feat: add support for Kimi K2 Thinking model embedded tool calls

[roomote]

Related GitHub Issue

Closes: #10064

Roo Code Task Context (Optional)

This PR was created by Roomote to address the issue.

Description

This PR adds support for extracting tool calls from Kimi K2 Thinking model responses when using the OpenAI Compatible provider.

The Kimi K2 Thinking model embeds tool calls in the reasoning_content field using special tags (<|tool_calls_section_begin|>, <|tool_call_begin|>, etc.) rather than the standard OpenAI tool_calls field. This caused the "no message content" error reported in the issue.

Key implementation details:

• Added kimi-tool-call-extractor.ts utility to parse embedded tool calls from reasoning content
• Updated the OpenAI handler to detect kimi-k2-thinking models and extract embedded tool calls
• Supports both streaming and non-streaming modes
• Handles multiple tool calls in a single response

Test Procedure

1. Unit tests added for the new kimi-tool-call-extractor utility covering:

   • Detection of embedded tool call markers
   • Extraction of single and multiple tool calls
   • Model name detection (case insensitive, various formats)
   • Cleaning of reasoning content after extraction

2. Existing OpenAI handler tests continue to pass

Run tests:

cd src && npx vitest run api/providers/utils/__tests__/kimi-tool-call-extractor.spec.ts
cd src && npx vitest run api/providers/__tests__/openai.spec.ts

Pre-Submission Checklist

• Issue Linked: This PR is linked to an approved GitHub Issue (see "Related GitHub Issue" above).
• Scope: My changes are focused on the linked issue (one major feature/fix per PR).
• Self-Review: I have performed a thorough self-review of my code.
• Testing: New and/or updated tests have been added to cover my changes (if applicable).
• Documentation Impact: I have considered if my changes require documentation updates (see "Documentation Updates" section below).
• Contribution Guidelines: I have read and agree to the Contributor Guidelines.

Screenshots / Videos

N/A - Backend API handling changes only

Documentation Updates

• No documentation updates are required.

Additional Notes

This implementation is based on the code example provided by @bozoweed in issue #10064 and the Kimi K2 Thinking tool call guidance.

Feedback and guidance are welcome!

Get in Touch

Roomote Bot

[roomote]

Rooviewer   See task on Roo Cloud

Review complete. No issues found.

The latest commit correctly addresses the feedback about handling tool call markers with or without <| |> delimiters:

• Detection function now checks for core marker text (tool_calls_section_begin)
• Regex patterns updated to make delimiters optional
• Test coverage added for both delimiter formats
• Changes are backward compatible

Previous reviews

• 464a7b5: Review #1

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}

[bozoweed]

take care some time it's formed as <|tool_calls_section_begin|> use only tool_calls_section_begin is working like a charmed

[bozoweed]

@roomote fix that please

[roomote]

Fixaroo   See task on Roo Cloud

Fixed the tool call marker detection to handle both <|marker|> and marker formats. All local checks passed.

_{View commit | Revert commit}

[bozoweed]

Date/time: 2025-12-17T07:22:48.014Z
Extension version: 3.36.9
Provider: openai
Model: moonshotai/kimi-k2-thinking

Unexpected API Response: The language model did not provide any assistant messages. This may indicate an issue with the API or the model's output.

any plan to work on that ..... you just add deepseek reasoner but seem that you don't support OpenAi Compatible for deepseek-reasoner .... please .... i hardly need to use that model on OpenAi Compatible way ....