Skip to main content

Roo Code 3.18 Release Notes (2025-05-21)

This release introduces advanced AI models with enhanced reasoning capabilities, comprehensive context condensing improvements, YAML support for custom modes, streamlined development infrastructure, and numerous quality-of-life improvements and bug fixes.

Codebase Indexing (Experimental)​

Our new experimental Codebase Indexing feature enables semantic search across your entire project using AI embeddings. Remember, this is disabled by default and requires additional setup (enable in Settings (⚙️) > "Experimental").

Codebase Indexing Settings

Key capabilities:

  • Semantic Code Search: Find relevant code using natural language queries instead of exact keyword matches
  • AI-Powered Understanding: The new codebase_search tool understands code meaning and relationships
  • Cross-Project Discovery: Search your entire indexed codebase, not just open files
  • Flexible Setup: Choose between OpenAI embeddings or local Ollama processing
  • Vector-Powered Search: Advanced embedding technology with Qdrant vector database storage
  • Enhanced Settings: Improved configuration saving and Ollama indexing performance (thanks daniel-lxs!)

Transform how you navigate large codebases - ask "find authentication logic" and get meaningful results instead of hunting through files.

For setup instructions and detailed usage, see the main Codebase Indexing documentation.

Context Condensing Upgrades​

Our Intelligent Context Condensing feature sees significant enhancements for better control and clarity. Note: As of version 3.19.0, this feature is enabled by default.

Watch a quick overview:

Key updates:

  • Adjustable Condensing Threshold & Manual Control: Fine-tune automatic condensing or trigger it manually. Learn more.
  • Advanced Controls & UI Enhancements: New settings for fine-tuning conversation compression behavior with improved interface components for managing condensing settings. (thanks SannidhyaSah!)
  • Clear UI Indicators: Better visual feedback during condensing. Details.
  • Improved Compression: Better conversation summarization while preserving important context during compression. (thanks SannidhyaSah!)
  • Accurate Token Counting: Improved accuracy for context and cost calculations. More info.

For full details, see the main Intelligent Context Condensing documentation.

Custom Modes: YAML Support​

Custom mode configuration is now significantly improved with YAML support for both global and project-level (.roomodes) definitions. YAML is the new default, offering superior readability with cleaner syntax, support for comments (#), and easier multi-line string management. While JSON remains supported for backward compatibility, YAML streamlines mode creation, sharing, and version control.

For comprehensive details on YAML benefits, syntax, and migrating existing JSON configurations, please see our updated Custom Modes documentation. (thanks R-omk! Additional thanks to GreatScottyMac for the YAML inspiration!)

API Cost Control: Request Limits​

To enhance API cost management, you can now set a Max Requests limit for auto-approved actions. This prevents Roo Code from making an excessive number of consecutive API calls without your re-approval.

Setting the Max Requests limit for auto-approved actions in Roo Code settings.

Setting the "Max Requests" for auto-approved actions.

Warning message indicating the auto-approved request limit has been reached.

Notification when the auto-approved request limit is met.

Learn more about configuring this safeguard in our Rate Limits and Costs documentation. (Inspired by Cline, thanks hassoncs!)

Documentation Updates​

  • README Display: Fixed README GIF display across all localized versions, ensuring consistent visual guides for users in all 17 supported languages
  • Referrer URL: Fixed technical referrer URL issues in documentation links

New Model Version: Gemini 2.5 Flash Preview (May 2025)​

Access the latest gemini-2.5-flash-preview-05-20 model, including its thinking variant. This cutting-edge addition is available via both the generic Gemini provider and the Vertex provider, further expanding your AI model options. (thanks shariqriazz, daniel-lxs!)

QOL Improvements​

  • Better Tool Selection: Improved tool descriptions to guide AI in making better file editing decisions
  • MCP Server Loading: Fixed MCP Server error when running with npx and bunx (thanks devxpain!)
  • Auto-Approve Input: Fixed issue where chat input would incorrectly clear when auto-approve was enabled (thanks Ruakij!)
  • Resizable Prompt Textareas: Restored vertical resizing for prompt input textareas (role definition, custom instructions, support prompts) for easier viewing and editing of longer prompts.
  • Settings Import: Error messages are now provided in multiple languages if an import fails. (thanks ChuKhaLi!)
  • Traditional Chinese Locale: Improve zh-TW Traditional Chinese locale (thanks PeterDaveHello and chrarnoldus!)

