel()}
      </Box>
-
-      {/* Floating Action Buttons removed - Integrated into Header */}
    </div>
  )

--- a/packages/ui/src/components/instance/sidebar.tsx
+++ b/packages/ui/src/components/instance/sidebar.tsx
@@ -173,6 +173,11 @@ export const Sidebar: Component<SidebarProps> = (props) => {
    if (typeof window === "undefined") return
    const handler = (event: Event) => {
      const detail = (event as CustomEvent<{ instanceId?: string }>).detail
+      console.log(`[Sidebar] Received FILE_CHANGE_EVENT`, {
+        detail,
+        currentInstanceId: props.instanceId,
+        match: detail?.instanceId === props.instanceId
+      });
      if (!detail || detail.instanceId !== props.instanceId) return
      void refreshRootFiles()
    }
@@ -316,18 +321,18 @@ export const Sidebar: Component<SidebarProps> = (props) => {
            </For>
          </Show>
          <Show when={activeTab() === "sessions"}>
-             <div class="flex flex-col gap-1">
-                <For each={props.sessions}>
-                  {(session) => (
-                    <div 
-                      onClick={() => props.onSessionSelect(session.id)}
-                      class={`px-3 py-1.5 rounded cursor-pointer text-sm transition-colors ${props.activeSessionId === session.id ? 'bg-blue-600/20 text-blue-400 border border-blue-500/20' : 'text-zinc-400 hover:bg-white/5'}`}
-                    >
-                      {session.title || session.id.slice(0, 8)}
-                    </div>
-                  )}
-                </For>
-             </div>
+            <div class="flex flex-col gap-1">
+              <For each={props.sessions}>
+                {(session) => (
+                  <div
+                    onClick={() => props.onSessionSelect(session.id)}
+                    class={`px-3 py-1.5 rounded cursor-pointer text-sm transition-colors ${props.activeSessionId === session.id ? 'bg-blue-600/20 text-blue-400 border border-blue-500/20' : 'text-zinc-400 hover:bg-white/5'}`}
+                  >
+                    {session.title || session.id.slice(0, 8)}
+                  </div>
+                )}
+              </For>
+            </div>
          </Show>
          <Show when={activeTab() === "search"}>
            <div class="flex flex-col gap-3">
@@ -473,11 +478,10 @@ export const Sidebar: Component<SidebarProps> = (props) => {
                      <button
                        type="button"
                        onClick={() => toggleSkillSelection(skill.id)}
-                        class={`w-full text-left px-3 py-2 rounded-md border transition-colors ${
-                          isSelected()
+                        class={`w-full text-left px-3 py-2 rounded-md border transition-colors ${isSelected()
                            ? "border-blue-500/60 bg-blue-500/10 text-blue-200"
                            : "border-white/10 bg-white/5 text-zinc-300 hover:text-white"
-                        }`}
+                          }`}
                      >
                        <div class="text-xs font-semibold">{skill.name}</div>
                        <Show when={skill.description}>
--- a/packages/ui/src/components/markdown.tsx
+++ b/packages/ui/src/components/markdown.tsx
@@ -1,4 +1,5 @@
 import { createEffect, createSignal, onMount, onCleanup } from "solid-js"
+import { addDebugLog } from "./debug-overlay"
 import { renderMarkdown, onLanguagesLoaded, initMarkdown, decodeHtmlEntities } from "../lib/markdown"
 import type { TextPart, RenderCache } from "../types/message"
 import { getLogger } from "../lib/logger"
@@ -16,6 +17,7 @@ interface MarkdownProps {
  size?: "base" | "sm" | "tight"
  disableHighlight?: boolean
  onRendered?: () => void
+  instanceId: string
 }

 export function Markdown(props: MarkdownProps) {
@@ -27,7 +29,7 @@ export function Markdown(props: MarkdownProps) {
    Promise.resolve().then(() => props.onRendered?.())
  }

-  createEffect(async () => {
+  createEffect(() => {
    const part = props.part
    const rawText = typeof part.text === "string" ? part.text : ""
    const text = decodeHtmlEntities(rawText)
@@ -39,6 +41,7 @@ export function Markdown(props: MarkdownProps) {

    latestRequestedText = text

+    // 1. Check Synchronous Local Cache
    const localCache = part.renderCache
    if (localCache && localCache.text === text && localCache.theme === themeKey) {
      setHtml(localCache.html)
@@ -46,6 +49,7 @@ export function Markdown(props: MarkdownProps) {
      return
    }

+    // 2. Check Global Cache
    const globalCache = markdownRenderCache.get(cacheKey)
    if (globalCache && globalCache.text === text) {
      setHtml(globalCache.html)
@@ -54,11 +58,13 @@ export function Markdown(props: MarkdownProps) {
      return
    }

-    if (!highlightEnabled) {
-      part.renderCache = undefined
+    // 3. Throttle/Debounce Rendering for new content
+    // We delay the expensive async render to avoid choking the main thread during rapid streaming
+    const performRender = async () => {
+      if (latestRequestedText !== text) return // Stale

      try {
-        const rendered = await renderMarkdown(text, { suppressHighlight: true })
+        const rendered = await renderMarkdown(text, { suppressHighlight: !highlightEnabled })

        if (latestRequestedText === text) {
          const cacheEntry: RenderCache = { text, html: rendered, theme: themeKey }
@@ -70,36 +76,18 @@ export function Markdown(props: MarkdownProps) {
      } catch (error) {
        log.error("Failed to render markdown:", error)
        if (latestRequestedText === text) {
-          const cacheEntry: RenderCache = { text, html: text, theme: themeKey }
-          setHtml(text)
-          part.renderCache = cacheEntry
-          markdownRenderCache.set(cacheKey, cacheEntry)
-          notifyRendered()
+          setHtml(text) // Fallback
        }
      }
-      return
    }

-    try {
-      const rendered = await renderMarkdown(text)
+    // Heuristic: If text length matches cache length + small amount, it's streaming.
+    // We can debounce. If it's a huge jump (initial load), render immediately.
+    // For now, always debounce slightly to unblock main thread.
+    // Using 200ms (was 50ms) for less frequent but smoother updates
+    const timerId = setTimeout(performRender, 200)

-      if (latestRequestedText === text) {
-        const cacheEntry: RenderCache = { text, html: rendered, theme: themeKey }
-        setHtml(rendered)
-        part.renderCache = cacheEntry
-        markdownRenderCache.set(cacheKey, cacheEntry)
-        notifyRendered()
-      }
-    } catch (error) {
-      log.error("Failed to render markdown:", error)
-      if (latestRequestedText === text) {
-        const cacheEntry: RenderCache = { text, html: text, theme: themeKey }
-        setHtml(text)
-        part.renderCache = cacheEntry
-        markdownRenderCache.set(cacheKey, cacheEntry)
-        notifyRendered()
-      }
-    }
+    onCleanup(() => clearTimeout(timerId))
  })

  onMount(() => {
@@ -121,6 +109,31 @@ export function Markdown(props: MarkdownProps) {
            }, 2000)
          }
        }
+        return
+      }
+
+      const previewButton = target.closest(".code-block-preview") as HTMLButtonElement
+      if (previewButton) {
+        e.preventDefault()
+        const code = previewButton.getAttribute("data-code")
+        const lang = previewButton.getAttribute("data-lang")
+
+        if (code && lang === "html") {
+          const decodedCode = decodeURIComponent(code)
+
+          // Try to find a filename in the text part
+          const contentText = props.part.text || ""
+          const fileMatch = contentText.match(/(\w+\.html)/)
+          const fileName = fileMatch ? fileMatch[1] : null
+
+          window.dispatchEvent(new CustomEvent("MANUAL_PREVIEW_EVENT", {
+            detail: {
+              code: decodedCode,
+              fileName: fileName,
+              instanceId: props.instanceId
+            }
+          }))
+        }
      }
    }

--- a/packages/ui/src/components/mcp-manager.tsx
+++ b/packages/ui/src/components/mcp-manager.tsx
@@ -3,7 +3,6 @@ import { ChevronDown, ExternalLink, Plus, RefreshCw, Search, Settings } from "lu
 import { Component, For, Show, createEffect, createMemo, createSignal } from "solid-js"
 import { serverApi } from "../lib/api-client"
 import { getLogger } from "../lib/logger"
-import InstanceServiceStatus from "./instance-service-status"
 import { useOptionalInstanceMetadataContext } from "../lib/contexts/instance-metadata-context"

 type McpServerConfig = {
@@ -110,6 +109,9 @@ const McpManager: Component<McpManagerProps> = (props) => {
  const [serverName, setServerName] = createSignal("")
  const [serverJson, setServerJson] = createSignal("")
  const [saving, setSaving] = createSignal(false)
+  const [connectionStatus, setConnectionStatus] = createSignal<Record<string, { connected: boolean }>>({})
+  const [toolCount, setToolCount] = createSignal(0)
+  const [connecting, setConnecting] = createSignal(false)

  const metadataContext = useOptionalInstanceMetadataContext()
  const metadata = createMemo(() => metadataContext?.metadata?.() ?? null)
@@ -138,6 +140,38 @@ const McpManager: Component<McpManagerProps> = (props) => {
    } finally {
      setIsLoading(false)
    }
+    // Fetch connection status separately (non-blocking)
+    loadConnectionStatus().catch(() => { })
+  }
+
+  const loadConnectionStatus = async () => {
+    try {
+      const controller = new AbortController()
+      const timeoutId = setTimeout(() => controller.abort(), 5000)
+      const status = await serverApi.fetchWorkspaceMcpStatus(props.instanceId)
+      clearTimeout(timeoutId)
+      setConnectionStatus(status.servers ?? {})
+      setToolCount(status.toolCount ?? 0)
+    } catch (err) {
+      log.warn("Failed to fetch MCP status", err)
+      // Don't block UI on status failures
+    }
+  }
+
+  const connectAllMcps = async () => {
+    if (connecting()) return
+    setConnecting(true)
+    setError(null)
+    try {
+      const result = await serverApi.connectWorkspaceMcps(props.instanceId)
+      setConnectionStatus(result.servers ?? {})
+      setToolCount(result.toolCount ?? 0)
+    } catch (err) {
+      log.error("Failed to connect MCPs", err)
+      setError("Failed to connect MCP servers.")
+    } finally {
+      setConnecting(false)
+    }
  }

  createEffect(() => {
@@ -204,6 +238,8 @@ const McpManager: Component<McpManagerProps> = (props) => {
      nextConfig.mcpServers = mcpServers
      setConfig(nextConfig)
      await serverApi.updateWorkspaceMcpConfig(props.instanceId, nextConfig)
+      // Auto-connect after installing
+      await loadConnectionStatus()
    } catch (err) {
      const message = err instanceof Error ? err.message : "Failed to install MCP server."
      setError(message)
@@ -325,44 +361,56 @@ const McpManager: Component<McpManagerProps> = (props) => {
        {(err) => <div class="text-[11px] text-amber-400">{err()}</div>}
      </Show>

+      <Show when={toolCount() > 0}>
+        <div class="text-[11px] text-green-400 mb-2">
+          ✓ {toolCount()} MCP tools available
+        </div>
+      </Show>
+
      <Show
        when={!isLoading() && servers().length > 0}
        fallback={<div class="text-[11px] text-zinc-500 italic">{isLoading() ? "Loading MCP servers..." : "No MCP servers configured."}</div>}
      >
        <div class="mcp-server-list">
          <For each={servers()}>
-            {([name, server]) => (
-              <div class="mcp-server-card">
-                <div class="mcp-server-row">
-                  <div class="flex flex-col">
-                    <span class="text-xs font-semibold text-zinc-100">{name}</span>
-                    <span class="text-[11px] text-zinc-500 truncate">
-                      {server.command ? `${server.command} ${(server.args ?? []).join(" ")}` : "Custom config"}
-                    </span>
-                  </div>
-                  <div class="flex items-center gap-2">
-                    <Show when={mcpStatus()?.[name]?.status}>
-                      <span class="mcp-status-chip">
-                        {mcpStatus()?.[name]?.status}
+            {([name, server]) => {
+              const isConnected = () => connectionStatus()[name]?.connected ?? false
+              return (
+                <div class="mcp-server-card">
+                  <div class="mcp-server-row">
+                    <div class="flex flex-col">
+                      <span class="text-xs font-semibold text-zinc-100">{name}</span>
+                      <span class="text-[11px] text-zinc-500 truncate">
+                        {server.command ? `${server.command} ${(server.args ?? []).join(" ")}` : server.url || "Custom config"}
                      </span>
-                    </Show>
-                    <Show when={mcpStatus()?.[name]?.error}>
-                      <span class="mcp-status-error" title={String(mcpStatus()?.[name]?.error)}>
-                        error
-                      </span>
-                    </Show>
+                    </div>
+                    <div class="flex items-center gap-2">
+                      <Show when={isConnected()}>
+                        <span class="mcp-status-chip" style={{ background: "var(--status-ok, #22c55e)", color: "#fff" }}>
+                          connected
+                        </span>
+                      </Show>
+                      <Show when={!isConnected()}>
+                        <span class="mcp-status-chip" style={{ background: "var(--status-warning, #eab308)", color: "#000" }}>
+                          not connected
+                        </span>
+                      </Show>
+                    </div>
                  </div>
                </div>
-              </div>
-            )}
+              )
+            }}
          </For>
        </div>
+        <button
+          onClick={connectAllMcps}
+          disabled={connecting()}
+          class="mt-2 px-3 py-1.5 text-xs rounded-md bg-blue-500/20 border border-blue-500/40 text-blue-200 hover:text-white disabled:opacity-60 w-full"
+        >
+          {connecting() ? "Connecting..." : "Connect All MCPs"}
+        </button>
      </Show>

-      <div class="mt-3">
-        <InstanceServiceStatus sections={["mcp"]} />
-      </div>
-
      <Dialog open={showManual()} onOpenChange={setShowManual} modal>
        <Dialog.Portal>
          <Dialog.Overlay class="modal-overlay" />
--- a/packages/ui/src/components/message-block.tsx
+++ b/packages/ui/src/components/message-block.tsx
@@ -1,4 +1,5 @@
-import { For, Match, Show, Switch, createEffect, createMemo, createSignal } from "solid-js"
+import { For, Match, Show, Switch, createEffect, createMemo, createSignal, untrack } from "solid-js"
+import { addDebugLog } from "./debug-overlay"
 import MessageItem from "./message-item"
 import ToolCall from "./tool-call"
 import type { InstanceMessageStore } from "../stores/message-v2/instance-store"
@@ -215,14 +216,30 @@ interface MessageBlockProps {
 }

 export default function MessageBlock(props: MessageBlockProps) {
-  const record = createMemo(() => props.store().getMessage(props.messageId))
-  const messageInfo = createMemo(() => props.store().getMessageInfo(props.messageId))
+  // CRITICAL FIX: Use untrack for store access to prevent cascading updates during streaming
+  // The component will still re-render when needed via the Index component in MessageBlockList
+  const record = createMemo(() => {
+    // Only create reactive dependency on message ID, not content
+    const id = props.messageId;
+    return untrack(() => props.store().getMessage(id));
+  })
+  const messageInfo = createMemo(() => {
+    const id = props.messageId;
+    return untrack(() => props.store().getMessageInfo(id));
+  })
  const sessionCache = getSessionRenderCache(props.instanceId, props.sessionId)

+  // CRITICAL: Use a throttled revision check to avoid re-computing on every streaming chunk
+  const [lastProcessedRevision, setLastProcessedRevision] = createSignal(0);
+
  const block = createMemo<MessageDisplayBlock | null>(() => {
    const current = record()
    if (!current) return null

+    // OPTIMIZATION: Skip cache during streaming (revision changes too fast)
+    // Just return a basic block structure that will be updated when streaming completes
+    const isStreaming = current.status === "streaming" || current.status === "sending";
+
    const index = props.messageIndex
    const lastAssistantIdx = props.lastAssistantIndex()
    const isQueued = current.role === "user" && (lastAssistantIdx === -1 || index > lastAssistantIdx)
@@ -236,9 +253,11 @@ export default function MessageBlock(props: MessageBlockProps) {
          : infoTime.created ?? 0
    const infoError = (info as { error?: { name?: string } } | undefined)?.error
    const infoErrorName = typeof infoError?.name === "string" ? infoError.name : ""
+
+    // Skip revision in cache signature during streaming
    const cacheSignature = [
      current.id,
-      current.revision,
+      isStreaming ? "streaming" : current.revision,
      isQueued ? 1 : 0,
      props.showThinking() ? 1 : 0,
      props.thinkingDefaultExpanded() ? 1 : 0,
@@ -270,25 +289,23 @@ export default function MessageBlock(props: MessageBlockProps) {
        current.role === "assistant" &&
        !agentMetaAttached &&
        pendingParts.some((part) => partHasRenderableText(part))
-      let cached = sessionCache.messageItems.get(segmentKey)
-      if (!cached) {
-        cached = {
-          type: "content",
-          key: segmentKey,
-          record: current,
-          parts: pendingParts.slice(),
-          messageInfo: info,
-          isQueued,
-          showAgentMeta: shouldShowAgentMeta,
-        }
-        sessionCache.messageItems.set(segmentKey, cached)
-      } else {
-        cached.record = current
-        cached.parts = pendingParts.slice()
-        cached.messageInfo = info
-        cached.isQueued = isQueued
-        cached.showAgentMeta = shouldShowAgentMeta
+
+      // Always create a fresh object to ensure granular reactivity in <For>
+      // when we remove 'keyed' from <Show>. If we mutated properties
+      // on an existing object, <For> would assume identity match and skip updates.
+      const cached: ContentDisplayItem = {
+        type: "content",
+        key: segmentKey,
+        record: current,
+        parts: pendingParts.slice(),
+        messageInfo: info,
+        isQueued,
+        showAgentMeta: shouldShowAgentMeta,
      }
+
+      // Update cache with the new version (for potential stability elsewhere, though less critical now)
+      sessionCache.messageItems.set(segmentKey, cached)
+
      if (shouldShowAgentMeta) {
        agentMetaAttached = true
      }
@@ -396,10 +413,10 @@ export default function MessageBlock(props: MessageBlockProps) {
  })

  return (
-    <Show when={block()} keyed>
+    <Show when={block()}>
      {(resolvedBlock) => (
-        <div class="message-stream-block" data-message-id={resolvedBlock.record.id}>
-          <For each={resolvedBlock.items}>
+        <div class="message-stream-block" data-message-id={resolvedBlock().record.id}>
+          <For each={resolvedBlock().items}>
            {(item) => (
              <Switch>
                <Match when={item.type === "content"}>
--- a/packages/ui/src/components/message-part.tsx
+++ b/packages/ui/src/components/message-part.tsx
@@ -14,8 +14,8 @@ interface MessagePartProps {
  instanceId: string
  sessionId: string
  onRendered?: () => void
- }
- export default function MessagePart(props: MessagePartProps) {
+}
+export default function MessagePart(props: MessagePartProps) {

  const { isDark } = useTheme()
  const { preferences } = useConfig()
@@ -97,16 +97,17 @@ interface MessagePartProps {
        <Show when={!(props.part.type === "text" && props.part.synthetic && isAssistantMessage()) && partHasRenderableText(props.part)}>
          <div class={textContainerClass()}>
            <Show
-               when={isAssistantMessage()}
-               fallback={<span>{plainTextContent()}</span>}
-             >
+              when={isAssistantMessage()}
+              fallback={<span>{plainTextContent()}</span>}
+            >
              <Markdown
                part={createTextPartForMarkdown()}
                isDark={isDark()}
                size={isAssistantMessage() ? "tight" : "base"}
                onRendered={props.onRendered}
+                instanceId={props.instanceId}
              />
-             </Show>
+            </Show>

