58 Commits

• v3.11.6: Antigravity loop breakers, vision/OCR preprocessing, has_content fix, auth config error fix, install.ps1
  Roman | RyzenAdvanced · 2026-05-26 18:07:42 +04:00
  e59ef6f28a
• v3.11.5: token-aware compaction, vision filter, universal adaptive compaction, smart-continue text detection
  Roman | RyzenAdvanced · 2026-05-26 17:30:19 +04:00
  b029e7cb5e
• v3.11.0: merge cobra PR, smart-continue, hot-reload, XML extraction

  - Merge PR #5 from cobra91: concurrency semaphore, auto-continue, SO_REUSEADDR,
    proxy-stderr.log, stream diagnostics, timeout handler, restart proxy fix
  - Tool call argument normalizer, smart-continue loop, XML extraction
  - API key hot-reload with mtime tracking + /admin/ endpoints
  - GUI hot-reload on endpoint edit with upstream verification
  - Synthetic tool-results disabled (caused deepseek-v4-pro truncation)
  - Version bump 3.10.12 -> 3.11.0, rebuild .deb

  Roman | RyzenAdvanced · 2026-05-26 15:02:02 +04:00
  c3ba3286ff
• docs: update CHANGELOG + lib with all TRAE fixes (Claude guards, guardrail skip)
  Roman | RyzenAdvanced · 2026-05-26 13:05:19 +04:00
  92ac4e5b87
• v3.10.12: sticky endpoint, parallel discovery, anti-stall, smart errors, missing headers
  Roman | RyzenAdvanced · 2026-05-26 00:57:16 +04:00
  0896ba5e55
• v3.10.11: hybrid endpoint fallback, 429 logging, SERVICE_DISABLED fallthrough restored
  Roman | RyzenAdvanced · 2026-05-26 00:15:01 +04:00
  fced1653f2
• v3.10.10: fix normalizer preserves compaction summaries, dedupes goal_context
  Roman | RyzenAdvanced · 2026-05-25 23:59:30 +04:00
  8638769a73
• v3.10.9: Antigravity context normalizer, Claude thinking fix, endpoint lockdown, cobra91 PR #4

  Antigravity-only changes (no other providers touched):
  - Production-only endpoints (cloudcode-pa.googleapis.com), sandbox blocked
  - AntigravityContextNormalizer: bounded context for every request
  - Simple message detector: 'hi' sends minimal payload
  - Auto-reset polluted context (200+ items with simple msg)
  - Claude thinking: maxOutputTokens=64000, snake_case config, VALIDATED toolConfig
  - Claude budgets: low=8192, medium=16384, high=32768
  
  z.ai/OpenRouter (cobra91 PR #4):
  - Full OpenClaw attribution headers for z.ai
  - X-OpenRouter-Cache header for OpenRouter
  
  Other fixes:
  - Linux Re-OAuth: load_oauth_secrets() undefined, now inline
  - GLib.idle_add lambda truthy tuple fix
  - Project discovery uses production endpoint

  Roman | RyzenAdvanced · 2026-05-25 22:28:43 +04:00
  a1d0fc3707
• v3.10.8: Fix Re-OAuth buttons, block staging/sandbox for Antigravity, prefer production endpoint

  - Fix Linux GUI Re-OAuth: load_oauth_secrets() was undefined, now loads inline
  - Fix GLib.idle_add lambda returning truthy tuple (repeated callbacks)
  - Proxy: production cloudcode-pa.googleapis.com tried first, sandbox as fallback
  - Proxy: 403 SERVICE_DISABLED falls through to next endpoint
  - Project discovery validates against production endpoint, not staging
  - Antigravity preset base_url changed to production
  - Windows GUI project discovery also uses production endpoint

