LM Studio: Loaded models appear twice in Provider Configuration Profile

[fbuechler]

App Version

3.25.10

API Provider

LM Studio

Model Used

N/A

Roo Code Task Links (Optional)

No response

🔁 Steps to Reproduce

1. Start LM Studio (Windows 11) and load a model to be used with Roo Code
2. Open Roo Code settings -> Providers
3. Add a Configuration Profile with API Provider "LM Studio"
4. The loaded model is displayed twice in the model selection list: once with the model path and once with the model id

💥 Outcome Summary

Expected models loaded in LM Studio to appear only once in the Configuration Profile, but got them twice under different names.

📄 Relevant Logs or Errors (Optional)

This might be a bug in getLMStudioModels() (https://github.com/RooCodeInc/Roo-Code/blob/main/src/api/providers/fetchers/lmstudio.ts#L53) where model.path is used as key for downloaded models but model.modelKey is used as key for loaded models.

[roomote]

Hi @fbuechler! I'm here to help out the maintainers and am going to see if I can fix this issue. Thanks for the detailed bug report and for identifying the likely cause in the getLMStudioModels() function - that's very helpful! I'll investigate and work on a fix.

[roomote]

I've submitted a pull request to fix this issue: #6956

The fix ensures that LM Studio models use a consistent key (model.path) for both downloaded and loaded models, preventing duplicates from appearing in the Provider Configuration Profile.

The PR is currently undergoing CI checks. Once they pass and the PR is merged, this issue should be resolved.

[daniel-lxs]

After reviewing the attempted fix in PR #6956, I'm updating the scope for this issue.

The root problem is that we're fetching models from two different LM Studio APIs:

• listDownloadedModels() - shows all models on disk
• listLoaded() - shows models currently in memory

This causes duplicates when a model is both downloaded and loaded.

Updated approach: Only show loaded models.

Downloaded models that aren't loaded can't actually handle requests - they're not usable until LM Studio loads them into memory. There's no value in showing them in our model list.

Implementation:

• Remove the listDownloadedModels() call and its try/catch block (lines 74-83)
• Keep only the loaded models logic
• Use model.path as the consistent key
• Update tests accordingly

Benefits:

• Eliminates duplicates entirely (no deduplication needed)
• Only shows models that can actually process requests
• Provides accurate runtime metadata (contextLength vs maxContextLength)
• Simpler, more maintainable code

Note: This changes the UX slightly - users will need to load models in LM Studio before they appear in Roo Code. But this is already required to use them anyway.

Consider adding a migration for existing saved model IDs if the key format changes affect stored configurations.

[daniel-lxs]

@roomote-agent Create a PR based on my previous comment