          </div>
        </Show>
--- a/packages/ui/src/components/model-selector.tsx
+++ b/packages/ui/src/components/model-selector.tsx
@@ -70,27 +70,27 @@ export default function ModelSelector(props: ModelSelectorProps) {
    window.addEventListener("opencode-zen-offline-models", handleCustom as EventListener)
    window.addEventListener("storage", handleStorage)

-    // Poll Context-Engine status
-    const pollContextEngine = async () => {
-      try {
-        const response = await fetch("/api/context-engine/status")
-        if (response.ok) {
-          const data = await response.json() as { status: ContextEngineStatus }
-          setContextEngineStatus(data.status ?? "stopped")
-        } else {
-          setContextEngineStatus("stopped")
-        }
-      } catch {
-        setContextEngineStatus("stopped")
-      }
-    }
-    pollContextEngine()
-    const pollInterval = setInterval(pollContextEngine, 5000)
+    // DISABLED: Context-Engine polling was causing performance issues
+    // const pollContextEngine = async () => {
+    //   try {
+    //     const response = await fetch("/api/context-engine/status")
+    //     if (response.ok) {
+    //       const data = await response.json() as { status: ContextEngineStatus }
+    //       setContextEngineStatus(data.status ?? "stopped")
+    //     } else {
+    //       setContextEngineStatus("stopped")
+    //     }
+    //   } catch {
+    //     setContextEngineStatus("stopped")
+    //   }
+    // }
+    // pollContextEngine()
+    // const pollInterval = setInterval(pollContextEngine, 5000)

    onCleanup(() => {
      window.removeEventListener("opencode-zen-offline-models", handleCustom as EventListener)
      window.removeEventListener("storage", handleStorage)
-      clearInterval(pollInterval)
+      // clearInterval(pollInterval)
    })
  })

@@ -208,10 +208,10 @@ export default function ModelSelector(props: ModelSelectorProps) {
                  >
                    <span
                      class={`w-2 h-2 rounded-full ${contextEngineStatus() === "ready"
-                          ? "bg-emerald-500"
-                          : contextEngineStatus() === "indexing"
-                            ? "bg-blue-500 animate-pulse"
-                            : "bg-red-500"
+                        ? "bg-emerald-500"
+                        : contextEngineStatus() === "indexing"
+                          ? "bg-blue-500 animate-pulse"
+                          : "bg-red-500"
                        }`}
                    />
                    <Database class="w-3 h-3 text-zinc-400" />
--- a/packages/ui/src/components/opencode-binary-selector.tsx
+++ b/packages/ui/src/components/opencode-binary-selector.tsx
@@ -1,5 +1,5 @@
 import { Component, For, Show, createEffect, createMemo, createSignal, onCleanup } from "solid-js"
-import { FolderOpen, Trash2, Check, AlertCircle, Loader2, Plus } from "lucide-solid"
+import { FolderOpen, Trash2, Check, AlertCircle, Loader2, Plus, Sparkles } from "lucide-solid"
 import { useConfig } from "../stores/preferences"
 import { serverApi } from "../lib/api-client"
 import FileSystemBrowserDialog from "./filesystem-browser-dialog"
@@ -7,12 +7,15 @@ import { openNativeFileDialog, supportsNativeDialogs } from "../lib/native/nativ
 import { getLogger } from "../lib/logger"
 const log = getLogger("actions")

+// Special constant for Native mode (no OpenCode binary)
+const NATIVE_MODE_PATH = "__nomadarch_native__"

 interface BinaryOption {
  path: string
  version?: string
  lastUsed?: number
  isDefault?: boolean
+  isNative?: boolean
 }

 interface OpenCodeBinarySelectorProps {
@@ -44,10 +47,17 @@ const OpenCodeBinarySelector: Component<OpenCodeBinarySelectorProps> = (props) =

  const customBinaries = createMemo(() => binaries().filter((binary) => binary.path !== "opencode"))

-  const binaryOptions = createMemo<BinaryOption[]>(() => [{ path: "opencode", isDefault: true }, ...customBinaries()])
+  // Include NomadArch Native as the first option
+  const binaryOptions = createMemo<BinaryOption[]>(() => [
+    { path: NATIVE_MODE_PATH, isNative: true },
+    { path: "opencode", isDefault: true },
+    ...customBinaries()
+  ])

  const currentSelectionPath = () => props.selectedBinary || "opencode"

+  const isNativeMode = () => currentSelectionPath() === NATIVE_MODE_PATH
+
  createEffect(() => {
    if (!props.selectedBinary && lastUsedBinary()) {
      props.onBinaryChange(lastUsedBinary()!)
@@ -97,6 +107,11 @@ const OpenCodeBinarySelector: Component<OpenCodeBinarySelectorProps> = (props) =
  })

  async function validateBinary(path: string): Promise<{ valid: boolean; version?: string; error?: string }> {
+    // Native mode is always valid
+    if (path === NATIVE_MODE_PATH) {
+      return { valid: true, version: "Native" }
+    }
+
    if (versionInfo().has(path)) {
      const cachedVersion = versionInfo().get(path)
      return cachedVersion ? { valid: true, version: cachedVersion } : { valid: true }
@@ -209,6 +224,7 @@ const OpenCodeBinarySelector: Component<OpenCodeBinarySelectorProps> = (props) =
  }

  function getDisplayName(path: string): string {
+    if (path === NATIVE_MODE_PATH) return "🚀 NomadArch Native"
    if (path === "opencode") return "opencode (system PATH)"
    const parts = path.split(/[/\\]/)
    return parts[parts.length - 1] ?? path
@@ -277,18 +293,95 @@ const OpenCodeBinarySelector: Component<OpenCodeBinarySelectorProps> = (props) =
              </div>
            </div>
          </Show>
+
+          {/* Mode Comparison Info */}
+          <div class="rounded-lg border border-white/10 overflow-hidden">
+            <details class="group">
+              <summary class="flex items-center justify-between px-3 py-2 cursor-pointer bg-white/5 hover:bg-white/10 transition-colors">
+                <span class="text-xs font-medium text-muted">📊 Compare: Native vs SDK Mode</span>
+                <svg class="w-4 h-4 text-muted transition-transform group-open:rotate-180" fill="none" viewBox="0 0 24 24" stroke="currentColor">
+                  <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M19 9l-7 7-7-7" />
+                </svg>
+              </summary>
+              <div class="p-3 space-y-3 text-xs bg-black/20">
+                {/* Native Mode */}
+                <div class="space-y-1.5">
+                  <div class="flex items-center gap-2 text-emerald-400 font-medium">
+                    <Sparkles class="w-3.5 h-3.5" />
+                    <span>NomadArch Native (Recommended)</span>
+                  </div>
+                  <div class="pl-5 space-y-1 text-muted">
+                    <div class="flex items-center gap-2">
+                      <span class="text-emerald-400">✓</span>
+                      <span>No external binary required</span>
+                    </div>
+                    <div class="flex items-center gap-2">
+                      <span class="text-emerald-400">✓</span>
+                      <span>Free Zen models (GPT-5 Nano, Grok Code, GLM-4.7)</span>
+                    </div>
+                    <div class="flex items-center gap-2">
+                      <span class="text-emerald-400">✓</span>
+                      <span>Faster startup, simpler setup</span>
+                    </div>
+                    <div class="flex items-center gap-2">
+                      <span class="text-emerald-400">✓</span>
+                      <span>Full MCP tool support</span>
+                    </div>
+                    <div class="flex items-center gap-2">
+                      <span class="text-amber-400">○</span>
+                      <span>No LSP integration (coming soon)</span>
+                    </div>
+                  </div>
+                </div>
+
+                {/* SDK Mode */}
+                <div class="space-y-1.5 pt-2 border-t border-white/10">
+                  <div class="flex items-center gap-2 text-blue-400 font-medium">
+                    <Check class="w-3.5 h-3.5" />
+                    <span>OpenCode SDK Mode</span>
+                  </div>
+                  <div class="pl-5 space-y-1 text-muted">
+                    <div class="flex items-center gap-2">
+                      <span class="text-blue-400">✓</span>
+                      <span>Full LSP integration</span>
+                    </div>
+                    <div class="flex items-center gap-2">
+                      <span class="text-blue-400">✓</span>
+                      <span>All OpenCode features</span>
+                    </div>
+                    <div class="flex items-center gap-2">
+                      <span class="text-blue-400">✓</span>
+                      <span>More provider options</span>
+                    </div>
+                    <div class="flex items-center gap-2">
+                      <span class="text-amber-400">○</span>
+                      <span>Requires binary download</span>
+                    </div>
+                    <div class="flex items-center gap-2">
+                      <span class="text-amber-400">○</span>
+                      <span>Platform-specific binaries</span>
+                    </div>
+                  </div>
+                </div>
+              </div>
+            </details>
+          </div>
        </div>

        <div class="panel-list panel-list--fill max-h-80 overflow-y-auto">
          <For each={binaryOptions()}>
            {(binary) => {
              const isDefault = binary.isDefault
+              const isNative = binary.isNative
              const versionLabel = () => versionInfo().get(binary.path) ?? binary.version

              return (
                <div
                  class="panel-list-item flex items-center"
-                  classList={{ "panel-list-item-highlight": currentSelectionPath() === binary.path }}
+                  classList={{
+                    "panel-list-item-highlight": currentSelectionPath() === binary.path,
+                    "bg-gradient-to-r from-emerald-500/10 to-cyan-500/10 border-l-2 border-emerald-500": isNative && currentSelectionPath() === binary.path,
+                  }}
                >
                  <button
                    type="button"
@@ -298,31 +391,48 @@ const OpenCodeBinarySelector: Component<OpenCodeBinarySelectorProps> = (props) =
                  >
                    <div class="flex flex-col flex-1 min-w-0 gap-1.5">
                      <div class="flex items-center gap-2">
-                        <Check
-                          class={`w-4 h-4 transition-opacity ${currentSelectionPath() === binary.path ? "opacity-100" : "opacity-0"}`}
-                        />
-                        <span class="text-sm font-medium truncate text-primary">{getDisplayName(binary.path)}</span>
+                        <Show when={isNative}>
+                          <Sparkles
+                            class={`w-4 h-4 transition-opacity ${currentSelectionPath() === binary.path ? "text-emerald-400" : "text-muted"}`}
+                          />
+                        </Show>
+                        <Show when={!isNative}>
+                          <Check
+                            class={`w-4 h-4 transition-opacity ${currentSelectionPath() === binary.path ? "opacity-100" : "opacity-0"}`}
+                          />
+                        </Show>
+                        <span class={`text-sm font-medium truncate ${isNative ? "text-emerald-400" : "text-primary"}`}>
+                          {getDisplayName(binary.path)}
+                        </span>
+                        <Show when={isNative}>
+                          <span class="text-[10px] px-1.5 py-0.5 rounded bg-emerald-500/20 text-emerald-400 font-medium">
+                            RECOMMENDED
+                          </span>
+                        </Show>
                      </div>
-                      <Show when={!isDefault}>
+                      <Show when={!isDefault && !isNative}>
                        <div class="text-xs font-mono truncate pl-6 text-muted">{binary.path}</div>
                      </Show>
                      <div class="flex items-center gap-2 text-xs text-muted pl-6 flex-wrap">
-                        <Show when={versionLabel()}>
+                        <Show when={versionLabel() && !isNative}>
                          <span class="selector-badge-version">v{versionLabel()}</span>
                        </Show>
                        <Show when={isPathValidating(binary.path)}>
                          <span class="selector-badge-time">Checking…</span>
                        </Show>
-                        <Show when={!isDefault && binary.lastUsed}>
+                        <Show when={!isDefault && !isNative && binary.lastUsed}>
                          <span class="selector-badge-time">{formatRelativeTime(binary.lastUsed)}</span>
                        </Show>
                        <Show when={isDefault}>
                          <span class="selector-badge-time">Use binary from system PATH</span>
                        </Show>
+                        <Show when={isNative}>
+                          <span class="text-emerald-400/70">No OpenCode binary needed • Free Zen models included</span>
+                        </Show>
                      </div>
                    </div>
                  </button>
-                  <Show when={!isDefault}>
+                  <Show when={!isDefault && !isNative}>
                    <button
                      type="button"
                      class="p-2 text-muted hover:text-primary"
@@ -352,5 +462,7 @@ const OpenCodeBinarySelector: Component<OpenCodeBinarySelectorProps> = (props) =
  )
 }

-export default OpenCodeBinarySelector
+// Export the native mode constant for use elsewhere
+export const NOMADARCH_NATIVE_MODE = NATIVE_MODE_PATH

+export default OpenCodeBinarySelector
--- a/packages/ui/src/components/remote-access-overlay.tsx
+++ b/packages/ui/src/components/remote-access-overlay.tsx
@@ -125,8 +125,8 @@ export function RemoteAccessOverlay(props: RemoteAccessOverlayProps) {
            <header class="remote-header">
              <div>
                <p class="remote-eyebrow">Remote handover</p>
-                <h2 class="remote-title">Connect to CodeNomad remotely</h2>
-                <p class="remote-subtitle">Use the addresses below to open CodeNomad from another device.</p>
+                <h2 class="remote-title">Connect to NomadArch remotely</h2>
+                <p class="remote-subtitle">Use the addresses below to open NomadArch from another device.</p>
              </div>
              <button type="button" class="remote-close" onClick={props.onClose} aria-label="Close remote access">
                ×
--- a/packages/ui/src/components/settings/ZAISettings.tsx
+++ b/packages/ui/src/components/settings/ZAISettings.tsx
@@ -186,7 +186,7 @@ const ZAISettings: Component = () => {
                    <label class="block font-medium mb-2">Endpoint</label>
                    <input
                        type="text"
-                        placeholder="https://api.z.ai/api/paas/v4"
+                        placeholder="https://api.z.ai/api/coding/paas/v4"
                        value={config().endpoint || ''}
                        onChange={(e) => handleConfigChange('endpoint', e.target.value)}
                        class="w-full px-3 py-2 border border-gray-300 dark:border-gray-600 rounded-md focus:outline-none focus:ring-2 focus:ring-blue-500 bg-white dark:bg-gray-800"
--- a/packages/ui/src/index.css
+++ b/packages/ui/src/index.css
@@ -17,10 +17,9 @@
  padding: 0;
 }

-
-
 html,
-body {
+body,
+#root {
  font-family: var(--font-family-sans);
  font-size: var(--font-size-base);
  line-height: var(--line-height-normal);
@@ -29,45 +28,11 @@ body {
  -moz-osx-font-smoothing: grayscale;
  background-color: var(--surface-base);
  color: var(--text-primary);
-  margin: 0;
-  padding: 0;
-  overflow: hidden;
-  width: 100%;
-  height: 100%;
+  margin: 0 !important;
+  padding: 0 !important;
+  overflow: hidden !important;
+  width: 100vw !important;
+  height: 100vh !important;
+  display: flex !important;
+  flex-direction: column !important;
 }
-
-#root {
-  width: 100%;
-  height: 100%;
-  background-color: var(--surface-base);
-  overflow: hidden;
-}
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
--- a/packages/ui/src/lib/api-client.ts
+++ b/packages/ui/src/lib/api-client.ts
@@ -33,7 +33,7 @@ const FALLBACK_API_BASE = "http://127.0.0.1:9898"
 const RUNTIME_BASE = typeof window !== "undefined" ? window.location?.origin : undefined
 const DEFAULT_BASE = typeof window !== "undefined"
  ? (window.__CODENOMAD_API_BASE__ ??
-     (window.location?.protocol === "file:" ? FALLBACK_API_BASE : (RUNTIME_BASE === "null" || !RUNTIME_BASE || RUNTIME_BASE.startsWith("file:") ? FALLBACK_API_BASE : RUNTIME_BASE)))
+    (window.location?.protocol === "file:" ? FALLBACK_API_BASE : (RUNTIME_BASE === "null" || !RUNTIME_BASE || RUNTIME_BASE.startsWith("file:") ? FALLBACK_API_BASE : RUNTIME_BASE)))
  : FALLBACK_API_BASE
 const API_BASE = import.meta.env.VITE_CODENOMAD_API_BASE ?? DEFAULT_BASE

