Enhancement: Expose richer model metadata in /engines/v1/models (similar to OpenRouter)

[Stru-Publish]

Hello team,

I would like to propose an enhancement to the Model Runner API.

Current behavior
The /engines/v1/models endpoint currently returns only a minimal schema for each model (e.g., id, object, created, owned_by).

Proposed enhancement
Consider expanding the model schema to include additional metadata about each model — similar to what OpenRouter’s Models API exposes. OpenRouter’s API provides comprehensive metadata for each model in its standard response, including a human-readable name, description, context length, technical architecture (input/output modalities, tokenizer, instruct_type), pricing information and supported parameters. This metadata allows client applications and agent frameworks to understand a model’s capabilities and limitations without hard-coding them. Their documentation emphasises that the API is designed to make important information about LLMs freely available for integration.

Why it matters
Applications like E‑Worker Studio dynamically adjust UI capabilities based on the available model features. When only a model ID is provided, it’s difficult to know what modalities are supported (text, image, audio), maximum context window, or any special parameters. By exposing richer metadata, Model Runner could enable downstream tools to present more accurate options to users and better utilise models’ full capabilities. It also aligns the Model Runner API with emerging standards in the ecosystem.

[xenoscopic]

Hey @Stru-Publish, at the moment we keep the /v1/models endpoints compatible with the OpenAI API specification (which has a limited model object type). Adding additional fields there might break clients.

However, we do have an additional Docker-specific endpoint (GET /models) which exposes significantly more information. See, e.g., curl http://model-runner.docker.internal/models. The model object type we return there is defined here. More specifically, its "config" field offers more detailed parameters.

We could add additional information though (e.g. modalities and maximum context window size). And I think we're still filling in some of those fields for existing models.

CC @ekcasey and @ilopezluna since this might be more in their wheelhouse.

[ericcurtin]

@Stru-Publish just curious we are putting in effort to make contributing to docker model-runner more contributor friendly, would you be interested in working on this?

[Stru-Publish]

@ericcurtin how to contact you to get more info?

[ericcurtin]

We just set up a slack for this purpose:

https://app.slack.com/client/T0JK1PCN6/C09H9P5E57B

[Stru-Publish]

@ericcurtin the slack is private, add or email: info@eworker.ca

[ericcurtin]

@Stru-Publish it's supposed to be public, are you sure you can't access it?