Skip to main content

TRMR 3.19 Release Notes (2025-06-02)

This release series focuses on streamlining the user interface with intelligent context condensing enabled by default, improved mode organization, enhanced file protection, experimental multi-file reading capabilities, and numerous quality-of-life improvements.

Intelligent Context Condensing Now Default

We've enabled intelligent content condensing by default and relocated the condense button for easier access:

Context condensed message in chat
  • Default Activation: Intelligent context condensing is now enabled out of the box for all users
  • Improved Button Placement: Moved the condense button out of the expanded task menu for quicker access
  • Smart Error Handling: The system now skips condensing and shows clear error messages if context grows during the condensing process
Settings for Intelligent Context Condensing

Intelligent Context Condensing configuration options: automatic triggering toggle, threshold slider, API configuration selection, and custom prompt customization.

For comprehensive details about this feature, see the main Intelligent Context Condensing documentation.

Streamlined Mode Organization

We've reorganized the interface for better workflow organization:

  • Modes Tab: Transformed the Prompts tab into a dedicated Modes tab for clearer navigation
  • Settings Relocation: Moved support prompts to Settings for better organization
  • Cleaner Interface: Simplified the main interface by grouping related features more logically

Navigate between different modes and prompts more intuitively.

Multi-File Reading (Experimental)

We've introduced an experimental feature for handling large codebases more efficiently:

  • Batch File Processing: Read up to 100 files in a single operation for comprehensive code analysis
  • Performance Control: Adjustable concurrent file read limits to optimize for your system
  • Smart Permission Management: Streamlined approval process for multi-file operations
  • Efficiency Boost: Dramatically reduces the time needed to analyze multiple related files

This experimental feature transforms how TRMR understands and works with your entire codebase at once (thanks samhvw8!).

Enterprise-Ready VPC Endpoint Support

We've added crucial enterprise security features for AWS Bedrock users:

  • Secure Corporate Access: Configure custom VPC endpoints to keep all LLM transactions within your firewall
  • Easy Configuration: Simple checkbox and text field to enable VPC endpoint usage
  • Compliance Ready: Meet your organization's security policies while using powerful AI models
  • Flexible Control: Toggle VPC endpoints on/off as needed with settings persistence

Enterprise customers can now use TRMR with AWS Bedrock while maintaining complete network security (thanks kcwhite!). See the AWS Bedrock configuration guide for setup instructions.

QOL Improvements

  • Fix reading PDF, DOCX, and IPYNB files in read_file tool: Ensures reliable reading of these file types (thanks samhvw8!)
  • Enhanced Sidebar Compatibility: Menu functionality now remains stable when moving Roo between primary and secondary sidebars (thanks chrarnoldus!)
  • Streamable HTTP Transport for MCP Servers: Added support for MCP servers using streamable HTTP-based transport (thanks taylorwilsdon!) (#3260)
  • MCP Server Instructions Support: Added support for server instructions in MCP configuration (thanks qingyuan1109!) (#4210)
  • Optimized Chat Layout and Styles: Fine-tuned the layout and styling of chat messages for improved readability (thanks zhangtony239!) (#4205)
  • MCP Server Connection Reliability: Fixed SSE connection issue for MCP servers using web-based transports (thanks taylorwilsdon!) (#4251)