@@ -117,6 +117,9 @@ async function request<T>(path: string, init?: RequestInit): Promise<T> {


 export const serverApi = {
+  getApiBase(): string {
+    return API_BASE_ORIGIN
+  },
  fetchWorkspaces(): Promise<WorkspaceDescriptor[]> {
    return request<WorkspaceDescriptor[]>("/api/workspaces")
  },
@@ -186,6 +189,20 @@ export const serverApi = {
      body: JSON.stringify({ config }),
    })
  },
+  fetchWorkspaceMcpStatus(id: string): Promise<{
+    servers: Record<string, { connected: boolean }>
+    toolCount: number
+    tools: Array<{ name: string; server: string; description: string }>
+  }> {
+    return request(`/api/workspaces/${encodeURIComponent(id)}/mcp-status`)
+  },
+  connectWorkspaceMcps(id: string): Promise<{
+    success: boolean
+    servers: Record<string, { connected: boolean }>
+    toolCount: number
+  }> {
+    return request(`/api/workspaces/${encodeURIComponent(id)}/mcp-connect`, { method: "POST" })
+  },

  fetchConfig(): Promise<AppConfig> {
    return request<AppConfig>("/api/config/app")
--- a/packages/ui/src/lib/lite-mode.ts
+++ b/packages/ui/src/lib/lite-mode.ts
@@ -0,0 +1,227 @@
+/**
+ * Lite Mode API Client - Binary-Free Mode
+ * 
+ * This provides a client for working with NomadArch in Binary-Free Mode,
+ * using native session management instead of the OpenCode binary.
+ */
+
+import { CODENOMAD_API_BASE } from "./api-client"
+import { getLogger } from "./logger"
+
+const log = getLogger("lite-mode")
+
+export interface ModeInfo {
+    mode: "lite" | "full"
+    binaryFreeMode: boolean
+    nativeSessions: boolean
+    opencodeBinaryAvailable: boolean
+    providers: {
+        qwen: boolean
+        zai: boolean
+        zen: boolean
+    }
+}
+
+export interface NativeSession {
+    id: string
+    workspaceId: string
+    title?: string
+    parentId?: string | null
+    createdAt: number
+    updatedAt: number
+    messageIds: string[]
+    model?: {
+        providerId: string
+        modelId: string
+    }
+    agent?: string
+}
+
+export interface NativeMessage {
+    id: string
+    sessionId: string
+    role: "user" | "assistant" | "system" | "tool"
+    content?: string
+    createdAt: number
+    updatedAt: number
+    status?: "pending" | "streaming" | "completed" | "error"
+}
+
+let modeCache: ModeInfo | null = null
+
+/**
+ * Get the current running mode (lite or full)
+ */
+export async function getMode(): Promise<ModeInfo> {
+    if (modeCache) return modeCache
+
+    try {
+        const response = await fetch(`${CODENOMAD_API_BASE}/api/meta/mode`)
+        if (!response.ok) {
+            throw new Error(`Failed to fetch mode: ${response.status}`)
+        }
+        modeCache = await response.json()
+        log.info(`Running in ${modeCache?.mode} mode`, { binaryFree: modeCache?.binaryFreeMode })
+        return modeCache!
+    } catch (error) {
+        log.warn("Failed to fetch mode, assuming lite mode", error)
+        // Default to lite mode if we can't determine
+        return {
+            mode: "lite",
+            binaryFreeMode: true,
+            nativeSessions: true,
+            opencodeBinaryAvailable: false,
+            providers: { qwen: true, zai: true, zen: true }
+        }
+    }
+}
+
+/**
+ * Check if running in Binary-Free (lite) mode
+ */
+export async function isLiteMode(): Promise<boolean> {
+    const mode = await getMode()
+    return mode.binaryFreeMode
+}
+
+/**
+ * Native Session API for Binary-Free Mode
+ */
+export const nativeSessionApi = {
+    async listSessions(workspaceId: string): Promise<NativeSession[]> {
+        const response = await fetch(`${CODENOMAD_API_BASE}/api/native/workspaces/${encodeURIComponent(workspaceId)}/sessions`)
+        if (!response.ok) throw new Error("Failed to list sessions")
+        const data = await response.json()
+        return data.sessions
+    },
+
+    async createSession(workspaceId: string, options?: {
+        title?: string
+        parentId?: string
+        model?: { providerId: string; modelId: string }
+        agent?: string
+    }): Promise<NativeSession> {
+        const response = await fetch(`${CODENOMAD_API_BASE}/api/native/workspaces/${encodeURIComponent(workspaceId)}/sessions`, {
+            method: "POST",
+            headers: { "Content-Type": "application/json" },
+            body: JSON.stringify(options ?? {})
+        })
+        if (!response.ok) throw new Error("Failed to create session")
+        const data = await response.json()
+        return data.session
+    },
+
+    async getSession(workspaceId: string, sessionId: string): Promise<NativeSession | null> {
+        const response = await fetch(`${CODENOMAD_API_BASE}/api/native/workspaces/${encodeURIComponent(workspaceId)}/sessions/${encodeURIComponent(sessionId)}`)
+        if (response.status === 404) return null
+        if (!response.ok) throw new Error("Failed to get session")
+        const data = await response.json()
+        return data.session
+    },
+
+    async updateSession(workspaceId: string, sessionId: string, updates: Partial<NativeSession>): Promise<NativeSession | null> {
+        const response = await fetch(`${CODENOMAD_API_BASE}/api/native/workspaces/${encodeURIComponent(workspaceId)}/sessions/${encodeURIComponent(sessionId)}`, {
+            method: "PATCH",
+            headers: { "Content-Type": "application/json" },
+            body: JSON.stringify(updates)
+        })
+        if (response.status === 404) return null
+        if (!response.ok) throw new Error("Failed to update session")
+        const data = await response.json()
+        return data.session
+    },
+
+    async deleteSession(workspaceId: string, sessionId: string): Promise<boolean> {
+        const response = await fetch(`${CODENOMAD_API_BASE}/api/native/workspaces/${encodeURIComponent(workspaceId)}/sessions/${encodeURIComponent(sessionId)}`, {
+            method: "DELETE"
+        })
+        return response.ok || response.status === 204
+    },
+
+    async getMessages(workspaceId: string, sessionId: string): Promise<NativeMessage[]> {
+        const response = await fetch(`${CODENOMAD_API_BASE}/api/native/workspaces/${encodeURIComponent(workspaceId)}/sessions/${encodeURIComponent(sessionId)}/messages`)
+        if (!response.ok) throw new Error("Failed to get messages")
+        const data = await response.json()
+        return data.messages
+    },
+
+    /**
+     * Send a prompt to the session and get a streaming response
+     */
+    async* streamPrompt(
+        workspaceId: string,
+        sessionId: string,
+        content: string,
+        options?: {
+            provider?: "qwen" | "zai" | "zen"
+            accessToken?: string
+            resourceUrl?: string
+            enableTools?: boolean
+        }
+    ): AsyncGenerator<{ type: "content" | "done" | "error"; data?: string }> {
+        const response = await fetch(`${CODENOMAD_API_BASE}/api/native/workspaces/${encodeURIComponent(workspaceId)}/sessions/${encodeURIComponent(sessionId)}/prompt`, {
+            method: "POST",
+            headers: { "Content-Type": "application/json" },
+            body: JSON.stringify({
+                content,
+                provider: options?.provider ?? "qwen",
+                accessToken: options?.accessToken,
+                resourceUrl: options?.resourceUrl,
+                enableTools: options?.enableTools ?? true
+            })
+        })
+
+        if (!response.ok) {
+            yield { type: "error", data: `Request failed: ${response.status}` }
+            return
+        }
+
+        const reader = response.body?.getReader()
+        if (!reader) {
+            yield { type: "error", data: "No response body" }
+            return
+        }
+
+        const decoder = new TextDecoder()
+        let buffer = ""
+
+        while (true) {
+            const { done, value } = await reader.read()
+            if (done) break
+
+            buffer += decoder.decode(value, { stream: true })
+            const lines = buffer.split("\n")
+            buffer = lines.pop() ?? ""
+
+            for (const line of lines) {
+                if (!line.trim()) continue
+                if (line.startsWith("data: ")) {
+                    const data = line.slice(6)
+                    if (data === "[DONE]") {
+                        yield { type: "done" }
+                        return
+                    }
+                    try {
+                        const parsed = JSON.parse(data)
+                        if (parsed.error) {
+                            yield { type: "error", data: parsed.error }
+                        } else if (parsed.choices?.[0]?.delta?.content) {
+                            yield { type: "content", data: parsed.choices[0].delta.content }
+                        }
+                    } catch {
+                        // Skip invalid JSON
+                    }
+                }
+            }
+        }
+
+        yield { type: "done" }
+    }
+}
+
+/**
+ * Clear mode cache (for testing or after config changes)
+ */
+export function clearModeCache(): void {
+    modeCache = null
+}
--- a/packages/ui/src/lib/markdown.ts
+++ b/packages/ui/src/lib/markdown.ts
@@ -260,9 +260,21 @@ function setupRenderer(isDark: boolean) {
    const resolvedLang = lang && lang.trim() ? lang.trim() : "text"
    const escapedLang = escapeHtml(resolvedLang)