Bug Fixes​

  • Codebase Indexing: Fixed settings saving and improved Ollama indexing performance for more reliable semantic search (thanks daniel-lxs!)
  • File Handling: Fixed handling of byte order mark (BOM) when users reject apply_diff operations (thanks avtc!)
  • Context Tracking: Fixed mark-as-read logic in the context tracker (thanks samhvw8!)
  • Model Cleanup: Removed deprecated claude-3.7-sonnet models from vscode-lm (thanks shariqriazz!)
  • Global Settings Export: Fixed global settings export functionality to ensure configurations transfer properly between setups
  • File Picker Icons: Fixed vscode-material-icons display issues in the file picker for better visual consistency
  • Terminal Integration: Fixed terminal integration to properly respect user-configured timeout settings (thanks KJ7LNW!)
  • VSCode Command Execution: Command execution is now type-safe, reducing potential errors.
  • Settings Import: The system now gracefully handles missing globalSettings in import files. (thanks ChuKhaLi!)
  • Diff View Scrolling: The diff view no longer scrolls unnecessarily during content updates. (thanks qdaxb!)
  • Syntax Highlighting: Improved syntax highlighting consistency across file listings, code definitions, and error messages. (thanks KJ7LNW!)
  • Grey Screen Issue: Implemented a fix for the "grey screen" issue by gracefully handling errors when closing diff views, preventing memory leaks. (thanks xyOz-dev!)
  • Settings Import Error Message: The settings_import_failed message is now correctly categorized as an error. (thanks ChuKhaLi!)
  • File Links to Line 0: Links to files with line number 0 now correctly open at the first line, preventing errors. (thanks RSO!)
  • Packaging tiktoken.wasm: Fixed packaging to include the correct tiktoken/lite/tiktoken_bg.wasm file for accurate token counting. (thanks vagadiya!)
  • Audio Playback: Moved audio playback to the webview to ensure cross-platform compatibility and reliability for notifications and sounds. (thanks SmartManoj, samhvw8!)

Provider Updates​

  • Claude 4.0 Models: Added support for Claude Sonnet 4 and Claude Opus 4 models with thinking variants for Anthropic, Bedrock, and Vertex providers (thanks shariqriazz!)
  • Enhanced Reasoning: Added reasoning support for Claude 4 and Gemini 2.5 Flash on OpenRouter, plus fixed o1-pro compatibility
  • Improved Defaults: Changed default model from Sonnet 3.7 to Sonnet 4 for better performance
  • xAI Models: Updated xAI model list and default model configuration for improved performance (thanks PeterDaveHello!)
  • Prompt Caching: Enabled prompt caching for Gemini 2.5 Flash Preview to reduce costs and improve response times (thanks shariqriazz!)
  • OpenRouter Settings: Model settings now persist when selecting specific OpenRouter providers
  • LiteLLM Refresh: Added ability to refresh LiteLLM models list for up-to-date model availability
  • LM Studio and Ollama Token Tracking: Token usage is now tracked for LM Studio and Ollama providers. (thanks xyOz-dev!)
  • LM Studio Reasoning Support: Added support for parsing "think" tags in LM Studio responses for enhanced transparency into the AI's process. (thanks avtc!)
  • Qwen3 Model Series for Chutes: Added new Qwen3 models to the Chutes provider (e.g., Qwen/Qwen3-235B-A22B). (thanks zeozeozeo!)
  • Unbound Provider Model Refresh: Added a refresh button for Unbound models to easily update the list of available models and get immediate feedback on API key validity. (thanks pugazhendhi-m!)
  • Requesty Thinking Controls: Add thinking controls for Requesty provider (thanks dtrugman!)
  • LiteLLM Metadata: Improve model metadata for LiteLLM provider

Misc Improvements​

  • Enhanced Debugging: Added metadata to API requests across all providers for better troubleshooting and monitoring (thanks dtrugman!)
  • Build Process: Improved pnpm bootstrapping and fixed bootstrap script parameters for better compatibility (thanks KJ7LNW!, ChuKhaLi!)
  • Code Quality: Simplified object assignment and modernized code patterns (thanks noritaka1166!)
  • Monorepo Migration: Switched codebase to a monorepo structure for improved development workflow
  • Automated Nightly Builds: Created automated nightly release system to deliver the latest features and fixes more quickly
  • Enhanced Toolchain: Improved build processes and dependency management for better reliability
  • Modern Linting: ESLint 9 upgrade with shared configurations catches issues before they reach users
  • E2E Test Suite: Reorganized end-to-end testing infrastructure for better quality assurance
  • Build Optimizations: Enhanced build process optimizations reduce bundle size and improve startup times
  • Quality Assurance: Advanced testing infrastructure catches more issues during development
  • Release Automation: Improved release process automation reduces manual errors and speeds up delivery
  • Auto-reload in Dev Mode: Core file changes now automatically trigger a window reload in development mode for a faster workflow. (thanks hassoncs!)
  • Simplified Loop Syntax: Refactored loop structures in multiple components to use for...of loops for improved readability. (thanks noritaka1166!)
  • Telemetry: Re-enable telemetry

These improvements work behind the scenes to provide a more reliable and stable extension experience with faster updates and higher quality releases.