  Roman | RyzenAdvanced · 2026-05-25 21:57:30 +04:00
  f645e92908
• v3.10.7 — Prompt Enhancer: offline + AI-powered modes, per-provider toggle
  Roman | RyzenAdvanced · 2026-05-25 21:04:30 +04:00
  f3f536e428
• v3.10.6: Linux/Windows feature parity — OAuth Secrets all providers, Re-OAuth, Sync from Preset, Codebuff OAuth
  Roman | RyzenAdvanced · 2026-05-25 19:37:42 +04:00
  869a2625fc
• v3.10.6: update changelogs with OAuth Secrets all-providers + Re-OAuth
  Roman | RyzenAdvanced · 2026-05-25 18:49:21 +04:00
  9796c56451
• v3.10.6 — Freebuff integration, Codebuff OAuth fix, cobra91 PR #3 (CROF gate, data consolidation, sticky port)
  Roman | RyzenAdvanced · 2026-05-25 17:59:30 +04:00
  b0f1b287a4
• v3.10.5 — Add Windows GUI (tkinter) in windows/ folder + README
  Roman | RyzenAdvanced · 2026-05-25 16:00:26 +04:00
  d28195dc1b
• v3.10.5 — Update CHANGELOG + README: compaction, OAuth secrets, model mapping
  Roman · 2026-05-25 14:48:05 +04:00
  db2b33befc
• v3.10.4 — Clean release: OAuth Secrets editor, Import JSON, no leaked credentials
  Roman · 2026-05-25 14:14:54 +04:00
  cce7c2e38c
• v3.10.3 — Fix Antigravity 404: verified REST model IDs
  Roman · 2026-05-25 13:17:44 +04:00
  419d916399
• v3.10.2 — Fix Antigravity models: use display names not slugs
  Roman · 2026-05-25 12:53:00 +04:00
  38e585c9d5
• v3.10.0 — Provider model editor: Remove/Clear/Sync buttons + Antigravity refresh
  Roman · 2026-05-25 12:38:57 +04:00
  539186ff40
• v3.9.9 — Refresh Antigravity models: Gemini 3.5 Flash, Claude 4.6, GPT-OSS 120B
  Roman · 2026-05-25 12:29:41 +04:00
  b39800497b
• v3.9.8 — Fix Desktop model leak, global BrokenPipeError protection
  Roman · 2026-05-25 12:07:40 +04:00
  aa7007d2d9
• v3.9.7 — Rename codebuff to Codebuff in public docs
  Roman · 2026-05-25 11:22:35 +04:00
  e730b035c5
• v3.9.7 — Forward real codebuff error messages, fix BrokenPipeError crash, fix SyntaxWarnings
  Roman · 2026-05-25 11:12:05 +04:00
  c93707745f
• v3.9.6: update CHANGELOG.md
  admin · 2026-05-24 20:21:17 +00:00
  9a55f60782
• v3.9.0: update CHANGELOG.md
  admin · 2026-05-24 19:32:39 +00:00
  3f3ed516c8
• v3.8.4: Fix codebuff DeepSeek V4 tool-call reasoning_content round-trip

  - Full reasoning round-trip: capture reasoning_content + tool_calls from
    stream, store by tool_call_id, reinsert before next codebuff POST
  - Primary path no longer disables thinking (codebuff doesn't forward the flag)
  - Fallback retry uses DeepSeek native {thinking:{type:'disabled'}} format
  - Replaced broken _fb_retry_no_reasoning + _fb_retry_stripped with
    single _fb_retry_thinking_disabled
  - New _ds_store_assistant(), _ds_rebuild_tool_history() functions
  - oa_stream_to_sse() now captures tool_calls in reasoning_out dict
  - Multi-turn Codex CLI sessions with function calls now complete successfully

  admin · 2026-05-24 21:48:00 +04:00
  8fd6f280f2
• v3.8.3: Fix codebuff streaming — SSE events now reach Codex client

  - Root cause: _handle_codebuff streaming loop collected events but never
    wrote them to self.wfile (stream_buffered_events was not called)
  - Fix: Replaced manual loop with stream_buffered_events() + on_event callback
  - Confirmed working: raw API streaming, non-stream, and stream through proxy
  - Updated CHANGELOG.md, README.md, version labels to 3.8.3

  admin · 2026-05-24 19:30:10 +04:00
  2d4c1a9c2d
• v3.8.1: update CHANGELOG.md + README.md with codebuff docs
  admin · 2026-05-24 16:43:48 +04:00
  e265584af9
• docs: update CHANGELOG, README, GUI changelog for v3.8.0 AI Monitoring

  - CHANGELOG.md: full v3.8.0 section with 3-tier system, 30 fault types, safety guards
  - README.md: AI Monitoring badge, features section, Phase 9 dev journey, troubleshooting rows
  - GUI CHANGELOG: v3.8.0 entry with 9 bullet points

  admin · 2026-05-22 23:22:26 +04:00
  6adabb67e6
• v3.7.0: Intelligence Routing — self-healing parser system

  Layer 1 (FIX 23): Deep URL extraction from nested JSON in explore_agent blocks.
  Layer 2 (FIX 24): Auto-proceed on require_escalation / request_escalation_permission.
  Layer 3 (FIX 25): Intent-based command synthesis with 5 heuristics when all parsers fail.
  