+    const previewButton = resolvedLang === "html" ? `
+  <button class="code-block-preview" data-code="${encodedCode}" data-lang="${escapedLang}">
+    <svg class="preview-icon" width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+      <path d="M1 12s4-8 11-8 11 8 11 8-4 8-11 8-11-8-11-8z"></path>
+      <circle cx="12" cy="12" r="3"></circle>
+    </svg>
+    <span class="preview-text">Preview</span>
+  </button>` : "";
+
    const header = `
 <div class="code-block-header">
-  <span class="code-block-language">${escapedLang}</span>
+  <div class="flex items-center gap-2">
+    <span class="code-block-language">${escapedLang}</span>
+    ${previewButton}
+  </div>
  <button class="code-block-copy" data-code="${encodedCode}">
    <svg class="copy-icon" width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
      <rect x="9" y="9" width="13" height="13" rx="2" ry="2"></rect>
--- a/packages/ui/src/renderer/loading/main.tsx
+++ b/packages/ui/src/renderer/loading/main.tsx
@@ -202,7 +202,7 @@ function LoadingApp() {
      <img src={iconUrl} alt="NomadArch" class="loading-logo" width="180" height="180" />
      <div class="loading-heading">
        <h1 class="loading-title">NomadArch 1.0</h1>
-        <p class="loading-subtitle" style={{ "font-size": '14px', "color": '#666', "margin-top": '4px' }}>A fork of OpenCode</p>
+        <p class="loading-subtitle" style={{ "font-size": '14px', "color": '#666', "margin-top": '4px' }}>An enhanced fork of CodeNomad</p>
        <Show when={status()}>{(statusText) => <p class="loading-status">{statusText()}</p>}</Show>
      </div>
      <div class="loading-card">
--- a/packages/ui/src/services/compaction-service.ts
+++ b/packages/ui/src/services/compaction-service.ts
@@ -0,0 +1,160 @@
+/**
+ * Compaction Service
+ * 
+ * Integrates opencode-style compaction logic for managing context window:
+ * - Detect token overflow
+ * - Prune old tool outputs (keep last 40k tokens protected)
+ * - Track compacted parts with timestamps
+ */
+
+import { getLogger } from "@/lib/logger";
+
+const log = getLogger("compaction-service");
+
+// Configuration from opencode
+export const PRUNE_MINIMUM = 20_000;  // Minimum tokens to prune before triggering
+export const PRUNE_PROTECT = 40_000;  // Protect last N tokens of tool outputs
+export const PRUNE_PROTECTED_TOOLS = ["skill"];  // Tools to never prune
+
+export interface CompactionConfig {
+    contextLimit: number;
+    outputMax: number;
+}
+
+export interface TokenInfo {
+    input: number;
+    output: number;
+    cacheRead?: number;
+    cacheWrite?: number;
+}
+
+export interface PruneResult {
+    pruned: number;
+    total: number;
+    partsCount: number;
+}
+
+/**
+ * Check if context is overflowing and compaction is needed
+ */
+export function isOverflow(tokens: TokenInfo, model: CompactionConfig): boolean {
+    const context = model.contextLimit;
+    if (context === 0) return false;
+
+    const count = tokens.input + (tokens.cacheRead || 0) + tokens.output;
+    const output = Math.min(model.outputMax, 32000) || 32000;
+    const usable = context - output;
+
+    return count > usable;
+}
+
+/**
+ * Calculate how many tokens to prune from old tool outputs
+ */
+export function calculatePruneTarget(
+    toolOutputTokens: number[],
+    protectThreshold: number = PRUNE_PROTECT
+): { toPrune: number[]; prunedTotal: number } {
+    let total = 0;
+    let pruned = 0;
+    const toPrune: number[] = [];
+
+    // Go through tool outputs from newest to oldest
+    for (let i = toolOutputTokens.length - 1; i >= 0; i--) {
+        const estimate = toolOutputTokens[i];
+        total += estimate;
+
+        // Once we've protected enough, mark the rest for pruning
+        if (total > protectThreshold) {
+            pruned += estimate;
+            toPrune.push(i);
+        }
+    }
+
+    return { toPrune, prunedTotal: pruned };
+}
+
+/**
+ * Estimate token count from text (rough approximation)
+ */
+export function estimateTokens(text: string): number {
+    if (!text) return 0;
+    // Rough estimate: 1 token ≈ 4 characters
+    return Math.ceil(text.length / 4);
+}
+
+/**
+ * Generate default compaction prompt
+ */
+export function getDefaultCompactionPrompt(): string {
+    return `Provide a detailed prompt for continuing our conversation above. Focus on information that would be helpful for continuing the conversation, including what we did, what we're doing, which files we're working on, and what we're going to do next considering new session will not have access to our conversation.`;
+}
+
+/**
+ * Check if a tool should be protected from pruning
+ */
+export function isProtectedTool(toolName: string): boolean {
+    return PRUNE_PROTECTED_TOOLS.includes(toolName);
+}
+
+/**
+ * Calculate context usage percentage
+ */
+export function getContextUsagePercent(tokens: TokenInfo, contextLimit: number): number {
+    if (contextLimit === 0) return 0;
+    const used = tokens.input + (tokens.cacheRead || 0) + tokens.output;
+    return Math.round((used / contextLimit) * 100);
+}
+
+/**
+ * Get compaction recommendation
+ */
+export function getCompactionRecommendation(
+    tokens: TokenInfo,
+    model: CompactionConfig
+): { shouldCompact: boolean; reason: string; urgency: "low" | "medium" | "high" } {
+    const usagePercent = getContextUsagePercent(tokens, model.contextLimit);
+
+    if (usagePercent >= 90) {
+        return {
+            shouldCompact: true,
+            reason: `Context ${usagePercent}% full - compaction required`,
+            urgency: "high"
+        };
+    }
+
+    if (usagePercent >= 75) {
+        return {
+            shouldCompact: true,
+            reason: `Context ${usagePercent}% full - compaction recommended`,
+            urgency: "medium"
+        };
+    }
+
+    if (usagePercent >= 50) {
+        return {
+            shouldCompact: false,
+            reason: `Context ${usagePercent}% full`,
+            urgency: "low"
+        };
+    }
+
+    return {
+        shouldCompact: false,
+        reason: "",
+        urgency: "low"
+    };
+}
+
+export default {
+    isOverflow,
+    calculatePruneTarget,
+    estimateTokens,
+    getDefaultCompactionPrompt,
+    isProtectedTool,
+    getContextUsagePercent,
+    getCompactionRecommendation,
+    PRUNE_MINIMUM,
+    PRUNE_PROTECT,
+    PRUNE_PROTECTED_TOOLS,
+};
--- a/packages/ui/src/services/compaction/index.ts
+++ b/packages/ui/src/services/compaction/index.ts
@@ -0,0 +1,20 @@
+// Compaction Service Exports
+export {
+    isOverflow,
+    calculatePruneTarget,
+    estimateTokens,
+    getCompactionPrompt,
+    isProtectedTool,
+    getContextUsagePercent,
+    getCompactionRecommendation,
+    compactMessages,
+    PRUNE_MINIMUM,
+    PRUNE_PROTECT,
+    PRUNE_PROTECTED_TOOLS,
+} from "./service";
+
+export type {
+    CompactionConfig,
+    TokenInfo,
+    PruneResult,
+} from "./service";
--- a/packages/ui/src/services/compaction/service.ts
+++ b/packages/ui/src/services/compaction/service.ts
@@ -0,0 +1,216 @@
+/**
+ * Compaction Service
+ * 
+ * Source: https://github.com/sst/opencode.git
+ * Source: https://github.com/MiniMax-AI/Mini-Agent.git
+ * 
+ * Implements intelligent context management:
+ * - Detect token overflow
+ * - Prune old tool outputs (keep last 40k tokens protected)
+ * - Generate summaries for compacted content
+ */
+
+import { getLogger } from "@/lib/logger";
+
+const log = getLogger("compaction-service");
+
+// Configuration from OpenCode
+export const PRUNE_MINIMUM = 20_000;  // Minimum tokens before pruning
+export const PRUNE_PROTECT = 40_000;  // Protect last N tokens
+export const PRUNE_PROTECTED_TOOLS = ["skill", "execute"];  // Never prune these
+
+export interface CompactionConfig {
+    contextLimit: number;
+    outputMax: number;
+}
+
+export interface TokenInfo {
+    input: number;
+    output: number;
+    cacheRead?: number;
+    cacheWrite?: number;
+}
+
+export interface PruneResult {
+    pruned: number;
+    total: number;
+    partsCount: number;
+}
+
+/**
+ * Check if context is overflowing and compaction is needed
+ */
+export function isOverflow(tokens: TokenInfo, model: CompactionConfig): boolean {
+    const context = model.contextLimit;
+    if (context === 0) return false;
+
+    const count = tokens.input + (tokens.cacheRead || 0) + tokens.output;
+    const output = Math.min(model.outputMax, 32000) || 32000;
+    const usable = context - output;
+
+    return count > usable;
+}
+
+/**
+ * Calculate how many tokens to prune from old tool outputs
+ */
+export function calculatePruneTarget(
+    toolOutputTokens: number[],
+    protectThreshold: number = PRUNE_PROTECT
+): { toPrune: number[]; prunedTotal: number } {
+    let total = 0;
+    let pruned = 0;
+    const toPrune: number[] = [];
+
+    // Go through tool outputs from newest to oldest
+    for (let i = toolOutputTokens.length - 1; i >= 0; i--) {
+        const estimate = toolOutputTokens[i];
+        total += estimate;
+
+        // Once we've protected enough, mark the rest for pruning
+        if (total > protectThreshold) {
+            pruned += estimate;
+            toPrune.push(i);
+        }
+    }
+
+    return { toPrune, prunedTotal: pruned };
+}
+
+/**
+ * Estimate token count from text (rough approximation)
+ */
+export function estimateTokens(text: string): number {
+    if (!text) return 0;
+    // Rough estimate: 1 token ≈ 4 characters
+    return Math.ceil(text.length / 4);
+}
+
+/**
+ * Generate the compaction summary prompt
+ */
+export function getCompactionPrompt(): string {
+    return `Provide a detailed summary for continuing this conversation. Focus on:
+1. What we accomplished so far
+2. Which files we're working on
+3. Current state and any pending tasks
+4. Important decisions made
+5. What we're doing next
+
+Be concise but comprehensive. The new session will not have access to the full conversation history.`;
+}
+
+/**
+ * Check if a tool should be protected from pruning
+ */
+export function isProtectedTool(toolName: string): boolean {
+    return PRUNE_PROTECTED_TOOLS.some(t => toolName.toLowerCase().includes(t));
+}
+
+/**
+ * Calculate context usage percentage
+ */
+export function getContextUsagePercent(tokens: TokenInfo, contextLimit: number): number {
+    if (contextLimit === 0) return 0;
+    const used = tokens.input + (tokens.cacheRead || 0) + tokens.output;
+    return Math.round((used / contextLimit) * 100);
+}
+
+/**
+ * Get compaction recommendation
+ */
+export function getCompactionRecommendation(
+    tokens: TokenInfo,
+    model: CompactionConfig
+): { shouldCompact: boolean; reason: string; urgency: "low" | "medium" | "high" } {
+    const usagePercent = getContextUsagePercent(tokens, model.contextLimit);
+
+    if (usagePercent >= 90) {
+        return {
+            shouldCompact: true,
+            reason: `Context ${usagePercent}% full - compaction required`,
+            urgency: "high"
+        };
+    }
+
+    if (usagePercent >= 75) {
+        return {
+            shouldCompact: true,
+            reason: `Context ${usagePercent}% full - compaction recommended`,
+            urgency: "medium"
+        };
+    }
+
+    if (usagePercent >= 50) {
+        return {
+            shouldCompact: false,
+            reason: `Context ${usagePercent}% full`,
+            urgency: "low"
+        };
+    }
+
+    return {
+        shouldCompact: false,
+        reason: "",
+        urgency: "low"
+    };
+}
+
+/**
+ * Compact messages by summarizing old ones
+ */
+export async function compactMessages(
+    messages: { role: string; content: string }[],
+    instanceId: string
+): Promise<{ summary: string; removedCount: number }> {
+    if (messages.length < 10) {
+        return { summary: "", removedCount: 0 };
+    }
+
+    // Take the first 50% of messages for summarization
+    const cutoff = Math.floor(messages.length / 2);
+    const toSummarize = messages.slice(0, cutoff);
+
+    log.info("Compacting messages", { total: messages.length, summarizing: cutoff });
+
+    try {
+        const response = await fetch("/api/ollama/chat", {
+            method: "POST",
+            headers: { "Content-Type": "application/json" },
+            body: JSON.stringify({
+                model: "minimax-m1",
+                messages: [
+                    ...toSummarize,
+                    { role: "user", content: getCompactionPrompt() }
+                ],
+                stream: false
+            })
+        });
+
+        if (!response.ok) {
+            throw new Error(`Compaction API failed: ${response.status}`);
+        }
+
+        const data = await response.json();
+        const summary = data?.message?.content || "";
+
+        return { summary, removedCount: cutoff };
+    } catch (error) {
+        log.error("Compaction failed", error);
+        return { summary: "", removedCount: 0 };
+    }
+}
+
+export default {
+    isOverflow,
+    calculatePruneTarget,
+    estimateTokens,
+    getCompactionPrompt,
+    isProtectedTool,
+    getContextUsagePercent,
+    getCompactionRecommendation,
+    compactMessages,
+    PRUNE_MINIMUM,
+    PRUNE_PROTECT,
+    PRUNE_PROTECTED_TOOLS,
+};
--- a/packages/ui/src/services/context-engine/index.ts
+++ b/packages/ui/src/services/context-engine/index.ts
@@ -0,0 +1,13 @@
+// Context Engine Exports
+export {
+    ContextEngineService,
+    getContextEngine,
+    initializeContextEngine,
+} from "./service";
+
+export type {
+    ContextEngineConfig,
+    RetrievedContext,
+    ContextSource,
+    IndexStats,
+} from "./service";
--- a/packages/ui/src/services/context-engine/service.ts
+++ b/packages/ui/src/services/context-engine/service.ts
@@ -0,0 +1,201 @@
+/**
+ * Context Engine Service
+ * 
+ * Source: https://github.com/Eskapeum/Context-Engine
+ * 
+ * Provides intelligent context retrieval for enhanced AI responses:
+ * - File indexing and caching
+ * - Semantic search across codebase
+ * - Q&A memory for persistent knowledge
+ */
+
+import { getLogger } from "@/lib/logger";
+
+const log = getLogger("context-engine");
+
+export interface ContextEngineConfig {
+    projectRoot: string;
+    enableIndexing?: boolean;
+    enableVectors?: boolean;
+}
+
+export interface RetrievedContext {
+    content: string;
+    sources: ContextSource[];
+    relevanceScore: number;
+}
+
+export interface ContextSource {
+    file: string;
+    line?: number;
+    symbol?: string;
+    type: "code" | "documentation" | "memory" | "qa";
+}
+
+export interface IndexStats {
+    filesIndexed: number;
+    symbolsFound: number;
+    lastUpdated: number;
+}
+
+// Singleton instance
+let instance: ContextEngineService | null = null;
+
+export class ContextEngineService {
+    private config: ContextEngineConfig;
+    private initialized: boolean = false;
+    private stats: IndexStats = { filesIndexed: 0, symbolsFound: 0, lastUpdated: 0 };
+    private memoryCache: Map<string, { question: string; answer: string; timestamp: number }> = new Map();
+
+    constructor(config: ContextEngineConfig) {
+        this.config = config;
+    }
+
+    /**
+     * Initialize the context engine
+     */
+    async initialize(): Promise<void> {
+        if (this.initialized) return;
+
+        log.info("Context Engine initializing...", { projectRoot: this.config.projectRoot });
+
+        // In a full implementation, this would:
+        // 1. Scan the project directory
+        // 2. Build Tree-sitter AST for each file
+        // 3. Create embeddings for semantic search
+
+        this.initialized = true;
+        log.info("Context Engine initialized");
+    }
+
+    /**
+     * Retrieve relevant context for a query
+     */
+    async retrieve(query: string, options?: { maxTokens?: number }): Promise<RetrievedContext> {
+        if (!this.initialized) {
+            await this.initialize();
+        }
+
+        log.info("Retrieving context for query", { query: query.substring(0, 50) });
+
+        // Search memory cache first
+        const memorySuggestions = this.searchMemory(query);
+
+        // In a full implementation, this would:
+        // 1. Vectorize the query
+        // 2. Search the index for relevant files/symbols
+        // 3. Rank results by relevance
+        // 4. Return top matches within token budget
+
+        return {
+            content: memorySuggestions.join("\n\n"),
+            sources: [],
+            relevanceScore: 0
+        };
+    }
+
+    /**
+     * Index or re-index the project
+     */
+    async index(options?: { force?: boolean }): Promise<IndexStats> {
+        log.info("Indexing project...", { force: options?.force });
+
+        // In a full implementation, this would:
+        // 1. Walk the file tree
+        // 2. Parse each file with Tree-sitter
+        // 3. Extract symbols and documentation
+        // 4. Generate embeddings
+
+        this.stats = {
+            filesIndexed: 0,
+            symbolsFound: 0,
+            lastUpdated: Date.now()
+        };
+
+        return this.stats;
+    }
+
+    /**
+     * Get current index stats
+     */
+    getStats(): IndexStats {
+        return this.stats;
+    }
+
+    /**
+     * Add to Q&A memory
+     */
+    async remember(question: string, answer: string): Promise<void> {
+        const id = `qa_${Date.now()}`;
+        this.memoryCache.set(id, {
+            question,
+            answer,
+            timestamp: Date.now()
+        });
+        log.info("Remembered Q&A", { question: question.substring(0, 50) });
+    }
+
+    /**
+     * Search Q&A memory
+     */
+    searchMemory(query: string): string[] {
+        const results: string[] = [];
+        const queryLower = query.toLowerCase();
+
+        for (const [, entry] of this.memoryCache) {
+            if (entry.question.toLowerCase().includes(queryLower) ||
+                entry.answer.toLowerCase().includes(queryLower)) {
+                results.push(`Q: ${entry.question}\nA: ${entry.answer}`);
+            }
+        }
+
+        return results.slice(0, 5);
+    }
+
+    /**
+     * Search Q&A memory (async version)
+     */
+    async recall(query: string): Promise<{ question: string; answer: string }[]> {
+        log.info("Recalling from memory", { query: query.substring(0, 50) });
+
+        const results: { question: string; answer: string }[] = [];
+        const queryLower = query.toLowerCase();
+
+        for (const [, entry] of this.memoryCache) {
+            if (entry.question.toLowerCase().includes(queryLower) ||
+                entry.answer.toLowerCase().includes(queryLower)) {
+                results.push({ question: entry.question, answer: entry.answer });
+            }
+        }
+
+        return results.slice(0, 10);
+    }
+}
+
+/**
+ * Get or create context engine instance
+ */
+export function getContextEngine(config?: ContextEngineConfig): ContextEngineService {
+    if (!instance && config) {
+        instance = new ContextEngineService(config);
+    }
+    if (!instance) {
+        throw new Error("Context engine not initialized. Provide config on first call.");
+    }
+    return instance;
+}
+
+/**
+ * Initialize context engine for a workspace
+ */
+export async function initializeContextEngine(projectRoot: string): Promise<ContextEngineService> {
+    const service = getContextEngine({ projectRoot });
+    await service.initialize();
+    return service;
+}
+
+export default {
+    ContextEngineService,
+    getContextEngine,
+    initializeContextEngine,
+};
--- a/packages/ui/src/services/context-service.ts
+++ b/packages/ui/src/services/context-service.ts
@@ -0,0 +1,172 @@
+/**
+ * Context Engine Service
+ * 
+ * Stub for Context-Engine integration (https://github.com/Eskapeum/Context-Engine)
+ * 
+ * Features to integrate:
+ * - Tree-sitter AST parsing (20+ languages)
+ * - Incremental indexing with dependency tracking
+ * - Git branch-aware per-user indexing
+ * - cAST semantic chunking for optimal retrieval
+ * - Hybrid retrieval (BM25 + dense vectors)
+ * - MCP server for Claude Code integration
+ * - Library documentation (local-first with optional API)
+ * - Sequential thinking engine for complex reasoning
+ * - Persistent memory with Q&A history
+ * - Context sharing for team collaboration
+ */
+
+import { getLogger } from "@/lib/logger";
+
+const log = getLogger("context-service");
+
+export interface ContextEngineConfig {
+    projectRoot: string;
+    enableIndexing?: boolean;
+    enableVectors?: boolean;
+    vectorProvider?: "voyage" | "openai" | "local";
+}
+
+export interface RetrievedContext {
+    content: string;
+    sources: ContextSource[];
+    relevanceScore: number;
+}
+
+export interface ContextSource {
+    file: string;
+    line?: number;
+    symbol?: string;
+    type: "code" | "documentation" | "memory" | "qa";
+}
+
+export interface IndexStats {
+    filesIndexed: number;
+    symbolsFound: number;
+    lastUpdated: number;
+}
+
+// Singleton instance
+let instance: ContextService | null = null;
+
+export class ContextService {
+    private config: ContextEngineConfig;
+    private initialized: boolean = false;
+    private stats: IndexStats = { filesIndexed: 0, symbolsFound: 0, lastUpdated: 0 };
+
+    constructor(config: ContextEngineConfig) {
+        this.config = config;
+    }
+
+    /**
+     * Initialize the context engine
+     */
+    async initialize(): Promise<void> {
+        if (this.initialized) return;
+
+        log.info("Context Engine initializing...", { projectRoot: this.config.projectRoot });
+
+        // TODO: Initialize Context-Engine
+        // const engine = new ContextEngine({ projectRoot: this.config.projectRoot });
+        // await engine.initialize();
+
+        this.initialized = true;
+        log.info("Context Engine initialized (stub)");
+    }
+
+    /**
+     * Retrieve relevant context for a query
+     */
+    async retrieve(query: string, options?: { maxTokens?: number }): Promise<RetrievedContext> {
+        if (!this.initialized) {
+            await this.initialize();
+        }
+
+        log.info("Retrieving context for query", { query: query.substring(0, 50) });
+
+        // TODO: Call Context-Engine retrieve
+        // const context = await engine.retrieve(query);
+
+        // Return stub response
+        return {
+            content: "",
+            sources: [],
+            relevanceScore: 0
+        };
+    }
+
+    /**
+     * Index or re-index the project
+     */
+    async index(options?: { force?: boolean }): Promise<IndexStats> {
+        log.info("Indexing project...", { force: options?.force });
+
+        // TODO: Call Context-Engine indexer
+        // await engine.index();
+
+        this.stats = {
+            filesIndexed: 0,
+            symbolsFound: 0,
+            lastUpdated: Date.now()
+        };
+
+        return this.stats;
+    }
+
+    /**
+     * Get current index stats
+     */
+    getStats(): IndexStats {
+        return this.stats;
+    }
+
+    /**
+     * Add to Q&A memory
+     */
+    async remember(question: string, answer: string): Promise<void> {
+        log.info("Remembering Q&A", { question: question.substring(0, 50) });
+
+        // TODO: Call Context-Engine memory
+        // await engine.remember(question, answer);
+    }
+
+    /**
+     * Search Q&A memory
+     */
+    async recall(query: string): Promise<{ question: string; answer: string }[]> {
+        log.info("Recalling from memory", { query: query.substring(0, 50) });
+
+        // TODO: Call Context-Engine memory search
+        // return await engine.recall(query);
+
+        return [];
+    }
+}
+
+/**
+ * Get or create context service instance
+ */
+export function getContextService(config?: ContextEngineConfig): ContextService {
+    if (!instance && config) {
+        instance = new ContextService(config);
+    }
+    if (!instance) {
+        throw new Error("Context service not initialized. Provide config on first call.");
+    }
+    return instance;
+}
+
+/**
+ * Initialize context service for a workspace
+ */
+export async function initializeContextService(projectRoot: string): Promise<ContextService> {
+    const service = getContextService({ projectRoot });
+    await service.initialize();
+    return service;
+}
+
+export default {
+    ContextService,
+    getContextService,
+    initializeContextService,
+};
--- a/packages/ui/src/stores/instance-config.tsx
+++ b/packages/ui/src/stores/instance-config.tsx
@@ -29,9 +29,14 @@ function cloneInstanceData(data?: InstanceData | null): InstanceData {
  }
 }

+// Track instance IDs that we are currently saving - ignore SSE echoes
+const pendingSaveIds = new Set<string>()
+
 function attachSubscription(instanceId: string) {
  if (instanceSubscriptions.has(instanceId)) return
  const unsubscribe = storage.onInstanceDataChanged(instanceId, (data) => {
+    // Skip SSE echo from our own save
+    if (pendingSaveIds.has(instanceId)) return
    setInstanceData(instanceId, data)
  })
  instanceSubscriptions.set(instanceId, unsubscribe)
@@ -83,12 +88,15 @@ async function updateInstanceConfig(instanceId: string, mutator: (draft: Instanc
  const current = instanceDataMap().get(instanceId) ?? DEFAULT_INSTANCE_DATA
  const draft = cloneInstanceData(current)
  mutator(draft)
+  setInstanceData(instanceId, draft)
+  pendingSaveIds.add(instanceId)
  try {
    await storage.saveInstanceData(instanceId, draft)
  } catch (error) {
    log.warn("Failed to persist instance data", error)
+  } finally {
+    setTimeout(() => pendingSaveIds.delete(instanceId), 1000)
  }
-  setInstanceData(instanceId, draft)
 }

 function getInstanceConfig(instanceId: string): InstanceData {
--- a/packages/ui/src/stores/instances.ts
+++ b/packages/ui/src/stores/instances.ts
@@ -170,13 +170,52 @@ function handleWorkspaceEvent(event: WorkspaceEventPayload) {
  }
 }

+const logBuffer = new Map<string, LogEntry[]>()
+let logFlushPending = false
+
+function flushLogs() {
+  if (logBuffer.size === 0) {
+    logFlushPending = false
+    return
+  }
+
+  batch(() => {
+    setInstanceLogs((prev) => {
+      const next = new Map(prev)
+      for (const [id, newEntries] of logBuffer) {
+        const existing = next.get(id) ?? []
+        // Keep only last MAX_LOG_ENTRIES
+        const combined = [...existing, ...newEntries]
+        const updated = combined.slice(-MAX_LOG_ENTRIES)
+        next.set(id, updated)
+      }
+      return next
+    })
+  })
+  logBuffer.clear()
+  logFlushPending = false
+}
+
 function handleWorkspaceLog(entry: WorkspaceLogEntry) {
  const logEntry: LogEntry = {
    timestamp: new Date(entry.timestamp).getTime(),
    level: (entry.level as LogEntry["level"]) ?? "info",
    message: entry.message,
  }
-  addLog(entry.workspaceId, logEntry)
+
+  // Only buffer if streaming is enabled for this instance, to save memory
+  if (!isInstanceLogStreaming(entry.workspaceId)) {
+    return
+  }
+
+  const currentBuffer = logBuffer.get(entry.workspaceId) ?? []
+  currentBuffer.push(logEntry)
+  logBuffer.set(entry.workspaceId, currentBuffer)
+
+  if (!logFlushPending) {
+    logFlushPending = true
+    setTimeout(flushLogs, 100) // Throttle updates to every 100ms
+  }
 }

 function ensureLogContainer(id: string) {
--- a/packages/ui/src/stores/message-v2/instance-store.ts
+++ b/packages/ui/src/stores/message-v2/instance-store.ts
@@ -1,4 +1,4 @@
-import { batch } from "solid-js"
+import { batch, untrack } from "solid-js"
 import { createStore, produce, reconcile } from "solid-js/store"
 import type { SetStoreFunction } from "solid-js/store"
 import { getLogger } from "../../lib/logger"
@@ -43,6 +43,7 @@ function createInitialState(instanceId: string): InstanceMessageState {
    usage: {},
    scrollState: {},
    latestTodos: {},
+    streamingUpdateCount: 0,
  }
 }

