Support for OpenAI-compatible API server

[shamil]

As I understand it, the API server isn't OpenAI-compatible right now. Are there any plans to change that?

It would be great to have a standard /v1/chat/completions endpoint so we can plug third-party UIs like Open WebUI straight into the API Server.

[elgreco247]

Why don't you start your agent as an MCP server?
https://docs.openwebui.com/features/extensibility/mcp/

[tdabasinskas]

Why don't you start your agent as an MCP server? https://docs.openwebui.com/features/extensibility/mcp/

I would argue MCP is not the same - it abstracts things, so the chat is not really interactive.

We are also interested in this feature for Cogvel - I would be happy to contribute if you think that's something that could be accepted upstream.

[shamil]

Why don't you start your agent as an MCP server?
https://docs.openwebui.com/features/extensibility/mcp/

Yes, I can use MCP or even mcpo, but that is mainly to expose the agent as a tool. What I mean is being able to use the agents as models, not as tools. So we would get a TUI like experience, but in a Web UI.

A practical example:

I have an agent called pr-manager with the gemini-3-flash model wired to it. If I could serve it as an OpenAI API server, the /models endpoint would show this agent as a model, like pr-manager/gemini-3-flash, and then I could interact with it like with any other model. This would effectively make it close to how the current TUI works.

[elgreco247]

Have you seen https://github.com/tubone24/docker-cagent-playground ?