Skip to main content

Roo Code 3.25.8 Release Notes (2025-08-06)

This release fixes a critical memory leak in long conversations, prevents disabled MCP servers from starting, and provides more flexibility for Gemini 2.5 Pro thinking budgets, along with numerous stability improvements.

Disabled MCP Servers No Longer Start​

We've fixed a significant issue where disabled MCP servers were still starting processes and consuming resources (#6084):

  • No Wasted Resources: Disabled servers now truly stay disabled and don't start background processes
  • Clear Status Indicators: Server status now accurately reflects whether a server is enabled or disabled
  • Immediate Cleanup: When MCP is globally disabled, all servers immediately disconnect

This ensures your system resources are only used for the MCP servers you actually want to run.

Memory Leak Fix for Long Conversations​

We've resolved a critical memory leak that was causing excessive memory usage and grey screens in long conversations (thanks xyOz-dev!) (#6697):

  • Bounded Memory Usage: Virtual scrolling now limits viewport rendering to 1000px below the current view instead of rendering infinitely
  • Optimized Caching: Reduced cache size and TTL to prevent memory accumulation
  • Stable Performance: Memory usage remains constant regardless of conversation length

Your long conversations will now remain responsive and stable without the dreaded grey screen appearing.

Gemini 2.5 Pro Thinking Budget Flexibility​

Gemini 2.5 Pro users now have much more control over the thinking budget (#6588):

  • Lower Minimum: The minimum thinking budget is reduced from 1024 to 128 tokens
  • More Concise Coding: Perfect for tasks that need quick, focused responses without extensive reasoning
  • Manual Adjustment Required: You'll need to manually adjust the thinking budget to 128 in your settings to take advantage of this feature

This change gives you the flexibility to optimize for either deep reasoning or more concise, rapid responses depending on your needs.

Bug Fixes​

  • MCP Server Refresh: Settings changes no longer trigger unnecessary MCP server refreshes (#6779)
  • Codebase Search: The tool now correctly searches the entire workspace when using "." as the path (#6517)
  • Swift File Support: Fixed VS Code crashes when indexing projects containing Swift files (thanks niteshbalusu11, sealad886!) (#6724)
  • Context Management: Model max tokens now intelligently capped at 20% of context window to prevent excessive condensing (#6761)
  • OpenAI Configuration: Extra whitespace in base URLs no longer breaks model detection (thanks vauhochzett!) (#6560)
  • URL Fetching: Better error recovery when fetching content from URLs (thanks QuinsZouls!) (#6635)
  • Qdrant Recovery: Code indexing automatically recovers when Qdrant becomes available after startup errors (#6661)
  • Chat Scrolling: Eliminated scroll jitter during message streaming (#6780)