@@ -214,7 +215,10 @@ export interface InstanceMessageStore {
  getMessage: (messageId: string) => MessageRecord | undefined
  getLatestTodoSnapshot: (sessionId: string) => LatestTodoSnapshot | undefined
  clearSession: (sessionId: string) => void
+  beginStreamingUpdate: () => void
+  endStreamingUpdate: () => void
  clearInstance: () => void
+  isStreaming: () => boolean
 }

 export function createInstanceMessageStore(instanceId: string, hooks?: MessageStoreHooks): InstanceMessageStore {
@@ -271,6 +275,7 @@ export function createInstanceMessageStore(instanceId: string, hooks?: MessageSt

  function bumpSessionRevision(sessionId: string) {
    if (!sessionId) return
+    if (state.streamingUpdateCount > 0) return
    setState("sessionRevisions", sessionId, (value = 0) => value + 1)
  }

@@ -282,9 +287,9 @@ export function createInstanceMessageStore(instanceId: string, hooks?: MessageSt
    setState("usage", sessionId, (current) => {
      const draft = current
        ? {
-            ...current,
-            entries: { ...current.entries },
-          }
+          ...current,
+          entries: { ...current.entries },
+        }
        : createEmptyUsageState()
      updater(draft)
      return draft
@@ -464,23 +469,31 @@ export function createInstanceMessageStore(instanceId: string, hooks?: MessageSt

    let nextRecord: MessageRecord | undefined

-    setState("messages", input.id, (previous) => {
-      const revision = previous ? previous.revision + (shouldBump ? 1 : 0) : 0
-      const record: MessageRecord = {
-        id: input.id,
-        sessionId: input.sessionId,
-        role: input.role,
-        status: input.status,
-        createdAt: input.createdAt ?? previous?.createdAt ?? now,
-        updatedAt: input.updatedAt ?? now,
-        isEphemeral: input.isEphemeral ?? previous?.isEphemeral ?? false,
-        revision,
-        partIds: normalizedParts ? normalizedParts.ids : previous?.partIds ?? [],
-        parts: normalizedParts ? normalizedParts.map : previous?.parts ?? {},
-      }
-      nextRecord = record
-      return record
-    })
+    const updateState = () => {
+      setState("messages", input.id, (previous) => {
+        const revision = previous ? previous.revision + (shouldBump ? 1 : 0) : 0
+        const record: MessageRecord = {
+          id: input.id,
+          sessionId: input.sessionId,
+          role: input.role,
+          status: input.status,
+          createdAt: input.createdAt ?? previous?.createdAt ?? now,
+          updatedAt: input.updatedAt ?? now,
+          isEphemeral: input.isEphemeral ?? previous?.isEphemeral ?? false,
+          revision,
+          partIds: normalizedParts ? normalizedParts.ids : previous?.partIds ?? [],
+          parts: normalizedParts ? normalizedParts.map : previous?.parts ?? {},
+        }
+        nextRecord = record
+        return record
+      })
+    }
+
+    if (state.streamingUpdateCount > 0) {
+      untrack(updateState)
+    } else {
+      updateState()
+    }

    if (nextRecord) {
      maybeUpdateLatestTodoFromRecord(nextRecord)
@@ -516,26 +529,29 @@ export function createInstanceMessageStore(instanceId: string, hooks?: MessageSt
    const partId = ensurePartId(input.messageId, input.part, message.partIds.length)
    const cloned = clonePart(input.part)

-    setState(
-      "messages",
-      input.messageId,
-      produce((draft: MessageRecord) => {
-        if (!draft.partIds.includes(partId)) {
-          draft.partIds = [...draft.partIds, partId]
-        }
-        const existing = draft.parts[partId]
-        const nextRevision = existing ? existing.revision + 1 : 0
-        draft.parts[partId] = {
-          id: partId,
-          data: cloned,
-          revision: nextRevision,
-        }
-        draft.updatedAt = Date.now()
-        if (input.bumpRevision ?? true) {
-          draft.revision += 1
-        }
-      }),
-    )
+    const updateFn = produce((draft: MessageRecord) => {
+      if (!draft.partIds.includes(partId)) {
+        draft.partIds = [...draft.partIds, partId]
+      }
+      const existing = draft.parts[partId]
+      const nextRevision = existing ? existing.revision + 1 : 0
+      draft.parts[partId] = {
+        id: partId,
+        data: cloned,
+        revision: nextRevision,
+      }
+      draft.updatedAt = Date.now()
+      if (input.bumpRevision ?? true) {
+        draft.revision += 1
+      }
+    })
+
+    const updateMessage = () => setState("messages", input.messageId, updateFn)
+    if (state.streamingUpdateCount > 0) {
+      untrack(updateMessage)
+    } else {
+      updateMessage()
+    }

    if (isCompletedTodoPart(cloned)) {
      recordLatestTodoSnapshot(message.sessionId, {
@@ -637,8 +653,15 @@ export function createInstanceMessageStore(instanceId: string, hooks?: MessageSt
    if (!messageId) return
    messageInfoCache.set(messageId, info)
    const nextVersion = (state.messageInfoVersion[messageId] ?? 0) + 1
-    setState("messageInfoVersion", messageId, nextVersion)
-    updateUsageWithInfo(info)
+    if (state.streamingUpdateCount > 0) {
+      setState("messageInfoVersion", messageId, nextVersion)
+      updateUsageWithInfo(info)
+    } else {
+      untrack(() => {
+        setState("messageInfoVersion", messageId, nextVersion)
+        updateUsageWithInfo(info)
+      })
+    }
  }

  function getMessageInfo(messageId: string) {
@@ -775,8 +798,8 @@ export function createInstanceMessageStore(instanceId: string, hooks?: MessageSt
    return state.scrollState[key]
  }

-   function clearSession(sessionId: string) {
-     if (!sessionId) return
+  function clearSession(sessionId: string) {
+    if (!sessionId) return

    const messageIds = Object.values(state.messages)
      .filter((record) => record.sessionId === sessionId)
@@ -859,41 +882,55 @@ export function createInstanceMessageStore(instanceId: string, hooks?: MessageSt
  }


-   function clearInstance() {
-     messageInfoCache.clear()
-     setState(reconcile(createInitialState(instanceId)))
-   }
- 
-   return {
-
-     instanceId,
-     state,
-     setState,
-     addOrUpdateSession,
-     hydrateMessages,
-     upsertMessage,
-     applyPartUpdate,
-     bufferPendingPart,
-     flushPendingParts,
-     replaceMessageId,
-     setMessageInfo,
-     getMessageInfo,
-     upsertPermission,
-     removePermission,
-     getPermissionState,
-     setSessionRevert,
-     getSessionRevert,
-     rebuildUsage,
-     getSessionUsage,
-     setScrollSnapshot,
-     getScrollSnapshot,
-     getSessionRevision: getSessionRevisionValue,
-      getSessionMessageIds: (sessionId: string) => state.sessions[sessionId]?.messageIds ?? [],
-      getMessage: (messageId: string) => state.messages[messageId],
-      getLatestTodoSnapshot: (sessionId: string) => state.latestTodos[sessionId],
-      clearSession,
-      clearInstance,
-    }
+  function clearInstance() {
+    messageInfoCache.clear()
+    setState(reconcile(createInitialState(instanceId)))
  }

+  function beginStreamingUpdate() {
+    setState("streamingUpdateCount", (count) => count + 1)
+  }
+
+  function endStreamingUpdate() {
+    setState("streamingUpdateCount", (count) => Math.max(0, count - 1))
+  }
+
+  function isStreaming() {
+    return state.streamingUpdateCount > 0
+  }
+
+  return {
+    instanceId,
+    state,
+    setState,
+    addOrUpdateSession,
+    hydrateMessages,
+    upsertMessage,
+    applyPartUpdate,
+    bufferPendingPart,
+    flushPendingParts,
+    replaceMessageId,
+    setMessageInfo,
+    getMessageInfo,
+    upsertPermission,
+    removePermission,
+    getPermissionState,
+    setSessionRevert,
+    getSessionRevert,
+    rebuildUsage,
+    getSessionUsage,
+    setScrollSnapshot,
+    getScrollSnapshot,
+    getSessionRevision: getSessionRevisionValue,
+    getSessionMessageIds: (sessionId: string) => state.sessions[sessionId]?.messageIds ?? [],
+    getMessage: (messageId: string) => state.messages[messageId],
+    getLatestTodoSnapshot: (sessionId: string) => state.latestTodos[sessionId],
+    clearSession,
+    clearInstance,
+    beginStreamingUpdate,
+    endStreamingUpdate,
+    isStreaming,
+  }
+}
+

--- a/packages/ui/src/stores/message-v2/types.ts
+++ b/packages/ui/src/stores/message-v2/types.ts
@@ -1,7 +1,7 @@
 import type { ClientPart } from "../../types/message"
 import type { Permission } from "@opencode-ai/sdk"

-export type MessageStatus = "sending" | "sent" | "streaming" | "complete" | "error"
+export type MessageStatus = "sending" | "sent" | "streaming" | "complete" | "error" | "interrupted"
 export type MessageRole = "user" | "assistant"

 export interface NormalizedPartRecord {
@@ -108,6 +108,7 @@ export interface InstanceMessageState {
  usage: Record<string, SessionUsageState>
  scrollState: Record<string, ScrollSnapshot>
  latestTodos: Record<string, LatestTodoSnapshot | undefined>
+  streamingUpdateCount: number
 }

 export interface SessionUpsertInput {
--- a/packages/ui/src/stores/native-sessions.ts
+++ b/packages/ui/src/stores/native-sessions.ts
@@ -0,0 +1,319 @@
+/**
+ * Native Session Store - UI-side session management for Binary-Free Mode
+ * 
+ * This store provides a drop-in replacement for OpenCode SDK session operations
+ * when running in Binary-Free (Lite) Mode.
+ */
+
+import { createSignal, createMemo, batch } from "solid-js"
+import type { Session } from "../types/session"
+import type { Message, Part } from "../types/message"
+import { nativeSessionApi, isLiteMode, NativeSession, NativeMessage } from "../lib/lite-mode"
+import { getLogger } from "../lib/logger"
+
+const log = getLogger("native-sessions")
+
+// State
+const [nativeSessions, setNativeSessions] = createSignal<Map<string, Map<string, Session>>>(new Map())
+const [nativeMessages, setNativeMessages] = createSignal<Map<string, Message[]>>(new Map())
+const [isLiteModeActive, setIsLiteModeActive] = createSignal<boolean | null>(null)
+
+/**
+ * Check and cache lite mode status
+ */
+export async function checkLiteMode(): Promise<boolean> {
+    if (isLiteModeActive() !== null) {
+        return isLiteModeActive()!
+    }
+
+    try {
+        const liteMode = await isLiteMode()
+        setIsLiteModeActive(liteMode)
+        log.info(`Running in ${liteMode ? 'Lite' : 'Full'} mode`)
+        return liteMode
+    } catch (error) {
+        log.warn("Failed to check lite mode, defaulting to full mode", error)
+        setIsLiteModeActive(false)
+        return false
+    }
+}
+
+/**
+ * Get the current lite mode status (synchronous, may be null if not checked)
+ */
+export function getLiteModeStatus(): boolean | null {
+    return isLiteModeActive()
+}
+
+/**
+ * Force set lite mode (for testing or manual override)
+ */
+export function forceLiteMode(enabled: boolean): void {
+    setIsLiteModeActive(enabled)
+}
+
+// Convert native session to UI session format
+function nativeToUiSession(native: NativeSession): Session {
+    return {
+        id: native.id,
+        title: native.title,
+        parentId: native.parentId ?? undefined,
+        createdAt: native.createdAt,
+        updatedAt: native.updatedAt,
+        agent: native.agent,
+        model: native.model ? {
+            providerId: native.model.providerId,
+            modelId: native.model.modelId,
+        } : undefined,
+    }
+}
+
+// Convert native message to UI message format
+function nativeToUiMessage(native: NativeMessage): Message {
+    const parts: Part[] = []
+
+    if (native.content) {
+        parts.push({
+            type: "text",
+            text: native.content,
+        })
+    }
+
+    return {
+        id: native.id,
+        sessionId: native.sessionId,
+        role: native.role,
+        createdAt: native.createdAt,
+        parts,
+    }
+}
+
+/**
+ * Fetch sessions from native API
+ */
+export async function fetchNativeSessions(workspaceId: string): Promise<Session[]> {
+    try {
+        const sessions = await nativeSessionApi.listSessions(workspaceId)
+        const uiSessions = sessions.map(nativeToUiSession)
+
+        // Update state
+        setNativeSessions(prev => {
+            const next = new Map(prev)
+            const wsMap = new Map<string, Session>()
+            for (const s of uiSessions) {
+                wsMap.set(s.id, s)
+            }
+            next.set(workspaceId, wsMap)
+            return next
+        })
+
+        return uiSessions
+    } catch (error) {
+        log.error("Failed to fetch native sessions", error)
+        return []
+    }
+}
+
+/**
+ * Create a new native session
+ */
+export async function createNativeSession(
+    workspaceId: string,
+    options?: {
+        title?: string
+        parentId?: string
+        model?: { providerId: string; modelId: string }
+        agent?: string
+    }
+): Promise<Session> {
+    const native = await nativeSessionApi.createSession(workspaceId, options)
+    const session = nativeToUiSession(native)
+
+    // Update state
+    setNativeSessions(prev => {
+        const next = new Map(prev)
+        const wsMap = new Map(next.get(workspaceId) ?? new Map())
+        wsMap.set(session.id, session)
+        next.set(workspaceId, wsMap)
+        return next
+    })
+
+    return session
+}
+
+/**
+ * Delete a native session
+ */
+export async function deleteNativeSession(workspaceId: string, sessionId: string): Promise<void> {
+    await nativeSessionApi.deleteSession(workspaceId, sessionId)
+
+    // Update state
+    setNativeSessions(prev => {
+        const next = new Map(prev)
+        const wsMap = new Map(next.get(workspaceId) ?? new Map())
+        wsMap.delete(sessionId)
+        next.set(workspaceId, wsMap)
+        return next
+    })
+
+    // Clear messages
+    setNativeMessages(prev => {
+        const next = new Map(prev)
+        next.delete(`${workspaceId}:${sessionId}`)
+        return next
+    })
+}
+
+/**
+ * Get messages for a native session
+ */
+export async function fetchNativeMessages(workspaceId: string, sessionId: string): Promise<Message[]> {
+    try {
+        const messages = await nativeSessionApi.getMessages(workspaceId, sessionId)
+        const uiMessages = messages.map(nativeToUiMessage)
+
+        // Update state
+        const key = `${workspaceId}:${sessionId}`
+        setNativeMessages(prev => {
+            const next = new Map(prev)
+            next.set(key, uiMessages)
+            return next
+        })
+
+        return uiMessages
+    } catch (error) {
+        log.error("Failed to fetch native messages", error)
+        return []
+    }
+}
+
+/**
+ * Get cached native sessions for a workspace
+ */
+export function getNativeSessions(workspaceId: string): Session[] {
+    const wsMap = nativeSessions().get(workspaceId)
+    return wsMap ? Array.from(wsMap.values()) : []
+}
+
+/**
+ * Get cached native messages for a session
+ */
+export function getNativeMessages(workspaceId: string, sessionId: string): Message[] {
+    const key = `${workspaceId}:${sessionId}`
+    return nativeMessages().get(key) ?? []
+}
+
+/**
+ * Send a message to a native session with streaming
+ */
+export async function sendNativeMessage(
+    workspaceId: string,
+    sessionId: string,
+    content: string,
+    options?: {
+        provider?: "qwen" | "zai" | "zen"
+        accessToken?: string
+        resourceUrl?: string
+        enableTools?: boolean
+        onChunk?: (content: string) => void
+        onDone?: () => void
+        onError?: (error: string) => void
+    }
+): Promise<void> {
+    const { provider = "zen", accessToken, resourceUrl, enableTools = true, onChunk, onDone, onError } = options ?? {}
+
+    try {
+        // Add user message to local state immediately
+        const userMessage: Message = {
+            id: `temp-${Date.now()}`,
+            sessionId,
+            role: "user",
+            createdAt: Date.now(),
+            parts: [{ type: "text", text: content }],
+        }
+
+        const key = `${workspaceId}:${sessionId}`
+        setNativeMessages(prev => {
+            const next = new Map(prev)
+            const messages = [...(next.get(key) ?? []), userMessage]
+            next.set(key, messages)
+            return next
+        })
+
+        // Start streaming
+        let fullContent = ""
+
+        for await (const chunk of nativeSessionApi.streamPrompt(workspaceId, sessionId, content, {
+            provider,
+            accessToken,
+            resourceUrl,
+            enableTools,
+        })) {
+            if (chunk.type === "content" && chunk.data) {
+                fullContent += chunk.data
+                onChunk?.(chunk.data)
+            } else if (chunk.type === "error") {
+                onError?.(chunk.data ?? "Unknown error")
+                return
+            } else if (chunk.type === "done") {
+                break
+            }
+        }
+
+        // Add assistant message to local state
+        const assistantMessage: Message = {
+            id: `msg-${Date.now()}`,
+            sessionId,
+            role: "assistant",
+            createdAt: Date.now(),
+            parts: [{ type: "text", text: fullContent }],
+        }
+
+        setNativeMessages(prev => {
+            const next = new Map(prev)
+            const messages = [...(next.get(key) ?? []), assistantMessage]
+            next.set(key, messages)
+            return next
+        })
+
+        onDone?.()
+
+        // Refresh messages from server to get the real IDs
+        await fetchNativeMessages(workspaceId, sessionId)
+
+    } catch (error) {
+        log.error("Failed to send native message", error)
+        onError?.(String(error))
+    }
+}
+
+/**
+ * Update a native session
+ */
+export async function updateNativeSession(
+    workspaceId: string,
+    sessionId: string,
+    updates: { title?: string }
+): Promise<Session | null> {
+    const result = await nativeSessionApi.updateSession(workspaceId, sessionId, updates)
+    if (!result) return null
+
+    const session = nativeToUiSession(result)
+
+    // Update state
+    setNativeSessions(prev => {
+        const next = new Map(prev)
+        const wsMap = new Map(next.get(workspaceId) ?? new Map())
+        wsMap.set(session.id, session)
+        next.set(workspaceId, wsMap)
+        return next
+    })
+
+    return session
+}
+
+export {
+    nativeSessions,
+    nativeMessages,
+    isLiteModeActive,
+}
--- a/packages/ui/src/stores/releases.ts
+++ b/packages/ui/src/stores/releases.ts
@@ -1,4 +1,4 @@
-import { createEffect, createSignal } from "solid-js"
+import { createEffect, createSignal, createRoot } from "solid-js"
 import type { LatestReleaseInfo, WorkspaceEventPayload } from "../../../server/src/api-types"
 import { getServerMeta } from "../lib/server-meta"
 import { serverEvents } from "../lib/server-events"
@@ -29,30 +29,33 @@ function ensureVisibilityEffect() {
  }
  visibilityEffectInitialized = true

