Release v3.34.8

[3.34.8] - 2025-11-27

• Fix: Race condition in new_task tool for native protocol (PR #9655 by @daniel-lxs)

Release v3.34.7

[3.34.7] - 2025-11-27

• Support native tools in the Anthropic provider for improved tool calling (PR #9644 by @mrubens)
• Enable native tool calling for z.ai models (PR #9645 by @mrubens)
• Enable native tool calling for Moonshot models (PR #9646 by @mrubens)
• Fix: OpenRouter tool calls handling improvements (PR #9642 by @mrubens)
• Fix: OpenRouter GPT-5 strict schema validation for read_file tool (PR #9633 by @daniel-lxs)
• Fix: Create parent directories early in write_to_file to prevent ENOENT errors (#9634 by @ivanenev, PR #9640 by @daniel-lxs)
• Fix: Disable native tools and temperature support for claude-code provider (PR #9643 by @hannesrudolph)
• Add 'taking you to cloud' screen after provider welcome for improved onboarding (PR #9652 by @mrubens)

Release v3.34.6

[3.34.6] - 2025-11-26

• Add support for AWS Bedrock embeddings in code indexing (#8658 by @kyle-hobbs, PR #9475 by @ggoranov-smar)
• Add native tool calling support for Mistral provider (PR #9625 by @hannesrudolph)
• Wire MULTIPLE_NATIVE_TOOL_CALLS experiment to OpenAI parallel_tool_calls for parallel tool execution (PR #9621 by @hannesrudolph)
• Add fine grained tool streaming for OpenRouter Anthropic (PR #9629 by @mrubens)
• Allow global inference selection for Bedrock when cross-region is enabled (PR #9616 by @roomote)
• Fix: Filter non-Anthropic content blocks before sending to Vertex API (#9583 by @cardil, PR #9618 by @hannesrudolph)
• Fix: Restore content undefined check in WriteToFileTool.handlePartial() (#9611 by @Lissanro, PR #9614 by @daniel-lxs)
• Fix: Prevent model cache from persisting empty API responses (#9597 by @zx2021210538, PR #9623 by @daniel-lxs)
• Fix: Exclude access_mcp_resource tool when MCP has no resources (PR #9615 by @daniel-lxs)
• Fix: Update default settings for inline terminal and codebase indexing (PR #9622 by @roomote)
• Fix: Convert line_ranges strings to lineRanges objects in native tool calls (PR #9627 by @daniel-lxs)
• Fix: Defer new_task tool_result until subtask completes for native protocol (PR #9628 by @daniel-lxs)

Release v3.34.5

[3.34.5] - 2025-11-25

• Experimental feature to enable multiple native tool calls per turn (PR #9273 by @daniel-lxs)
• Add Bedrock Opus 4.5 to global inference model list (PR #9595 by @roomote)
• Fix: Update API handler when toolProtocol changes (PR #9599 by @mrubens)
• Set native tools as default for minimax-m2 and claude-haiku-4.5 (PR #9586 by @daniel-lxs)
• Make single file read only apply to XML tools (PR #9600 by @mrubens)
• Enhance web-evals dashboard with dynamic tool columns and UX improvements (PR #9592 by @hannesrudolph)
• Revert "Add support for Roo Code Cloud as an embeddings provider" while we fix some issues (PR #9602 by @mrubens)

Release v3.34.4

[3.34.4] - 2025-11-25

• Add new Black Forest Labs image generation models, free on Roo Code Cloud and also available on OpenRouter (PR #9587 and #9589 by @mrubens)
• Fix: Preserve dynamic MCP tool names in native mode API history to prevent tool name mismatches (PR #9559 by @daniel-lxs)
• Fix: Preserve tool_use blocks in summary message during condensing with native tools to maintain conversation context (PR #9582 by @daniel-lxs)

