feat: add GLM model detection for LM Studio and OpenAI-compatible providers

[roomote]

Related GitHub Issue

Closes: #11071

Description

This PR addresses the questions raised in issue #11071 about GLM model detection and sharing Z.ai optimizations with LM Studio and OpenAI-compatible endpoints.

Question 1: How do you detect whether LM Studio or the OpenAI-compatible endpoint is serving a GLM model?

This PR introduces an isGlmModel() utility function that detects GLM models by checking if the model ID contains common GLM patterns:

• glm- prefix (official naming: glm-4, glm-4.5, glm-4.7)
• glm4 (compact naming without dash)
• chatglm (older ChatGLM models)

Question 2: Is there a way so that when improvements and fixes are made to how Roo Code communicates with GLM models on Z.ai, they will also become available to LM Studio and OpenAI-compatible endpoints running GLM models?

Yes! This PR enables automatic application of Z.ai optimizations for GLM models detected on LM Studio and OpenAI-compatible endpoints:

1. mergeToolResultText: Merges text content after tool_results into the last tool message, preventing GLM models from dropping reasoning_content
2. disableParallelToolCalls: Disables parallel_tool_calls by default for GLM models since they may not support it

Key Implementation Details

• Created src/api/providers/utils/model-detection.ts with:
  • isGlmModel(modelId): Detects GLM models from model ID
  • getGlmModelOptions(modelId): Returns GLM-specific configuration options
• Modified LmStudioHandler to detect GLM models and apply optimizations
• Modified BaseOpenAiCompatibleProvider to detect GLM models and apply optimizations
• Added comprehensive test coverage for model detection

Test Procedure

1. Unit tests for model detection utility:

   • Run: cd src && npx vitest run api/providers/utils/__tests__/model-detection.spec.ts
   • Tests verify GLM detection for various model name formats
   • Tests verify non-GLM models are not falsely detected

2. Manual testing (for users with GLM models):

   • Configure LM Studio or OpenAI-compatible endpoint with a GLM model
   • Start a task that uses tools
   • Verify the model no longer gets stuck in file read loops

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.

Documentation Updates

• No documentation updates are required.

Additional Notes

This implementation follows the same pattern used by Z.ai for GLM model handling. The key insight is that Z.ai uses convertToZAiFormat() with mergeToolResultText: true, which is functionally equivalent to using convertToOpenAiMessages() with the same option. This PR enables this option automatically when a GLM model is detected.

Future Z.ai improvements for GLM models can be easily extended to third-party providers by:

1. Adding new detection patterns to isGlmModel() if needed
2. Adding new configuration options to GlmModelOptions interface
3. Updating providers to use the new options

Feedback and guidance are welcome!

---

Important

Adds GLM model detection and Z.ai optimizations for LM Studio and OpenAI-compatible providers, with tests for model detection.

• Behavior:
  • Adds isGlmModel() and detectGlmModel() in glm-model-detection.ts to identify GLM models by model ID patterns.
  • Applies Z.ai optimizations for GLM models in BaseOpenAiCompatibleProvider and LmStudioHandler.
  • Disables parallel_tool_calls and enables mergeToolResultText for GLM models.
• Implementation:
  • Modifies BaseOpenAiCompatibleProvider to detect GLM models and apply optimizations during construction and stream creation.
  • Modifies LmStudioHandler to detect GLM models and apply optimizations during message creation.
  • Adds logGlmDetection() for logging detection results.
• Testing:
  • Adds tests in glm-model-detection.spec.ts to verify GLM model detection and configuration.

^{This description was created by for d010015. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Rooviewer   See task on Roo Cloud

Reviewed all changes since last review. No issues found.

The GLM model detection implementation is clean, well-documented, and thoroughly tested:

• detectGlmModel() correctly identifies GLM models from various ID formats (standard, prefixed paths, GGUF files, ChatGLM)
• Version detection (4.5, 4.6, 4.7) and variant detection (base, air, flash, vision, etc.) work correctly
• Both LmStudioHandler and BaseOpenAiCompatibleProvider properly integrate GLM-specific optimizations
• mergeToolResultText and disableParallelToolCalls flags are correctly applied for GLM models
• GLM-4.7 thinking mode is properly supported
• Test coverage is comprehensive with tests for positive cases, negative cases, and edge cases

Previous reviews

• 1379ba4: Review #1

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