Skip to main content

Roo Code 3.25.21 Release Notes (2025-01-22)

This release enables grounding features for Vertex AI, adds subtask todo list support, and includes the new Featherless provider.

Subtask Todo Listsโ€‹

Streamline hierarchical task planning with the new optional todo list parameter for subtasks (#6775):

  • Pass Todo Lists: Include predefined todo lists when creating subtasks
  • Maintain Context: Pass along context to the subtask in the form of a todo list
  • Optional Enforcement: The "New Task Require Todos" setting in VS Code can enforce todo lists for all new subtasks if desired
Subtask todo lists configuration in VS Code settings

This feature works out of the box, and you can optionally configure VS Code settings to require todos for all new tasks.

๐Ÿ“š Documentation: See Task Todo List for detailed information about todo list management.

Vertex AI Grounding Featuresโ€‹

Vertex AI users can now access powerful grounding capabilities previously exclusive to the Gemini provider (thanks anguslees!) (#6777):

  • Google Search Grounding: Get real-time information from web searches directly in your AI responses
  • URL Context: Provide specific web pages as context for more accurate, up-to-date responses
  • Enhanced Accuracy: Combine AI capabilities with current web information for better results

These features bring Vertex AI to feature parity with Gemini for grounding capabilities.

๐Ÿ“š Documentation: See Vertex AI Provider for configuration details.

Provider Updatesโ€‹

  • Featherless Provider: Access a wide range of open-source models through the new Featherless AI platform (thanks DarinVerheijke!) (#7235)

QOL Improvementsโ€‹

  • Context Window Error Handling: Improved error detection and automatic recovery when hitting context limits - the system now detects errors from multiple providers (OpenAI, Anthropic, Cerebras) and automatically truncates context by 25% before retrying (up to 3 attempts) (#6967)
  • Marketplace Filter: New "Show installed only" checkbox to easily view and manage your installed MCPs and modes (thanks semidark!) (#7007)
  • DeepSeek Context: Updated DeepSeek models to support 128k context window (#7269)

Bug Fixesโ€‹

  • OpenAI Responses: Fixed GPT-5 subtask results not being provided correctly in OpenAI Responses API (#7305)
  • Temperature Handling: Fixed temperature parameter issues for OpenAI Compatible providers (#7188)
  • Provider Validation: Added 'roo' provider to key validation function (#7239)
  • UI Stability: Fixed context condense UI crash when handling null/undefined token values (#6916)
  • Storage Checks: Improved filesystem checks for better storage handling (thanks elianiva!) (#7164)

Misc Improvementsโ€‹

  • MDM Authentication: Clear feedback when authentication is required by your organization's MDM policy (#7291)
  • Security Updates: Updated mermaid dependency to v11.10.0 (#7231) and tmp dependency to v0.2.4 (#6762)
  • Task Tracking: Pass task ID in Roo provider request headers (#7303)
  • Error Handling: Removed exceptions from RooHandler constructor (#7302)
  • Cloud Services: Various internal improvements for cloud services (#7299, #7258)
  • Build Optimization: Marked non-English package files as linguist-generated (#7271)