Bug Fixes

  • Major Memory Leak Fixes: Multiple memory leaks resolved for improved stability and performance:
    • ChatView: Fixed leaks from unmanaged async operations and setTimeouts (thanks kiwina!)
    • WorkspaceTracker: FileSystemWatcher and other disposables are now properly cleaned up (thanks kiwina!)
    • RooTips: setTimeout is now cleared to prevent state updates on unmounted components (thanks kiwina!)
    • RooIgnoreController: FileSystemWatcher leak resolved by ensuring Task.dispose() is always called (thanks kiwina!)
    • Clipboard: useCopyToClipboard now clears setTimeout to avoid memory leaks (thanks kiwina!)
    • ClineProvider: Instance cleanup improved to prevent lingering resources (thanks xyOz-dev!)
  • Directory Mention Protection: Fixed @directory not respecting .rooignore files (thanks xyOz-dev!)
  • Tool File Protection: Added rooignore checking for insert_content and search_and_replace tools
  • Memory Leak Fix: Resolved memory leak in ChatView by stabilizing callback props (thanks samhvw8!)
  • Empty File Creation: Fixed write_to_file to properly create empty files when content is empty (thanks Ruakij!)
  • Chat Input Stability: Fixed chat input clearing during running tasks (thanks xyOz-dev!)
  • MCP Server Authentication: Fixed SSE header passing to correctly authenticate with remote MCP servers
  • AWS Bedrock Conversations: Resolved context condensing error that required conversations to start with user messages
  • Terminal UTF-8 Encoding: Fixed encoding errors when running Ruby/CocoaPods commands (thanks mr-ryan-james!)
  • Sidebar Context Menus: Resolved empty menu items by properly setting sidebar name (thanks chrarnoldus!)
  • Documentation Links: Fixed broken CONTRIBUTING.md link in feature request template (thanks cannuri!)
  • Prevent Full File Dump on User Edit: Resolved an issue where editing a file before saving a proposed modification could dump the entire file content into the context. This fix prevents context bloat, maintains model focus, and reduces costs (#4219, #3654) (thanks KJ7LNW!).
  • Correct OpenAI-Compatible API Key Label: The label for OpenAI-compatible API keys in settings now correctly reads "API Key" instead of "OpenAI API Key" for better clarity (#4191).
  • Prevent Invalid Markers in apply_diff REPLACE: Added validation to ensure :start_line: and :end_line: markers are not incorrectly included in the REPLACE section of apply_diff operations, improving robustness (#4015) (thanks KJ7LNW!).
  • Fix Virtuoso Footer Re-rendering: Addressed an issue where the Virtuoso component's footer would continuously re-render, improving UI performance (#4209) (thanks kiwina!).
  • Release Memory in apply_diff: Optimized memory usage in the apply_diff tool by explicitly releasing the original content buffer after use, helping to prevent potential memory leaks (#4193) (thanks xyOz-dev!).
  • Fix Directory Link Handling in Markdown: Enhanced how directory links in markdown (e.g., [some/directory/](some/directory/)) are handled, ensuring they correctly open in the Explorer view (#3690) (thanks KJ7LNW!).
  • Typo Fixes: Corrected various typos in comments and tool use prompts (#4220, #4095) (thanks mrubens, noritaka1166!).

Provider Updates

  • Gemini 2.5 Pro Preview 06-05 Model Support: Added support for the newly released Gemini 2.5 Pro Preview 06-05 model (thanks daniel-lxs and shariqriazz!). This model is available in the Gemini, Vertex, and OpenRouter providers.
  • DeepSeek R1 0528: Add DeepSeek R1 0528 model support to Chutes provider (thanks zeozeozeo!)
  • AWS Regions: Updated AWS regions to include Spain and Hyderabad
  • Enhanced LiteLLM Provider: Added cached read and write tokens to stats and cost calculation (thanks mollux!) (#4206)

Misc Improvements

  • Enforce codebase_search as primary tool: TRMR now always uses codebase_search as the first step for code understanding tasks, improving accuracy and consistency (thanks hannesrudolph!)
  • Improved Docker setup for evals: Dockerfile and docker-compose updated for better isolation, real-time monitoring, and streamlined configuration
  • Move evals into pnpm workspace, switch from SQLite to Postgres: Evals are now managed in a pnpm workspace and use PostgreSQL for improved scalability
  • Refactor MCP to use getDefaultEnvironment for stdio client transport: Simplifies MCP client setup and improves maintainability (thanks samhvw8!)
  • Get rid of "partial" component in names referencing not necessarily partial messages: Improves code clarity (thanks wkordalski!)
  • Improve feature request template: Makes it easier to submit actionable feature requests (thanks elianiva!)
  • MCP Environment Variables: Add environment variable injection support for whole MCP config (thanks NamesMT!)
  • Shell Compatibility: Improve POSIX shell compatibility in pre-push hook (thanks PeterDaveHello and chrarnoldus!)
  • Windows Terminal: Update PAGER environment variable for Windows compatibility in Terminal (thanks SmartManoj!)
  • Search Clarity: Update codebase search description to emphasize English query requirements (thanks ChuKhaLi!)
  • Backend Task Tracking: Added metadata to Unbound API requests for improved service reliability (thanks pugazhendhi-m!)
  • Model-Specific Caching: Optimized cache strategies for Gemini and Anthropic models for better performance
  • Unified History Item UI: Refactored history item UI components for code consistency (thanks daniel-lxs!) (#4151)
  • Node.js Upgrade: Upgraded Node.js to v20.19.2 to incorporate the latest security enhancements and ensure a more secure environment (thanks PeterDaveHello!) (#4212)

Thank you to all our contributors for making TRMR better with each release!