✨ 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
2.9 KiB
2.9 KiB
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:
- Create Task 1. Type "Write a long story about space" and hit Launch.
- While Task 1 is streaming, click the + button to create Task 2.
- Type "Write a python script for fibonacci" in Task 2 and hit Launch.
- 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:
- Go to Task 1. Select Agent: "Software Engineer" and Model: "minimax-m2".
- Go to Task 2. Select Agent: "Writer" and Model: "deepseek-chat".
- Switch back and forth.
- 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:
- Open the Agent Selector dropdown.
- Click "✨ AI Agent Generator".
- Type: "A rust expert who is sarcastic and funny".
- Click "Generate Agent".
- Review the generated name, description, and system prompt.
- Click "Save & Use Agent".
- Result: The new agent is saved and immediately selected.
D. Prompt Enhancer
- Goal: strict Opus 4.5 prompt optimization.
- Steps:
- Type a simple prompt: "fix bug".
- Click the Magic Wand (✨) button in the input area.
- Result: The prompt is expanded into a professional, structured request using the active model.
E. Compaction System
- Goal: Manage context window usage.
- Steps:
- In a long chat, look for the "Compact suggested" banner at the top of the chat list.
- Click "Compact" in the banner or the header bar.
- 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.