  Module-level _build_explore_cmd() for reuse across parser + stream path.
  54 self-test patterns (up from 41).

  admin · 2026-05-22 16:29:45 +04:00
  51b89e6d08
• v3.6.0 — Performance & Stability Hardening

  P0: Connection pooling (http.client reuse per host), stream idle timeout
      (300s via selectors) on all streaming paths (OA/CC/Gemini/auto-continue)
  P1: Retry-After header support on all retry paths, preemptive OAuth token
      refresh (5min before expiry)
  P2: oa_convert_tools(strict=) for Responses vs Chat Completions, filter
      null/empty tool names
  P3: Response store TTL (600s eviction), bounded stream buffers (8MB cap),
      response.failed/error urgent flush, dual logging (proxy.log)
  
  .deb: v3.6.0 (71KB) — v3.5.0 and v3.3.0 kept as fallback

  admin · 2026-05-22 13:14:51 +04:00
  ee78d35aa7
• v3.5.0 — Major Release: Command Code Multi-Format Parser, AI Assist, Self-Revive Watchdog

  CC Adapter (17 fixes):
  - Multi-format tool-call parser chain: DSML → bash → explore → XML → raw JSON → fallback
  - Three-tier argument parser (direct/unescape/unicode_escape)
  - Recursive double/triple-wrap unwrapping (_unwrap_cmd)
  - Post-extraction sanitizer validation
  - DSML tag support (current CC model format)
  - Self-revive watchdog (50 restarts, progressive backoff)
  - Debug-to-file logging (cc-debug.log)
  - Inline self-test (19 tests via --self-test)
  - ErrorAnalyzer with 4xx learning on retry
  - Schema cache with 24h TTL
  
  Launcher:
  - AI Assist integration
  - Updated usage dashboard
  - Reasoning controls per-provider
  - Updated cleanup patterns
  
  .deb: v3.5.0 (70KB) — v3.3.0 kept as fallback

  admin · 2026-05-22 10:54:30 +04:00
  4bbfb4ada7
• v3.3.0: fix auto-continue class breakage, add MAX_TOKENS auto-continue for Gemini/Antigravity, bump version label
  Roman · 2026-05-20 22:00:49 +04:00
  e0cad357f2
• v3.3.0: Antigravity OAuth + Gemini CLI OAuth, full Codex agent loop with tool calls, history hardening, SSE fixes
  Roman · 2026-05-20 21:44:33 +04:00
  7aa1f10877
• v3.0.0: ThreadingHTTPServer, dynamic ports, health gating, atomic config, safe cleanup, buffered SSE, batched stats, graceful shutdown
  Roman · 2026-05-20 18:54:47 +04:00
  c0c4d7e420
• v2.7.0: Usage Dashboard redesign (OpenUsage-inspired), TCP_NODELAY streaming, Anthropic prompt caching
  Roman · 2026-05-20 18:11:39 +04:00
  cbd1f558dd
• v2.6.1: rebuild Google OAuth to emulate Gemini CLI

  - Uses Google's public OAuth client_id (no client_secret.json needed)
  - PKCE + CSRF state protection for secure auth
  - Scopes: cloud-platform, generative-language, userinfo
  - Just click OAuth Login -> browser -> authorize -> done
  - Zero setup required

  Roman · 2026-05-20 17:38:08 +04:00
  d55b0322e8
• v2.6.0: Usage Dashboard, per-provider tracking, OAuth file picker

  - Usage Dashboard: visual cards with success rate bars, token stats, latency
  - Per-model breakdown and error tracking per provider
  - Proxy records usage-stats.json after every request
  - Google OAuth: browse for client_secret.json instead of fixed path
  - Auto-copies selected file to ~/.cache/codex-proxy/

  Roman · 2026-05-20 17:23:14 +04:00
  bd4ccf1635
• v2.5.1: adaptive retry for 429/502/503, socket reuse, BGP retry

  - Exponential backoff retry (2s/4s/8s) for rate limits and transient errors
  - BGP routes retry before failing over to next route
  - Socket SO_REUSEADDR prevents 'Address already in use' crashes
  - Connection reset/broken pipe also retried
  - BGP route count shown at proxy startup

  Roman · 2026-05-20 17:10:40 +04:00
  0e70fa47f9
• v2.5.0: AI BGP multi-provider routing with automatic failover

  - New AI BGP pool manager (create/edit/delete pools)
  - Each pool has ordered routes from any configured endpoint
  - Failover: tries primary, falls back to next route on error
  - Pools appear in endpoint dropdown with shuffle icon
  - Pool editor with route add/remove/reorder
  - Fixed TOML breakage from multi-line paste
  - Added OpenAdapter preset with 0G models

  Roman · 2026-05-20 16:40:57 +04:00
  12ca136fba
• v2.4.0: add OpenAdapter preset, fix dialog crash, smarter OAuth UX
  Roman · 2026-05-20 15:28:41 +04:00
  866c07c2b5
• v2.3.2: add Google Gemini provider with OAuth support

  - Two presets: API Key and OAuth modes
  - OAuth Login button: full Google OAuth2 flow with auto-refresh
  - Auto-refreshes expired access tokens using refresh_token
  - Gemini OpenAI-compatible endpoint works with existing proxy
  - Models: gemini-2.5-flash, gemini-2.5-pro, gemini-2.0-flash, etc.

