Currently, when using vendor: "customendpoint" in chatLanguageModels.json (such as with OpenRouter or any OpenAI-compatible provider), models must be listed manually in the models array. This is tedious and error-prone when the provider offers many available models or updates its catalog frequently.
Request:
- Support optional, automatic model discovery for
customendpoint providers that expose a compatible model-list API (e.g., OpenAI's /v1/models).
- If
url points to a standard OpenAI-compatible endpoint and discovery is successful, VS Code should fetch and display available models automatically in the picker and config.
- Allow manual
models entries as an override or fallback, so users can customize/limit their view or supply capability details (tool calling, etc.) for specific models.
- If discovery fails (auth, network, unsupported schema), fall back gracefully to manual or prompt for user action.
Why this matters
- Providers like OpenRouter expose dozens (sometimes hundreds) of models via a models endpoint, but manual config is not scalable as models change.
- Auto-discovery reduces repetitive config, keeps lists in sync, and matches user expectations based on other provider experiences.
- This would align
customendpoint with capabilities offered by extensions and some built-in providers for model management.
Example:
- User sets up OpenRouter with
vendor: "customendpoint" and only configures the main endpoint and API key. VS Code should show all available and permitted models in the picker, just as the web portal does.
References
Thanks for considering!
Currently, when using
vendor: "customendpoint"inchatLanguageModels.json(such as with OpenRouter or any OpenAI-compatible provider), models must be listed manually in themodelsarray. This is tedious and error-prone when the provider offers many available models or updates its catalog frequently.Request:
customendpointproviders that expose a compatible model-list API (e.g., OpenAI's/v1/models).urlpoints to a standard OpenAI-compatible endpoint and discovery is successful, VS Code should fetch and display available models automatically in the picker and config.modelsentries as an override or fallback, so users can customize/limit their view or supply capability details (tool calling, etc.) for specific models.Why this matters
customendpointwith capabilities offered by extensions and some built-in providers for model management.Example:
vendor: "customendpoint"and only configures the main endpoint and API key. VS Code should show all available and permitted models in the picker, just as the web portal does.References
Thanks for considering!