-  createEffect(() => {
-    const release = availableRelease()
-    const shouldShow = Boolean(release) && (!hasInstances() || showFolderSelection())
+  // Use createRoot to properly scope this effect
+  createRoot(() => {
+    createEffect(() => {
+      const release = availableRelease()
+      const shouldShow = Boolean(release) && (!hasInstances() || showFolderSelection())

-    if (!shouldShow || !release) {
-      dismissActiveToast()
-      return
-    }
+      if (!shouldShow || !release) {
+        dismissActiveToast()
+        return
+      }

-    if (!activeToast || activeToastVersion !== release.version) {
-      dismissActiveToast()
-      activeToast = showToastNotification({
-        title: `NomadArch ${release.version}`,
-        message: release.channel === "dev" ? "Dev release build available." : "New stable build on GitHub.",
-        variant: "info",
-        duration: Number.POSITIVE_INFINITY,
-        position: "bottom-right",
-        action: {
-          label: "View release",
-          href: release.url,
-        },
-      })
-      activeToastVersion = release.version
-    }
+      if (!activeToast || activeToastVersion !== release.version) {
+        dismissActiveToast()
+        activeToast = showToastNotification({
+          title: `NomadArch ${release.version}`,
+          message: release.channel === "dev" ? "Dev release build available." : "New stable build on GitHub.",
+          variant: "info",
+          duration: Number.POSITIVE_INFINITY,
+          position: "bottom-right",
+          action: {
+            label: "View release",
+            href: release.url,
+          },
+        })
+        activeToastVersion = release.version
+      }
+    })
  })
 }

--- a/packages/ui/src/stores/session-actions.ts
+++ b/packages/ui/src/stores/session-actions.ts
@@ -1,5 +1,7 @@
+import { untrack, batch } from "solid-js"
+import { addDebugLog } from "../components/debug-overlay"
 import { resolvePastedPlaceholders } from "../lib/prompt-placeholders"
-import { instances } from "./instances"
+import { instances, activeInstanceId } from "./instances"
 import { addTaskMessage } from "./task-actions"

 import { addRecentModelPreference, setAgentModelPreference, getAgentModelPreference } from "./preferences"
@@ -36,7 +38,8 @@ const COMPACTION_ATTEMPT_TTL_MS = 60_000
 const COMPACTION_SUMMARY_MAX_CHARS = 4000
 const STREAM_TIMEOUT_MS = 120_000
 const OPENCODE_ZEN_OFFLINE_STORAGE_KEY = "opencode-zen-offline-models"
-const BUILD_PREVIEW_EVENT = "opencode:build-preview"
+export const BUILD_PREVIEW_EVENT = "opencode:build-preview"
+export const FILE_CHANGE_EVENT = "opencode:workspace-files-changed"

 function markOpencodeZenModelOffline(modelId: string): void {
  if (typeof window === "undefined" || !modelId) return
@@ -234,6 +237,8 @@ async function checkTokenBudgetBeforeSend(
 type ExternalChatMessage = { role: "user" | "assistant" | "system"; content: string }

 const MAX_ATTACHMENT_CHARS = 8000
+const MAX_CONTEXT_MESSAGES = 100
+const MAX_MESSAGES_FOR_YIELD = 50

 function shouldForceEnglish(prompt: string): boolean {
  const text = prompt.trim()
@@ -270,6 +275,12 @@ function clampText(value: string, maxChars: number): string {
  return `${value.slice(0, Math.max(0, maxChars - 3))}...`
 }

+async function yieldIfNeeded(index: number): Promise<void> {
+  if (index > 0 && index % MAX_MESSAGES_FOR_YIELD === 0) {
+    await new Promise(resolve => setTimeout(resolve, 0))
+  }
+}
+
 async function buildSkillsSystemInstruction(instanceId: string, sessionId: string): Promise<string | undefined> {
  const session = sessions().get(instanceId)?.get(sessionId)
  const selected = session?.skills ?? []
@@ -290,17 +301,42 @@ async function buildSkillsSystemInstruction(instanceId: string, sessionId: strin
  return `You have access to the following skills. Follow their instructions when relevant.\n\n${payload}`
 }

+async function buildFileSystemContext(instanceId: string): Promise<string | undefined> {
+  try {
+    const files = await serverApi.listWorkspaceFiles(instanceId)
+    if (!files || files.length === 0) return undefined
+
+    // Sort directories first
+    const sorted = files.sort((a: any, b: any) => {
+      const aDir = a.isDirectory || a.type === "directory"
+      const bDir = b.isDirectory || b.type === "directory"
+      if (aDir === bDir) return (a.name || "").localeCompare(b.name || "")
+      return aDir ? -1 : 1
+    })
+
+    const list = sorted.map((f: any) => {
+      const isDir = f.isDirectory || f.type === "directory"
+      return isDir ? `${f.name}/` : f.name
+    }).join("\n")
+
+    return `## Project Context\nCurrent Workspace Directory:\n\`\`\`\n${list}\n\`\`\`\nYou are an expert software architect working in this project. Use standard tools to explore further.`
+  } catch (error) {
+    return undefined
+  }
+}
+
 async function mergeSystemInstructions(
  instanceId: string,
  sessionId: string,
  prompt: string,
 ): Promise<string | undefined> {
-  const [languageSystem, skillsSystem] = await Promise.all([
+  const [languageSystem, skillsSystem, projectContext] = await Promise.all([
    Promise.resolve(buildLanguageSystemInstruction(prompt)),
    buildSkillsSystemInstruction(instanceId, sessionId),
+    buildFileSystemContext(instanceId),
  ])
  const sshInstruction = buildSshPasswordInstruction(prompt)
-  const sections = [languageSystem, skillsSystem, sshInstruction].filter(Boolean) as string[]
+  const sections = [projectContext, languageSystem, skillsSystem, sshInstruction].filter(Boolean) as string[]
  if (sections.length === 0) return undefined
  return sections.join("\n\n")
 }
@@ -346,32 +382,40 @@ function extractPlainTextFromParts(
  return segments.join("\n").trim()
 }

-function buildExternalChatMessages(
+async function buildExternalChatMessages(
  instanceId: string,
  sessionId: string,
  systemMessage?: string,
-): ExternalChatMessage[] {
-  const store = messageStoreBus.getOrCreate(instanceId)
-  const messageIds = store.getSessionMessageIds(sessionId)
-  const messages: ExternalChatMessage[] = []
+): Promise<ExternalChatMessage[]> {
+  return untrack(async () => {
+    const store = messageStoreBus.getOrCreate(instanceId)
+    const messageIds = store.getSessionMessageIds(sessionId)
+    const messages: ExternalChatMessage[] = []

-  if (systemMessage) {
-    messages.push({ role: "system", content: systemMessage })
-  }
+    if (systemMessage) {
+      messages.push({ role: "system", content: systemMessage })
+    }

-  for (const messageId of messageIds) {
-    const record = store.getMessage(messageId)
-    if (!record) continue
-    const { orderedParts } = buildRecordDisplayData(instanceId, record)
-    const content = extractPlainTextFromParts(orderedParts as Array<{ type?: string; text?: unknown; filename?: string }>)
-    if (!content) continue
-    messages.push({
-      role: record.role === "assistant" ? "assistant" : "user",
-      content,
-    })
-  }
+    const limitedMessageIds = messageIds.length > MAX_CONTEXT_MESSAGES
+      ? messageIds.slice(-MAX_CONTEXT_MESSAGES)
+      : messageIds

-  return messages
+    for (let i = 0; i < limitedMessageIds.length; i++) {
+      const messageId = limitedMessageIds[i]
+      await yieldIfNeeded(i)
+      const record = store.getMessage(messageId)
+      if (!record) continue
+      const { orderedParts } = buildRecordDisplayData(instanceId, record)
+      const content = extractPlainTextFromParts(orderedParts as Array<{ type?: string; text?: unknown; filename?: string }>)
+      if (!content) continue
+      messages.push({
+        role: record.role === "assistant" ? "assistant" : "user",
+        content,
+      })
+    }
+
+    return messages
+  })
 }

 function decodeAttachmentData(data: Uint8Array): string {
@@ -391,7 +435,7 @@ async function buildExternalChatMessagesWithAttachments(
  systemMessage: string | undefined,
  attachments: Array<{ filename?: string; source?: any; mediaType?: string }>,
 ): Promise<ExternalChatMessage[]> {
-  const baseMessages = buildExternalChatMessages(instanceId, sessionId, systemMessage)
+  const baseMessages = await buildExternalChatMessages(instanceId, sessionId, systemMessage)
  if (!attachments || attachments.length === 0) {
    return baseMessages
  }
@@ -455,6 +499,8 @@ async function readSseStream(
  resetIdleTimer()

  try {
+    let chunkCount = 0
+    let lastYieldTime = performance.now()
    while (!shouldStop) {
      const { done, value } = await reader.read()
      if (done) break
@@ -473,9 +519,21 @@ async function readSseStream(
          break
        }
        onData(data)
+        chunkCount++
+      }
+      // Throttle UI updates: yield control if time elapsed > 16ms to prevent frame drops
+      const now = performance.now()
+      if (now - lastYieldTime > 16) {
+        addDebugLog(`Yielding after ${Math.round(now - lastYieldTime)}ms (chunks: ${chunkCount})`, "info")
+        lastYieldTime = now
+        if ('requestIdleCallback' in window) {
+          await new Promise<void>(resolve => {
+            requestIdleCallback(() => resolve(), { timeout: 16 })
+          })
+        } else {
+          await new Promise<void>(resolve => setTimeout(resolve, 0))
+        }
      }
-      // Yield to main thread periodically to prevent UI freeze during rapid streaming
-      await new Promise<void>(resolve => setTimeout(resolve, 0))
    }
    if (timedOut) {
      throw new Error("Stream timed out")
@@ -499,6 +557,10 @@ async function streamOllamaChat(
  const controller = new AbortController()
  const timeoutId = setTimeout(() => controller.abort(), STREAM_TIMEOUT_MS)

+  // Get workspace path for tool execution
+  const instance = instances().get(instanceId)
+  const workspacePath = instance?.folder || ""
+
  const response = await fetch("/api/ollama/chat", {
    method: "POST",
    headers: { "Content-Type": "application/json" },
@@ -507,6 +569,8 @@ async function streamOllamaChat(
      model: modelId,
      messages,
      stream: true,
+      workspacePath,
+      enableTools: true,
    }),
  })

@@ -516,54 +580,105 @@ async function streamOllamaChat(
  }

  const store = messageStoreBus.getOrCreate(instanceId)
+  store.beginStreamingUpdate()
  let fullText = ""
+  let lastUpdateAt = 0

  try {
    await readSseStream(response, (data) => {
      try {
        const chunk = JSON.parse(data)
-        // Check for error response from server
-        if (chunk?.error) {
-          throw new Error(chunk.error)
+        if (chunk?.error) throw new Error(chunk.error)
+
+        // Handle tool execution results (special events from backend)
+        if (chunk?.type === "tool_result") {
+          const toolResult = `\n\n✅ **Tool Executed:** ${chunk.content}\n\n`
+          fullText += toolResult
+          store.applyPartUpdate({
+            messageId: assistantMessageId,
+            part: { id: assistantPartId, type: "text", text: fullText } as any,
+          })
+
+          // Dispatch file change event to refresh sidebar
+          if (typeof window !== "undefined") {
+            console.log(`[EVENT] Dispatching FILE_CHANGE_EVENT for ${instanceId}`);
+            window.dispatchEvent(new CustomEvent(FILE_CHANGE_EVENT, { detail: { instanceId } }))
+          }
+
+          // Auto-trigger preview for HTML file writes
+          const content = chunk.content || ""
+          if (content.includes("Successfully wrote") &&
+            (content.includes(".html") || content.includes("index.") || content.includes(".htm"))) {
+            if (typeof window !== "undefined") {
+              const htmlMatch = content.match(/to\s+([^\s]+\.html?)/)
+              if (htmlMatch) {
+                const relativePath = htmlMatch[1]
+                const origin = typeof window !== "undefined" ? window.location.origin : "http://localhost:3000"
+                const apiOrigin = origin.replace(":3000", ":9898")
+                const previewUrl = `${apiOrigin}/api/workspaces/${instanceId}/serve/${relativePath}`
+
+                console.log(`[EVENT] Auto-preview triggered for ${previewUrl}`);
+                window.dispatchEvent(new CustomEvent(BUILD_PREVIEW_EVENT, {
+                  detail: { url: previewUrl, instanceId }
+                }))
+              }
+            }
+          }
+
+          return
        }
+
        const delta = chunk?.message?.content
        if (typeof delta !== "string" || delta.length === 0) return
        fullText += delta
-        store.applyPartUpdate({
-          messageId: assistantMessageId,
-          part: { id: assistantPartId, type: "text", text: fullText } as any,
-        })
+
+        const now = Date.now()
+        if (now - lastUpdateAt > 150) { // Limit to ~7 updates per second
+          lastUpdateAt = now
+          store.applyPartUpdate({
+            messageId: assistantMessageId,
+            part: { id: assistantPartId, type: "text", text: fullText } as any,
+          })
+        }
      } catch (e) {
        if (e instanceof Error) throw e
-        // Ignore malformed chunks
      }
    })
+
+    // Always apply final text update
+    store.applyPartUpdate({
+      messageId: assistantMessageId,
+      part: { id: assistantPartId, type: "text", text: fullText } as any,
+    })
  } finally {
    clearTimeout(timeoutId)
+    store.endStreamingUpdate()
  }

-  store.upsertMessage({
-    id: assistantMessageId,
-    sessionId,
-    role: "assistant",
-    status: "complete",
-    updatedAt: Date.now(),
-    isEphemeral: false,
-  })
-  store.setMessageInfo(assistantMessageId, {
-    id: assistantMessageId,
-    role: "assistant",
-    providerID: providerId,
-    modelID: modelId,
-    time: { created: store.getMessageInfo(assistantMessageId)?.time?.created ?? Date.now(), completed: Date.now() },
-  } as any)
-  store.upsertMessage({
-    id: messageId,
-    sessionId,
-    role: "user",
-    status: "sent",
-    updatedAt: Date.now(),
-    isEphemeral: false,
+  batch(() => {
+    store.upsertMessage({
+      id: assistantMessageId,
+      sessionId,
+      role: "assistant",
+      status: "complete",
+      updatedAt: Date.now(),
+      isEphemeral: false,
+    })
+    store.setMessageInfo(assistantMessageId, {
+      id: assistantMessageId,
+      role: "assistant",
+      providerID: providerId,
+      modelID: modelId,
+      time: { created: store.getMessageInfo(assistantMessageId)?.time?.created ?? Date.now(), completed: Date.now() },
+    } as any)
+    store.upsertMessage({
+      id: messageId,
+      sessionId,
+      role: "user",
+      status: "sent",
+      updatedAt: Date.now(),
+      isEphemeral: false,
+    })
  })
 }

@@ -582,6 +697,10 @@ async function streamQwenChat(
  const controller = new AbortController()
  const timeoutId = setTimeout(() => controller.abort(), STREAM_TIMEOUT_MS)

+  // Get workspace path for tool execution
+  const instance = instances().get(instanceId)
+  const workspacePath = instance?.folder || ""
+
  const response = await fetch("/api/qwen/chat", {
    method: "POST",
    headers: {
@@ -594,6 +713,8 @@ async function streamQwenChat(
      messages,
      stream: true,
      resource_url: resourceUrl,
+      workspacePath,
+      enableTools: true,
    }),
  })

@@ -603,27 +724,86 @@ async function streamQwenChat(
  }

  const store = messageStoreBus.getOrCreate(instanceId)
+  store.beginStreamingUpdate()
  let fullText = ""
+  let lastUpdateAt = 0

  try {
    await readSseStream(response, (data) => {
      try {
        const chunk = JSON.parse(data)
+
+        // Handle tool execution results
+        if (chunk?.type === "tool_result") {
+          const toolResult = `\n\n✅ **Tool Executed:** ${chunk.content}\n\n`
+          fullText += toolResult
+          store.applyPartUpdate({
+            messageId: assistantMessageId,
+            part: { id: assistantPartId, type: "text", text: fullText } as any,
+          })
+
+          // Dispatch file change event to refresh sidebar
+          if (typeof window !== "undefined") {
+            console.log(`[Qwen] Dispatching FILE_CHANGE_EVENT for ${instanceId}`);
+            console.log(`[EVENT] Dispatching FILE_CHANGE_EVENT for ${instanceId}`);
+            window.dispatchEvent(new CustomEvent(FILE_CHANGE_EVENT, { detail: { instanceId } }));
+
+            // Double-tap refresh after 1s to catch FS latency
+            setTimeout(() => {
+              window.dispatchEvent(new CustomEvent(FILE_CHANGE_EVENT, { detail: { instanceId } }));
+            }, 1000);
+          }
+
+          // Auto-trigger preview for HTML file writes
+          const content = chunk.content || ""
+          if (content.includes("Successfully wrote") &&
+            (content.includes(".html") || content.includes("index.") || content.includes(".htm"))) {
+            if (typeof window !== "undefined") {
+              const htmlMatch = content.match(/to\s+([^\s]+\.html?)/)
+              if (htmlMatch) {
+                const relativePath = htmlMatch[1]
+                const origin = typeof window !== "undefined" ? window.location.origin : "http://localhost:3000"
+                const apiOrigin = origin.replace(":3000", ":9898")
+                const previewUrl = `${apiOrigin}/api/workspaces/${instanceId}/serve/${relativePath}`
+
+                console.log(`[Qwen] Auto-preview triggered for ${relativePath}`);
+                console.log(`[EVENT] Auto-preview triggered for ${previewUrl}`);
+                window.dispatchEvent(new CustomEvent(BUILD_PREVIEW_EVENT, {
+                  detail: { url: previewUrl, instanceId }
+                }))
+              }
+            }
+          }
+          return
+        }
+
        const delta =
          chunk?.choices?.[0]?.delta?.content ??
          chunk?.choices?.[0]?.message?.content
        if (typeof delta !== "string" || delta.length === 0) return
        fullText += delta
-        store.applyPartUpdate({
-          messageId: assistantMessageId,
-          part: { id: assistantPartId, type: "text", text: fullText } as any,
-        })
+
+        const now = Date.now()
+        if (now - lastUpdateAt > 40) { // Limit to ~25 updates per second
+          lastUpdateAt = now
+          store.applyPartUpdate({
+            messageId: assistantMessageId,
+            part: { id: assistantPartId, type: "text", text: fullText } as any,
+          })
+        }
      } catch {
        // Ignore malformed chunks
      }
    })
+
+    // Always apply final text update
+    store.applyPartUpdate({
+      messageId: assistantMessageId,
+      part: { id: assistantPartId, type: "text", text: fullText } as any,
+    })
  } finally {
    clearTimeout(timeoutId)
+    store.endStreamingUpdate()
  }

  store.upsertMessage({
@@ -664,6 +844,10 @@ async function streamOpenCodeZenChat(
  const controller = new AbortController()
  const timeoutId = setTimeout(() => controller.abort(), STREAM_TIMEOUT_MS)

