This is the original multi-model chat endpoint. Still fully supported — nothing has been removed — but new integrations should prefer the named v1 endpoints.

POST /api/chat

If you're starting fresh

Use the v1 endpoints instead — they pick the behaviour from the URL so you don't have to remember which mode string to send:

Two models side-by-side → POST /api/v1/battle
Up to 8 models with voting → POST /api/v1/fight
Single model (OpenAI-compatible) → POST /api/v1/chat/completions

How it works

You tell the server which mode you want with a mode field in the request body.

{
  "message": "What's the best IDE?",
  "mode": "fight",
  "models": "claude,gpt,grok",
  "rounds": 1
}

Under the hood, each mode value maps 1:1 to one of the v1 endpoints:

Body `mode`	Same as the v1 endpoint
`"chat"`	`POST /api/v1/chat/completions`
`"battle"`	`POST /api/v1/battle`
`"fight"`	`POST /api/v1/fight`

Migrating to the new endpoints

No rush — both surfaces will keep working. When you do move over:

Swap the URL. /api/chat → /api/v1/battle, /api/v1/fight, or /api/v1/chat/completions, depending on what you were using.
Drop the mode field from the request body — the new endpoint name already says which mode it is.
Everything else (headers, auth, the rest of the body, the response shape) stays exactly the same.

Full reference

The full request/response shape, SSE event list, and parameter reference are documented on the Chat API page. Everything described there applies to /api/chat too, just with the mode field added.

Chat API (Legacy)

How it works

Migrating to the new endpoints

Full reference

On this page