Roo Code 3.36.0 Release Notes (2025-12-04)
This release introduces non-destructive context management for reliable checkpoint restoration, adds reasoning visibility for supported models, and improves provider reliability across Roo, Minimax, Cerebras, and Vercel AI Gateway.
Non-Destructive Context Management
Context condensing and sliding window truncation now preserve your original messages internally rather than deleting them (#9665). When you rewind to an earlier checkpoint, the full conversation history is restored automatically. This applies to both automatic condensing and sliding window operations.
Provider Updates
-
Reasoning Details Support: The Roo provider now displays reasoning details from models with extended thinking capabilities, giving you visibility into how the model approaches your requests (#9796)
-
Native Tools Default: All Roo provider models now default to native tool protocol for improved reliability and performance (#9811)
-
Minimax search_and_replace: The Minimax M2 model now uses search_and_replace for more reliable file editing operations (#9780)
-
Cerebras Token Optimization: Conservative 8K token limits prevent premature rate limiting (thanks sebastiand-cerebras!), plus deprecated model cleanup (#9804)
-
Vercel AI Gateway: More reliable model fetching for models without complete pricing information (#9791)
Bug Fixes
-
Write Tool Validation: Resolved false positives where
write_to_fileincorrectly rejected complete markdown files containing inline code comments like# NEW:or// Step 1:(thanks Michaelzag!) (#9787) -
Download Count Display: Fixed homepage download count to display with proper precision for million-scale numbers (#9807)
Misc Improvements
-
Tool Consolidation: Removed the deprecated
insert_contenttool; useapply_difforwrite_to_filefor file modifications (#9751) -
Experimental Settings: Temporarily disabled the parallel tool calls experiment while improvements are in progress (#9798)
-
Infrastructure: Updated Next.js dependencies for web applications (#9799)