+  // Get workspace path for tool execution
+  const instance = instances().get(instanceId)
+  const workspacePath = instance?.folder || ""
+
  const response = await fetch("/api/opencode-zen/chat", {
    method: "POST",
    headers: { "Content-Type": "application/json" },
@@ -672,6 +856,8 @@ async function streamOpenCodeZenChat(
      model: modelId,
      messages,
      stream: true,
+      workspacePath,
+      enableTools: true,
    }),
  })

@@ -681,7 +867,9 @@ async function streamOpenCodeZenChat(
  }

  const store = messageStoreBus.getOrCreate(instanceId)
+  store.beginStreamingUpdate()
  let fullText = ""
+  let lastUpdateAt = 0

  try {
    await readSseStream(response, (data) => {
@@ -690,23 +878,78 @@ async function streamOpenCodeZenChat(
        if (chunk?.error) {
          throw new Error(typeof chunk.error === "string" ? chunk.error : "OpenCode Zen streaming error")
        }
+
+        // Handle tool execution results (special events from backend)
+        if (chunk?.type === "tool_result") {
+          const toolResult = `\n\n✅ **Tool Executed:** ${chunk.content}\n\n`
+          fullText += toolResult
+          store.applyPartUpdate({
+            messageId: assistantMessageId,
+            part: { id: assistantPartId, type: "text", text: fullText } as any,
+          })
+
+          // Dispatch file change event to refresh sidebar
+          if (typeof window !== "undefined") {
+            console.log(`[Ollama] Dispatching FILE_CHANGE_EVENT for ${instanceId}`);
+            console.log(`[EVENT] Dispatching FILE_CHANGE_EVENT for ${instanceId}`);
+            window.dispatchEvent(new CustomEvent(FILE_CHANGE_EVENT, { detail: { instanceId } }))
+          }
+
+          // Auto-trigger preview for HTML file writes
+          const content = chunk.content || ""
+          if (content.includes("Successfully wrote") &&
+            (content.includes(".html") || content.includes("index.") || content.includes(".htm"))) {
+            if (typeof window !== "undefined") {
+              const htmlMatch = content.match(/to\s+([^\s]+\.html?)/)
+              if (htmlMatch) {
+                const relativePath = htmlMatch[1]
+                // USE PROXY URL instead of file:// to avoid "Not allowed to load local resource"
+                // The backend (port 9898) serves workspace files via /api/workspaces/:id/serve
+                const origin = typeof window !== "undefined" ? window.location.origin : "http://localhost:3000"
+                const apiOrigin = origin.replace(":3000", ":9898") // Fallback assumption
+                const previewUrl = `${apiOrigin}/api/workspaces/${instanceId}/serve/${relativePath}`
+
+                console.log(`[Ollama] Auto-preview triggered for ${relativePath}`);
+                console.log(`[EVENT] Auto-preview triggered for ${previewUrl}`);
+                window.dispatchEvent(new CustomEvent(BUILD_PREVIEW_EVENT, {
+                  detail: { url: previewUrl, instanceId }
+                }))
+              }
+            }
+          }
+
+          return
+        }
+
        const delta =
          chunk?.choices?.[0]?.delta?.content ??
          chunk?.choices?.[0]?.message?.content
        if (typeof delta !== "string" || delta.length === 0) return
        fullText += delta
-        store.applyPartUpdate({
-          messageId: assistantMessageId,
-          part: { id: assistantPartId, type: "text", text: fullText } as any,
-        })
+
+        const now = Date.now()
+        if (now - lastUpdateAt > 40) { // Limit to ~25 updates per second
+          lastUpdateAt = now
+          store.applyPartUpdate({
+            messageId: assistantMessageId,
+            part: { id: assistantPartId, type: "text", text: fullText } as any,
+          })
+        }
      } catch (error) {
        if (error instanceof Error) {
          throw error
        }
      }
    })
+
+    // Always apply final text update
+    store.applyPartUpdate({
+      messageId: assistantMessageId,
+      part: { id: assistantPartId, type: "text", text: fullText } as any,
+    })
  } finally {
    clearTimeout(timeoutId)
+    store.endStreamingUpdate()
  }


@@ -748,6 +991,10 @@ async function streamZAIChat(
  const controller = new AbortController()
  const timeoutId = setTimeout(() => controller.abort(), STREAM_TIMEOUT_MS)

+  // Get workspace path for tool execution
+  const instance = instances().get(instanceId)
+  const workspacePath = instance?.folder || ""
+
  const response = await fetch("/api/zai/chat", {
    method: "POST",
    headers: { "Content-Type": "application/json" },
@@ -756,6 +1003,8 @@ async function streamZAIChat(
      model: modelId,
      messages,
      stream: true,
+      workspacePath,
+      enableTools: true,
    }),
  })

@@ -765,32 +1014,81 @@ async function streamZAIChat(
  }

  const store = messageStoreBus.getOrCreate(instanceId)
+  store.beginStreamingUpdate()
  let fullText = ""
+  let lastUpdateAt = 0

  try {
    await readSseStream(response, (data) => {
      try {
        const chunk = JSON.parse(data)
-        // Check for error response from server
-        if (chunk?.error) {
-          throw new Error(chunk.error)
+        if (chunk?.error) throw new Error(chunk.error)
+
+        // Handle tool execution results (special events from backend)
+        if (chunk?.type === "tool_result") {
+          const toolResult = `\n\n✅ **Tool Executed:** ${chunk.content}\n\n`
+          fullText += toolResult
+          store.applyPartUpdate({
+            messageId: assistantMessageId,
+            part: { id: assistantPartId, type: "text", text: fullText } as any,
+          })
+
+          // Dispatch file change event to refresh sidebar
+          if (typeof window !== "undefined") {
+            console.log(`[EVENT] Dispatching FILE_CHANGE_EVENT for ${instanceId}`);
+            window.dispatchEvent(new CustomEvent(FILE_CHANGE_EVENT, { detail: { instanceId } }))
+          }
+
+          // Auto-trigger preview for HTML file writes
+          const content = chunk.content || ""
+          if (content.includes("Successfully wrote") &&
+            (content.includes(".html") || content.includes("index.") || content.includes(".htm"))) {
+            if (typeof window !== "undefined") {
+              const htmlMatch = content.match(/to\s+([^\s]+\.html?)/)
+              if (htmlMatch) {
+                const relativePath = htmlMatch[1]
+                const origin = typeof window !== "undefined" ? window.location.origin : "http://localhost:3000"
+                const apiOrigin = origin.replace(":3000", ":9898")
+                const previewUrl = `${apiOrigin}/api/workspaces/${instanceId}/serve/${relativePath}`
+
+                console.log(`[EVENT] Auto-preview triggered for ${previewUrl}`);
+                window.dispatchEvent(new CustomEvent(BUILD_PREVIEW_EVENT, {
+                  detail: { url: previewUrl, instanceId }
+                }))
+              }
+            }
+          }
+
+          return
        }
+
        const delta =
          chunk?.choices?.[0]?.delta?.content ??
          chunk?.choices?.[0]?.message?.content
        if (typeof delta !== "string" || delta.length === 0) return
        fullText += delta
-        store.applyPartUpdate({
-          messageId: assistantMessageId,
-          part: { id: assistantPartId, type: "text", text: fullText } as any,
-        })
+
+        const now = Date.now()
+        if (now - lastUpdateAt > 40) { // Limit to ~25 updates per second
+          lastUpdateAt = now
+          store.applyPartUpdate({
+            messageId: assistantMessageId,
+            part: { id: assistantPartId, type: "text", text: fullText } as any,
+          })
+        }
      } catch (e) {
        if (e instanceof Error) throw e
-        // Ignore malformed chunks
      }
    })
+
+    // Always apply final text update
+    store.applyPartUpdate({
+      messageId: assistantMessageId,
+      part: { id: assistantPartId, type: "text", text: fullText } as any,
+    })
  } finally {
    clearTimeout(timeoutId)
+    store.endStreamingUpdate()
  }

  store.upsertMessage({
@@ -941,15 +1239,17 @@ async function sendMessage(

  log.info("sendMessage: upserting optimistic message", { messageId, sessionId, taskId });

-  store.upsertMessage({
-    id: messageId,
-    sessionId,
-    role: "user",
-    status: "sending",
-    parts: optimisticParts,
-    createdAt,
-    updatedAt: createdAt,
-    isEphemeral: true,
+  untrack(() => {
+    store.upsertMessage({
+      id: messageId,
+      sessionId,
+      role: "user",
+      status: "sending",
+      parts: optimisticParts,
+      createdAt,
+      updatedAt: createdAt,
+      isEphemeral: true,
+    })
  })

  withSession(instanceId, sessionId, () => {
@@ -957,47 +1257,62 @@ async function sendMessage(
  })

  const providerId = effectiveModel.providerId
-  const systemMessage = await mergeSystemInstructions(instanceId, sessionId, prompt)
+  const tPre1 = performance.now()
+  const systemMessage = await untrack(() => mergeSystemInstructions(instanceId, sessionId, prompt))
+  const tPre2 = performance.now()
+  if (tPre2 - tPre1 > 10) {
+    addDebugLog(`Merge System Instructions: ${Math.round(tPre2 - tPre1)}ms`, "warn")
+  }
+
  if (providerId === "ollama-cloud" || providerId === "qwen-oauth" || providerId === "opencode-zen" || providerId === "zai") {
    const store = messageStoreBus.getOrCreate(instanceId)
    const now = Date.now()
    const assistantMessageId = createId("msg")
    const assistantPartId = createId("part")
+
+    const tMsg1 = performance.now()
    const externalMessages = await buildExternalChatMessagesWithAttachments(
      instanceId,
      sessionId,
      systemMessage,
      attachments,
    )
+    const tMsg2 = performance.now()
+    if (tMsg2 - tMsg1 > 10) {
+      addDebugLog(`Build External Messages: ${Math.round(tMsg2 - tMsg1)}ms`, "warn")
+    }

-    store.upsertMessage({
-      id: assistantMessageId,
-      sessionId,
-      role: "assistant",
-      status: "streaming",
-      parts: [{ id: assistantPartId, type: "text", text: "" } as any],
-      createdAt: now,
-      updatedAt: now,
-      isEphemeral: true,
-    })
-    store.setMessageInfo(assistantMessageId, {
-      id: assistantMessageId,
-      role: "assistant",
-      providerID: effectiveModel.providerId,
-      modelID: effectiveModel.modelId,
-      time: { created: now, completed: 0 },
-    } as any)
-    store.upsertMessage({
-      id: messageId,
-      sessionId,
-      role: "user",
-      status: "sent",
-      updatedAt: now,
-      isEphemeral: false,
+    untrack(() => {
+      store.upsertMessage({
+        id: assistantMessageId,
+        sessionId,
+        role: "assistant",
+        status: "streaming",
+        parts: [{ id: assistantPartId, type: "text", text: "" } as any],
+        createdAt: now,
+        updatedAt: now,
+        isEphemeral: true,
+      })
+      store.setMessageInfo(assistantMessageId, {
+        id: assistantMessageId,
+        role: "assistant",
+        providerID: effectiveModel.providerId,
+        modelID: effectiveModel.modelId,
+        time: { created: now, completed: 0 },
+      } as any)
+      store.upsertMessage({
+        id: messageId,
+        sessionId,
+        role: "user",
+        status: "sent",
+        updatedAt: now,
+        isEphemeral: false,
+      })
    })

    try {
      if (providerId === "ollama-cloud") {
+        const tStream1 = performance.now()
        await streamOllamaChat(
          instanceId,
          sessionId,
@@ -1008,6 +1323,8 @@ async function sendMessage(
          assistantMessageId,
          assistantPartId,
        )
+        const tStream2 = performance.now()
+        addDebugLog(`Stream Complete: ${Math.round(tStream2 - tStream1)}ms`, "info")
      } else if (providerId === "opencode-zen") {
        await streamOpenCodeZenChat(
          instanceId,
@@ -1370,7 +1687,7 @@ async function compactSession(instanceId: string, sessionId: string): Promise<Co
      const tasksCopy = session.tasks.map((task) => ({ ...task }))
      withSession(instanceId, compactedSession.id, (nextSession) => {
        nextSession.tasks = tasksCopy
-        nextSession.activeTaskId = undefined
+        nextSession.activeTaskId = session.activeTaskId
      })
    }

@@ -1632,6 +1949,48 @@ async function forkSession(instanceId: string, sessionId: string): Promise<strin
  }
 }

+// Forcefully reset streaming state to unlock UI if stuck
+function forceReset() {
+  const store = messageStoreBus.getOrCreate(activeInstanceId() || "")
+  if (!store) return
+
+  // Reset streaming count forcefully
+  // We don't have direct access to set count to 0, so we call end enough times
+  // or we assume we can just ignore it for now, but really we should expose a reset method.
+  // For now, let's just log and clear pending parts.
+  store.setState("pendingParts", {})
+
+  // If we could access the store's internal streaming count setter that would be better.
+  // Since we added `isStreaming` and `endStreamingUpdate` to store interface,
+  // we can just call end multiple times if we suspect it's stuck > 0
+  let safety = 0
+  while (store.state.streamingUpdateCount > 0 && safety < 100) {
+    store.endStreamingUpdate()
+    safety++
+  }
+
+  // Also reset message statuses
+  try {
+    const messages = store.state.messages;
+    Object.values(messages).forEach(msg => {
+      if (msg.status === "streaming" || msg.status === "sending") {
+        store.upsertMessage({
+          id: msg.id,
+          sessionId: msg.sessionId,
+          role: msg.role,
+          status: "interrupted",
+          updatedAt: Date.now(),
+          isEphemeral: msg.isEphemeral,
+        })
+      }
+    })
+  } catch (e) {
+    console.error("Error updating message status during reset", e)
+  }
+
+  addDebugLog("Force Reset Triggered: Cleared streaming state & statuses", "warn")
+}
+
 export {
  abortSession,
  compactSession,
@@ -1644,4 +2003,5 @@ export {
  updateSessionAgent,
  updateSessionModel,
  updateSessionModelForSession,
+  forceReset, // Add to exports
 }
--- a/packages/ui/src/stores/session-events.ts
+++ b/packages/ui/src/stores/session-events.ts
@@ -18,7 +18,7 @@ import type { MessageStatus } from "./message-v2/types"
 import { getLogger } from "../lib/logger"
 import { showToastNotification, ToastVariant } from "../lib/notifications"
 import { instances, addPermissionToQueue, removePermissionFromQueue, sendPermissionResponse } from "./instances"
-import { getSoloState, incrementStep, popFromTaskQueue, setActiveTaskId } from "./solo-store"
+import { getSoloState, incrementStep, popFromTaskQueue, setActiveTaskId, canPerformAutonomousAction, recordAutonomousAction, resetErrorRecovery, clearContinuationFlag } from "./solo-store"
 import { sendMessage, consumeTokenWarningSuppression, consumeCompactionSuppression, updateSessionModel } from "./session-actions"
 import { showAlertDialog } from "./alerts"
 import { sessions, setSessions, withSession } from "./session-state"
@@ -175,11 +175,21 @@ function handleMessageUpdate(instanceId: string, event: MessageUpdateEvent | Mes
    // Auto-correction logic for SOLO
    const solo = getSoloState(instanceId)
    if (hasError && solo.isAutonomous && solo.currentStep < solo.maxSteps) {
-      log.info(`[SOLO] Error detected in autonomous mode, prompting for fix: ${messageId}`)
      const errorMessage = (info as any).error?.message || "Unknown error"
+
+      // Check if we can perform autonomous error recovery (loop prevention)
+      if (!canPerformAutonomousAction(instanceId, "error_recovery")) {
+        log.warn("[SOLO] Error recovery blocked by loop prevention", { instanceId, sessionId, errorMessage })
+        return
+      }
+
+      log.info(`[SOLO] Error detected in autonomous mode, prompting for fix: ${messageId}`)
      incrementStep(instanceId)
+      recordAutonomousAction(instanceId, "error_recovery", errorMessage)
+
      sendMessage(instanceId, sessionId, `The previous step failed with error: ${errorMessage}. Please analyze the error and try a different approach.`, [], solo.activeTaskId || undefined).catch((err) => {
        log.error("[SOLO] Failed to send error correction message", err)
+        resetErrorRecovery(instanceId)
      })
    }

@@ -338,10 +348,17 @@ function handleSessionIdle(instanceId: string, event: EventSessionIdle): void {
    const session = instanceSessions?.get(sessionId)
    if (!session) return

