Roo Code 3.25.4 Release Notes (2025-07-31)
This release introduces two new AI providers, enhances prompt suggestions with conversation context, and improves various tool functionalities.
Doubao Provider Support
We've added support for Doubao, ByteDance's AI model provider (thanks AntiMoron!) (#6345):
- Access to ByteDance AI Models: Connect to Doubao's language models for your AI-powered development tasks
- Full Integration: Complete provider implementation with API handling
- Easy Configuration: Simple setup through Roo Code's provider settings with internationalized prompts
Doubao expands your AI model options, giving you access to ByteDance's competitive language models alongside existing providers.
📚 Documentation: See Doubao Provider Guide for setup instructions and available models.
SambaNova Provider Integration
We've integrated SambaNova as a new LLM provider (thanks snova-jorgep!) (#6188):
- High-Speed Inference: Experience competitive inference speeds for faster AI responses
- Broader Model Selection: Access SambaNova's diverse range of language models
- Seamless Integration: Configure and use SambaNova models just like any other provider
SambaNova brings enterprise-grade AI infrastructure to Roo Code, offering powerful alternatives for your development workflows.
📚 Documentation: See SambaNova Provider Guide for configuration details.
Context-Aware Prompt Enhancement
Prompt enhancement now uses your conversation history for better suggestions (thanks liwilliam2021!) (#6343):
- Smarter Suggestions: Enhancement considers your last 10 messages to generate more relevant prompts
- Reduced Hallucinations: Context awareness prevents the AI from making unfounded suggestions
- Flexible Configuration: Use a separate API configuration for enhancement operations
- Toggle Control: Enable or disable task history inclusion based on your needs
To enable this feature: Settings → Prompts tab → Select "ENHANCE" → Check "Include task history in enhancement" for better context.
📚 Documentation: See Prompt Enhancement Guide for configuration options.
QOL Improvements
- Task Resumption Message: Improved LLM instructions by removing misleading message about resuming tasks (thanks KJ7LNW!) (#5851)
- AGENTS.md Support: Added symlink support for AGENTS.md file loading (#6326)
- Marketplace Updates: Auto-refresh marketplace data when organization settings change (#6446)
Bug Fixes
- Execute Command Kill Button: Fixed the kill button functionality for the execute_command tool (#6457)
- Token Counting: Fixed accuracy issues by extracting text from messages using VSCode LM API (thanks NaccOll!) (#6424)
Provider Updates
- Chutes AI: Added zai-org/GLM-4.5-FP8 model support (#6441)
- OpenRouter: Set horizon-alpha model max tokens to 32k (#6470)
- Databricks: Added support for /invocations endpoints pattern (thanks adambrand!) (#6317)