  Roman · 2026-05-20 14:45:43 +04:00
  ea60d74527
• v2.3.0: adaptive Crof self-healing system

  - Per-model success/failure tracking with dynamic item limits
  - Proactive compaction when above learned limit
  - Auto-retry on finish_reason=length with aggressive re-compaction
  - Tested: kimi-k2.6 (27 items) and mimo-v2.5-pro both completed
  - All previous fixes included: _ts crash, connection reset, timeout, orphaned fco

  Roman · 2026-05-20 14:32:36 +04:00
  27b22f4fd8
• v2.2.1: fix compaction orphaning tool outputs causing Crof incomplete

  - Fix compaction cutting between function_call and function_call_output pairs
  - Orphaned tool results confused Crof models causing finish_reason=length
  - reasoning_effort=none now always sends enable_thinking=false too
  - Added Crof upstream debug logging

  Roman · 2026-05-20 13:14:40 +04:00
  6c67642522
• v2.2.0: styled reasoning switch + error handling for dialogs

  - Reasoning switch: green ON, orange OFF, gentle rounded pill shape
  - Error handling on Add/Edit/Manage Endpoints dialogs
  - Updated CHANGELOG.md

  Roman · 2026-05-20 12:59:04 +04:00
  9bcb6998e0
• v2.2.0: per-provider reasoning controls (on/off + effort level)

  - Add Reasoning On/Off toggle and Effort selector in endpoint editor
  - Proxy sends enable_thinking=false when reasoning is OFF
  - Proxy sends reasoning_effort level when reasoning is ON
  - Strip reasoning_content from output, force max_tokens=64000 minimum
  - Fixes Crof mimo-v2.5-pro and similar reasoning model token exhaustion

  Roman · 2026-05-20 12:20:33 +04:00
  9532ba40f3
• v2.1.3: fix Crof mimo-v2.5-pro reasoning_content token exhaustion

  - Strip reasoning_content from proxy output (Codex doesn't use it)
  - Force max_tokens=64000 minimum for openai-compat providers
  - Prevents models that emit large reasoning from running out of tokens

  admin · 2026-05-19 21:59:38 +04:00
  77423c5c35
• feat: auto-compaction for long conversations (like Claude Code/Codex /compact)

  Instead of just truncating old items, the proxy now auto-compacts
  them into a structured summary preserving key context:
  - User requests, assistant responses, tool calls made, files touched
  - Keeps original query + system messages + last 10 recent items
  - 38 items -> 14 items in testing, with summary of dropped turns
  - Similar to Claude Code's auto-compact and Codex CLI's /compact
  - No extra API calls needed, instant, zero cost

  admin · 2026-05-19 21:49:55 +04:00
  662d8e961e
• fix: truncate large tool outputs to prevent Crof incomplete responses

  Crof models (mimo, deepseek-v4-pro) return status=incomplete when
  tool results contain too much text (e.g. full HTML pages at 8500+ tokens).
  Auto-truncate tool outputs exceeding 8000 chars with truncation notice.
  Combined with the 30-item conversation trim from previous commit.

  admin · 2026-05-19 21:37:34 +04:00
  c90912ed07
• fix: Crof multi-turn tool calls + auto-trim long conversations

  Root cause: Codex sends function_call items with id=None, causing
  tool_call_id mismatch between tool calls and tool results. Proxy now
  resolves IDs by call_id + positional fallback.
  
  Auto-trim: conversations exceeding 30 items are trimmed automatically,
  keeping system messages, original user query, and most recent items.
  This prevents context overflow on providers with smaller context
  windows (Crof mimo-v2.5-pro stops responding at ~40 items).
  
  - Fix None tool IDs in oa_input_to_messages with positional matching
  - Auto-trim input to 30 items max (keeps head + tail)
  - Add request/response logging to ~/.cache/codex-proxy/requests.log
  - Proxy stderr visible in launcher terminal for debugging
  - v2.1.2

  admin · 2026-05-19 21:25:35 +04:00
  aa377024d9