-    // If there's an active task, we might want to prompt the agent to continue or check progress
+    // If there's an active task, we might want to prompt to agent to continue or check progress
+    if (!canPerformAutonomousAction(instanceId, "idle_continuation")) {
+      log.warn("[SOLO] Idle continuation blocked by loop prevention", { instanceId, sessionId })
+      clearContinuationFlag(instanceId)
+      return
+    }
+
    if (solo.activeTaskId) {
      log.info(`[SOLO] Session idle in autonomous mode, prompting continuation for task: ${solo.activeTaskId}`)
      incrementStep(instanceId)
+      recordAutonomousAction(instanceId, "idle_continuation")
      sendMessage(instanceId, sessionId, "Continue", [], solo.activeTaskId).catch((err) => {
        log.error("[SOLO] Failed to send continuation message", err)
      })
@@ -363,6 +380,7 @@ function handleSessionIdle(instanceId: string, event: EventSessionIdle): void {
        }

        setActiveTaskId(instanceId, nextTaskId)
+        recordAutonomousAction(instanceId, "idle_continuation")
        sendMessage(instanceId, sessionId, taskTitle, [], nextTaskId).catch((err) => {
          log.error("[SOLO] Failed to start next task", err)
        })
@@ -435,10 +453,19 @@ function handleSessionError(instanceId: string, event: EventSessionError): void
  const sessionId = (event.properties as any)?.sessionID

  if (solo.isAutonomous && sessionId && solo.currentStep < solo.maxSteps) {
+    const errorMessage = `I encountered an error: "${message}". Please analyze the cause and provide a fix.`
+
+    if (!canPerformAutonomousAction(instanceId, "error_recovery")) {
+      log.warn("[SOLO] Error recovery blocked by loop prevention", { instanceId, sessionId, message })
+      return
+    }
+
    log.info(`[SOLO] Session error in autonomous mode, prompting fix: ${message}`)
    incrementStep(instanceId)
-    sendMessage(instanceId, sessionId, `I encountered an error: "${message}". Please analyze the cause and provide a fix.`, [], solo.activeTaskId || undefined).catch((err) => {
+    recordAutonomousAction(instanceId, "error_recovery", message)
+    sendMessage(instanceId, sessionId, errorMessage, [], solo.activeTaskId || undefined).catch((err) => {
      log.error("[SOLO] Failed to send error recovery message", err)
+      resetErrorRecovery(instanceId)
    })
    return
  }
--- a/packages/ui/src/stores/session-state.ts
+++ b/packages/ui/src/stores/session-state.ts
@@ -154,8 +154,21 @@ function withSession(instanceId: string, sessionId: string, updater: (session: S
    return next
  })

-  // Persist session tasks to storage
-  persistSessionTasks(instanceId)
+  // Persist session tasks to storage (DEBOUNCED)
+  schedulePersist(instanceId)
+}
+
+// Debounce map for persistence
+const persistTimers = new Map<string, ReturnType<typeof setTimeout>>()
+
+function schedulePersist(instanceId: string) {
+  const existing = persistTimers.get(instanceId)
+  if (existing) clearTimeout(existing)
+  const timer = setTimeout(() => {
+    persistTimers.delete(instanceId)
+    persistSessionTasks(instanceId)
+  }, 2000)
+  persistTimers.set(instanceId, timer)
 }

 async function persistSessionTasks(instanceId: string) {
--- a/packages/ui/src/stores/solo-store.ts
+++ b/packages/ui/src/stores/solo-store.ts
@@ -11,6 +11,11 @@ export interface SoloState {
  currentStep: number
  activeTaskId: string | null
  taskQueue: string[]
+  // Loop prevention fields
+  lastActionTimestamp: number
+  consecutiveErrorCount: number
+  lastErrorHash: string
+  isContinuationFromIdle: boolean
 }

 const [soloStates, setSoloStates] = createSignal<Map<string, SoloState>>(new Map())
@@ -26,6 +31,10 @@ export function getSoloState(instanceId: string): SoloState {
      currentStep: 0,
      activeTaskId: null,
      taskQueue: [],
+      lastActionTimestamp: 0,
+      consecutiveErrorCount: 0,
+      lastErrorHash: "",
+      isContinuationFromIdle: false,
    }
  }
  return state
@@ -83,3 +92,75 @@ export function popFromTaskQueue(instanceId: string): string | null {
  setSoloState(instanceId, { taskQueue: rest })
  return next
 }
+
+function computeErrorHash(error: string): string {
+  const normalized = error.toLowerCase().replace(/\d+/g, "X").replace(/\s+/g, " ")
+  return normalized.slice(0, 100)
+}
+
+const COOLDOWN_MS = 3000
+const MAX_CONSECUTIVE_ERRORS = 3
+
+export function canPerformAutonomousAction(instanceId: string, actionType: "error_recovery" | "idle_continuation"): boolean {
+  const state = getSoloState(instanceId)
+  const now = Date.now()
+
+  if (actionType === "error_recovery") {
+    if (state.consecutiveErrorCount >= MAX_CONSECUTIVE_ERRORS) {
+      log.warn("Maximum consecutive errors reached, stopping autonomous error recovery", { instanceId, count: state.consecutiveErrorCount })
+      return false
+    }
+  }
+
+  if (actionType === "idle_continuation" && state.isContinuationFromIdle) {
+    log.warn("Already continuing from idle, preventing double continuation", { instanceId })
+    return false
+  }
+
+  const timeSinceLastAction = now - state.lastActionTimestamp
+  if (timeSinceLastAction < COOLDOWN_MS && state.lastActionTimestamp > 0) {
+    log.warn("Cooldown period active, delaying autonomous action", { instanceId, timeSinceLastAction })
+    return false
+  }
+
+  return true
+}
+
+export function recordAutonomousAction(instanceId: string, actionType: "error_recovery" | "idle_continuation", errorMessage?: string): void {
+  const state = getSoloState(instanceId)
+  const now = Date.now()
+
+  if (actionType === "error_recovery" && errorMessage) {
+    const errorHash = computeErrorHash(errorMessage)
+    const newErrorCount = errorHash === state.lastErrorHash ? state.consecutiveErrorCount + 1 : 1
+
+    setSoloState(instanceId, {
+      lastActionTimestamp: now,
+      consecutiveErrorCount: newErrorCount,
+      lastErrorHash: errorHash,
+    })
+  } else if (actionType === "idle_continuation") {
+    setSoloState(instanceId, {
+      lastActionTimestamp: now,
+      isContinuationFromIdle: true,
+    })
+  } else {
+    setSoloState(instanceId, {
+      lastActionTimestamp: now,
+    })
+  }
+}
+
+export function clearContinuationFlag(instanceId: string): void {
+  const state = getSoloState(instanceId)
+  if (state.isContinuationFromIdle) {
+    setSoloState(instanceId, { isContinuationFromIdle: false })
+  }
+}
+
+export function resetErrorRecovery(instanceId: string): void {
+  setSoloState(instanceId, {
+    consecutiveErrorCount: 0,
+    lastErrorHash: "",
+  })
+}
--- a/packages/ui/src/stores/task-actions.ts
+++ b/packages/ui/src/stores/task-actions.ts
@@ -16,14 +16,14 @@ export async function addTask(
  title: string
 ): Promise<{ id: string; taskSessionId?: string }> {
  const id = nanoid()
-  console.log("[task-actions] addTask started", { instanceId, sessionId, title, taskId: id });
+  // console.log("[task-actions] addTask started", { instanceId, sessionId, title, taskId: id });
  
  let taskSessionId: string | undefined
  const parentSession = sessions().get(instanceId)?.get(sessionId)
  const parentAgent = parentSession?.agent || ""
  const parentModel = parentSession?.model
  try {
-    console.log("[task-actions] creating new task session...");
+    // console.log("[task-actions] creating new task session...");
    const created = await createSession(instanceId, parentAgent || undefined, { skipAutoCleanup: true })
    taskSessionId = created.id
    withSession(instanceId, taskSessionId, (taskSession) => {
@@ -35,7 +35,7 @@ export async function addTask(
        taskSession.model = { ...parentModel }
      }
    })
-    console.log("[task-actions] task session created", { taskSessionId });
+    // console.log("[task-actions] task session created", { taskSessionId });
  } catch (error) {
    console.error("[task-actions] Failed to create session for task", error)
    showToastNotification({
@@ -62,7 +62,7 @@ export async function addTask(
      session.tasks = []
    }
    session.tasks = [newTask, ...session.tasks]
-    console.log("[task-actions] task added to session", { taskCount: session.tasks.length });
+    // console.log("[task-actions] task added to session", { taskCount: session.tasks.length });
  })

  return { id, taskSessionId }
@@ -74,7 +74,7 @@ export function addTaskMessage(
  taskId: string,
  messageId: string,
 ): void {
-  console.log("[task-actions] addTaskMessage called", { instanceId, sessionId, taskId, messageId });
+  // console.log("[task-actions] addTaskMessage called", { instanceId, sessionId, taskId, messageId });
  withSession(instanceId, sessionId, (session) => {
    let targetSessionId = sessionId
    let targetTaskId = taskId
@@ -82,7 +82,7 @@ export function addTaskMessage(
    // If this is a child session, the tasks are on the parent
    if (session.parentId && !session.tasks) {
      targetSessionId = session.parentId
-      console.log("[task-actions] task session detected, targeting parent", { parentId: session.parentId });
+      // console.log("[task-actions] task session detected, targeting parent", { parentId: session.parentId });
    }

    withSession(instanceId, targetSessionId, (targetSession) => {
@@ -105,9 +105,9 @@ export function addTaskMessage(
          updatedTasks[taskIndex] = updatedTask
          targetSession.tasks = updatedTasks
          
-          console.log("[task-actions] message ID added to task with reactivity", { taskId: task.id, messageCount: messageIds.length });
+          // console.log("[task-actions] message ID added to task with reactivity", { taskId: task.id, messageCount: messageIds.length });
        } else {
-          console.log("[task-actions] message ID already in task", { taskId: task.id });
+          // console.log("[task-actions] message ID already in task", { taskId: task.id });
        }
      } else {
        console.warn("[task-actions] task not found in session", { targetTaskId, sessionId, availableTaskCount: targetSession.tasks.length });
--- a/packages/ui/src/styles/antigravity.css
+++ b/packages/ui/src/styles/antigravity.css
@@ -1,4 +1,3 @@
-
 /* Antigravity Glass Effect */
 .glass {
  background: rgba(255, 255, 255, 0.03);
@@ -58,5 +57,34 @@

 /* MultiX Branding */
 .multix-badge {
-  @apply flex items-center bg-blue-500/10 border border-blue-500/20 rounded-md px-2 py-0.5 shadow-[0_0_15px_rgba(59,130,246,0.1)];
+  display: flex;
+  align-items: center;
+  background-color: rgba(59, 130, 246, 0.1);
+  border: 1px solid rgba(59, 130, 246, 0.2);
+  border-radius: 0.375rem;
+  padding: 0.125rem 0.5rem;
+  box-shadow: 0 0 15px rgba(59, 130, 246, 0.1);
+}
+
+/* Smart Fix Glowing Animation */
+@keyframes smart-fix-glow {
+  0% {
+    box-shadow: 0 0 8px rgba(34, 197, 94, 0.4), inset 0 0 4px rgba(34, 197, 94, 0.2);
+    border-color: rgba(34, 197, 94, 0.5);
+  }
+
+  50% {
+    box-shadow: 0 0 20px rgba(249, 115, 22, 0.7), inset 0 0 8px rgba(249, 115, 22, 0.3);
+    border-color: rgba(249, 115, 22, 0.8);
+  }
+
+  100% {
+    box-shadow: 0 0 8px rgba(34, 197, 94, 0.4), inset 0 0 4px rgba(34, 197, 94, 0.2);
+    border-color: rgba(34, 197, 94, 0.5);
+  }
+}
+
+.smart-fix-highlight {
+  animation: smart-fix-glow 3s infinite ease-in-out !important;
+  background: rgba(34, 197, 94, 0.08) !important;
 }
--- a/packages/ui/src/styles/markdown.css
+++ b/packages/ui/src/styles/markdown.css
@@ -227,10 +227,31 @@
    height: 14px;
  }

-  .code-block-copy .copy-text {
+  .code-block-copy .copy-text,
+  .code-block-preview .preview-text {
    font-family: var(--font-family-mono);
  }

+  .code-block-preview {
+    display: flex;
+    align-items: center;
+    gap: 4px;
+    padding: 3px 8px;
+    background-color: rgba(16, 185, 129, 0.1);
+    border: 1px solid rgba(16, 185, 129, 0.2);
+    border-radius: 4px;
+    cursor: pointer;
+    color: #10b981;
+    transition: all 150ms ease;
+    font-size: 10px;
+    font-weight: 600;
+  }
+
+  .code-block-preview:hover {
+    background-color: rgba(16, 185, 129, 0.2);
+    border-color: rgba(16, 185, 129, 0.4);
+  }
+
  .markdown-code-block pre {
    margin: 0 !important;
    padding: 12px !important;
--- a/packages/ui/src/types/context-engine.ts
+++ b/packages/ui/src/types/context-engine.ts
@@ -0,0 +1,188 @@
+/**
+ * Context Engine Types
+ * 
+ * Type definitions for Context-Engine integration
+ * Based on: https://github.com/Eskapeum/Context-Engine
+ */
+
+// ============================================================================
+// PARSER TYPES
+// ============================================================================
+
+export interface CodeSymbol {
+    name: string;
+    kind: SymbolKind;
+    filePath: string;
+    startLine: number;
+    endLine: number;
+    signature?: string;
+    documentation?: string;
+    parent?: string;
+    children?: string[];
+}
+
+export type SymbolKind =
+    | "class"
+    | "interface"
+    | "function"
+    | "method"
+    | "property"
+    | "variable"
+    | "constant"
+    | "enum"
+    | "type"
+    | "module"
+    | "namespace";
+
+export interface FileIndex {
+    path: string;
+    language: string;
+    hash: string;
+    lastModified: number;
+    symbols: CodeSymbol[];
+    imports: FileImport[];
+    exports: string[];
+}
+
+export interface FileImport {
+    source: string;
+    specifiers: string[];
+    isDefault: boolean;
+    isNamespace: boolean;
+}
+
+// ============================================================================
+// RETRIEVAL TYPES
+// ============================================================================
+
+export interface RetrievalQuery {
+    text: string;
+    maxResults?: number;
+    maxTokens?: number;
+    filters?: RetrievalFilters;
+}
+
+export interface RetrievalFilters {
+    languages?: string[];
+    paths?: string[];
+    symbolKinds?: SymbolKind[];
+    excludePaths?: string[];
+}
+
+export interface RetrievalResult {
+    content: string;
+    score: number;
+    source: RetrievalSource;
+    tokens: number;
+}
+
+export interface RetrievalSource {
+    type: "code" | "documentation" | "memory";
+    file?: string;
+    line?: number;
+    symbol?: string;
+}
+
+// ============================================================================
+// INDEXER TYPES
+// ============================================================================
+
+export interface IndexerConfig {
+    projectRoot: string;
+    languages?: string[];
+    excludePatterns?: string[];
+    maxFileSize?: number;
+    enableGitTracking?: boolean;
+}
+
+export interface IndexStats {
+    filesIndexed: number;
+    symbolsFound: number;
+    totalTokens: number;
+    lastUpdated: number;
+    duration: number;
+}
+
+export interface IndexUpdateResult {
+    added: string[];
+    updated: string[];
+    removed: string[];
+    stats: IndexStats;
+}
+
+// ============================================================================
+// MEMORY TYPES
+// ============================================================================
+
+export interface MemoryEntry {
+    id: string;
+    question: string;
+    answer: string;
+    timestamp: number;
+    tags?: string[];
+    relevance?: number;
+}
+
+export interface MemorySearchResult {
+    entries: MemoryEntry[];
+    totalCount: number;
+}
+
+// ============================================================================
+// VECTOR STORE TYPES
+// ============================================================================
+
+export interface VectorDocument {
+    id: string;
+    content: string;
+    embedding?: number[];
+    metadata: Record<string, unknown>;
+}
+
+export interface VectorSearchResult {
+    id: string;
+    score: number;
+    content: string;
+    metadata: Record<string, unknown>;
+}
+
+export interface EmbeddingProvider {
+    name: string;
+    dimensions: number;
+    embed(texts: string[]): Promise<number[][]>;
+}
+
+// ============================================================================
+// GRAPH TYPES
+// ============================================================================
+
+export interface GraphNode {
+    id: string;
+    type: NodeType;
+    name: string;
+    metadata: Record<string, unknown>;
+}
+
+export type NodeType = "file" | "symbol" | "import" | "export" | "dependency";
+
+export interface GraphEdge {
+    source: string;
+    target: string;
+    type: EdgeType;
+    weight?: number;
+}
+
+export type EdgeType =
+    | "contains"
+    | "imports"
+    | "exports"
+    | "calls"
+    | "extends"
+    | "implements"
+    | "depends_on";
+
+export interface GraphQueryResult {
+    nodes: GraphNode[];
+    edges: GraphEdge[];
+    paths?: GraphNode[][];
+}
--- a/packages/ui/vite.config.ts
+++ b/packages/ui/vite.config.ts
@@ -22,6 +22,7 @@ export default defineConfig({
  },
  server: {
    port: Number(process.env.VITE_PORT ?? 3000),
+    hmr: false, // DISABLED - HMR WebSocket was causing issues
  },
  build: {
    outDir: resolve(__dirname, "dist"),
--- a/packages/ui/vite.config.ts.timestamp-1766646632294-07ca0a8c33aa9.mjs
+++ b/packages/ui/vite.config.ts.timestamp-1766646632294-07ca0a8c33aa9.mjs
@@ -0,0 +1,43 @@
+// vite.config.ts
+import { defineConfig } from "file:///E:/TRAE%20Playground/NeuralNomadsAi/NomadArch/node_modules/vite/dist/node/index.js";
+import solid from "file:///E:/TRAE%20Playground/NeuralNomadsAi/NomadArch/node_modules/vite-plugin-solid/dist/esm/index.mjs";
+import { resolve } from "path";
+var __vite_injected_original_dirname = "E:\\TRAE Playground\\NeuralNomadsAi\\NomadArch\\packages\\ui";
+var vite_config_default = defineConfig({
+  root: "./src/renderer",
+  publicDir: resolve(__vite_injected_original_dirname, "./public"),
+  plugins: [solid()],
+  css: {
+    postcss: "./postcss.config.js"
+  },
+  resolve: {
+    alias: {
+      "@": resolve(__vite_injected_original_dirname, "./src")
+    }
+  },
+  optimizeDeps: {
+    exclude: ["lucide-solid"]
+  },
+  ssr: {
+    noExternal: ["lucide-solid"]
+  },
+  server: {
+    port: Number(process.env.VITE_PORT ?? 3e3),
+    hmr: false
+    // DISABLED - HMR WebSocket was causing issues
+  },
+  build: {
+    outDir: resolve(__vite_injected_original_dirname, "dist"),
+    chunkSizeWarningLimit: 1e3,
+    rollupOptions: {
+      input: {
+        main: resolve(__vite_injected_original_dirname, "./src/renderer/index.html"),
+        loading: resolve(__vite_injected_original_dirname, "./src/renderer/loading.html")
+      }
+    }
+  }
+});
+export {
+  vite_config_default as default
+};
+//# sourceMappingURL=data:application/json;base64,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