Release v3.34.3

[3.34.3] - 2025-11-25

• Implement streaming for native tool calls, providing real-time feedback during tool execution (PR #9542 by @daniel-lxs)
• Add Claude Opus 4.5 model to Claude Code provider (PR #9560 by @mrubens)
• Add Claude Opus 4.5 model to Bedrock provider (#9571 by @pisicode, PR #9572 by @roomote)
• Enable caching for Opus 4.5 model to improve performance (#9567 by @iainRedro, PR #9568 by @roomote)
• Add support for Roo Code Cloud as an embeddings provider (PR #9543 by @mrubens)
• Fix ask_followup_question streaming issue and add missing tool cases (PR #9561 by @daniel-lxs)
• Add contact links to About Roo Code settings page (PR #9570 by @roomote)
• Switch from asdf to mise-en-place in bare-metal evals setup script (PR #9548 by @cte)

Release v3.34.2

[3.34.2] - 2025-11-24

• Add support for Claude Opus 4.5 in Anthropic and Vertex providers (PR #9541 by @daniel-lxs)
• Add support for Claude Opus 4.5 in OpenRouter with prompt caching and reasoning budget (PR #9540 by @daniel-lxs)
• Add Roo Code Cloud as an image generation provider (PR #9528 by @mrubens)
• Fix: Gracefully skip unsupported content blocks in Gemini transformer (PR #9537 by @daniel-lxs)
• Fix: Flush LiteLLM cache when credentials change on refresh (PR #9536 by @daniel-lxs)
• Fix: Ensure XML parser state matches tool protocol on config update (PR #9535 by @daniel-lxs)
• Update Cerebras models (PR #9527 by @sebastiand-cerebras)
• Fix: Support reasoning_details format for Gemini 3 models (PR #9506 by @daniel-lxs)

Release v3.34.1

[3.34.1] - 2025-11-23

• Show the prompt for image generation in the UI (PR #9505 by @mrubens)
• Fix double todo list display issue (PR #9517 by @mrubens)
• Add tracking for cloud synced messages (PR #9518 by @mrubens)
• Enable the Roo Code Cloud provider in evals (PR #9492 by @cte)

Release v3.34.0

[3.34.0] - 2025-11-21

• Add Browser Use 2.0 with enhanced browser interaction capabilities (PR #8941 by @hannesrudolph)
• Add support for Baseten as a new AI provider (PR #9461 by @AlexKer)
• Improve base OpenAI compatible provider with better error handling and configuration (PR #9462 by @mrubens)
• Add provider-oriented welcome screen to improve onboarding experience (PR #9484 by @mrubens)
• Pin Roo provider to the top of the provider list for better discoverability (PR #9485 by @mrubens)
• Enhance native tool descriptions with examples and clarifications for better AI understanding (PR #9486 by @daniel-lxs)
• Fix: Make cancel button immediately responsive during streaming (#9435 by @jwadow, PR #9448 by @daniel-lxs)
• Fix: Resolve apply_diff performance regression from earlier changes (PR #9474 by @daniel-lxs)
• Fix: Implement model cache refresh to prevent stale disk cache issues (PR #9478 by @daniel-lxs)
• Fix: Copy model-level capabilities to OpenRouter endpoint models correctly (PR #9483 by @daniel-lxs)
• Fix: Add fallback to yield tool calls regardless of finish_reason (PR #9476 by @daniel-lxs)

Release v3.33.3

[3.33.3] - 2025-11-20

• Add Google Gemini 3 Pro Image Preview to image generation models (PR #9440 by @app/roomote)
• Add support for Minimax as Anthropic-compatible provider (PR #9455 by @daniel-lxs)
• Store reasoning in conversation history for all providers (PR #9451 by @daniel-lxs)
• Fix: Improve preserveReasoning flag to control API reasoning inclusion (PR #9453 by @daniel-lxs)
• Fix: Prevent OpenAI Native parallel tool calls for native tool calling (PR #9433 by @hannesrudolph)
• Fix: Improve search and replace symbol parsing (PR #9456 by @daniel-lxs)
• Fix: Send tool_result blocks for skipped tools in native protocol (PR #9457 by @daniel-lxs)
• Fix: Improve markdown formatting and add reasoning support (PR #9458 by @daniel-lxs)
• Fix: Prevent duplicate environment_details when resuming cancelled tasks (PR #9442 by @daniel-lxs)
• Improve read_file tool description with examples (PR #9422 by @daniel-lxs)
• Update glob dependency to ^11.1.0 (PR #9449 by @jr)
• Update tar-fs to 3.1.1 via pnpm override (PR #9450 by @app/roomote)