Skip to main content

Roo Code 3.29 Release Notes (2025-10-31)

This combined summary covers the 3.29.x releases. It introduces the MiniMax provider, improves @ file search on large projects, clarifies the Logs tab and VS Code LM guidance, and includes multiple stability fixes.

Feature Highlights​

  • Cloud Agent: PR Reviewer for instant high‑quality code reviews (learn more: https://roocode.com/reviewer)
  • Intelligent file reading with token‑budget management and a 100 KB preview for very large files (#8789)
  • MiniMax provider for coding and reasoning with strong Chinese language support and competitive pricing (thanks Maosghoul!) (#8820)
  • Faster and more reliable @ file search on large repositories with configurable indexing limits and respect for VS Code ignore settings (thanks Naituw!) (#8805)
  • MCP tab renamed to “Logs” to match mixed-level messages (thanks hannesrudolph!) (#8894)

QOL Improvements​

  • Improve @ file search for large projects; add the roo-cline.maximumIndexedFilesForFileSearch setting with guidance on memory tradeoffs (#8805)
  • Rename MCP “Errors” tab to “Logs” to match info/warning/error messages; clearer empty state (“No logs yet”) (#8894)
  • Custom modes load from the configured storage path and persist across restarts (thanks elianiva!) (#8499)
  • Clarify VS Code LM API integration warning in Settings to reduce “model not supported” errors (#8493)
  • Keyboard shortcut: “Add to Context” moved to avoid Redo conflict; restores standard Redo (#8653)
  • Auto‑approve button responsiveness improved (#8798)
  • After “Add to Context,” input auto‑focuses and inserts two newlines to keep typing immediately (#8877)

Bug Fixes​

  • Reasoning effort selection auto-enables reasoning when needed so UI and behavior stay in sync (#8890)
  • Suppress noisy cloud‑agent exceptions for cleaner logs (#8577)
  • Prevent MCP server restart when toggling “Always allow” for MCP tools (#8633)
  • Reuse existing Qdrant index after outages to avoid full reindex and cut restart time (#8588)
  • Make code index initialization non‑blocking at activation to avoid startup hangs (#8933)
  • Honor maxReadFileLine across code definition listing and file reads to prevent context overflows (#8509)
  • Prevent infinite retry loop when canceling during auto‑retry (#8902)
  • Gate auth‑driven Roo model refresh to the active provider only to reduce background work (#8915)
  • search_files now respects .gitignore (including nested) by default; override when needed (#8804)
  • apply_diff export preserves trailing newlines (fix stripLineNumbers) (#8227)
  • Fix provider model loading race conditions to reduce timeouts and intermittent errors (#8836)
  • Rate limiting now uses a monotonic clock and enforces a hard cap to avoid long lockouts (#8456)
  • Messages typed during context condensing now send automatically once condensing finishes; per‑task queues no longer cross‑drain (#8478)
  • Checkpoint menu popover no longer clips long option text; items remain fully visible (#8867)

Provider Updates​

  • MiniMax provider added (MiniMax‑M2) with API key setup and model selection in UI (#8820)
  • Cerebras: add zai‑glm‑4.6 and change default to gpt‑oss‑120b; deprecate qwen‑3‑coder models (#8920)
  • Roo Code Cloud: dynamic model loading in the Model Picker with caching and graceful fallback (#8728)
  • Chutes AI: LongCat‑Flash‑Thinking‑FP8 models (200K, 128K) for longer coding sessions (#8426)
  • OpenRouter: add Anthropic Claude Haiku 4.5 to prompt‑caching models (#8764)
  • Z.ai (GLM‑4.5/4.6): “Enable reasoning” Deep Thinking toggle in UI (#8872)
  • Roo provider: Reasoning effort control to choose deeper thinking vs. faster/cheaper (#8874)
  • OpenAI‑compatible: centralized ~20% maxTokens cap to prevent context overruns (#8822)
  • Gemini: updated model list and “latest” aliases for easier selection (#8486)

Documentation Updates​