Docs
Chat Completions
The most widely used path. Request and response shapes match OpenAI Chat Completions, so existing code works unchanged.
Request body
Only model and messages are required. Put a routing path in model and candidates get re-picked on every request.
| Field | Type | Required | Note |
|---|---|---|---|
| model | string | Yes | Model ID or route path |
| messages | array | Yes | system, user, assistant, tool |
| stream | boolean | No | SSE deltas, default false |
| tools | array | No | Tool-capable models only |
| max_tokens | integer | No | Capped by model limit |
| temperature | number | No | 0–2 |
Streaming
Set stream to true to receive deltas over SSE. Once streaming has begun there is no fallback; the request fails as-is.
Tool calls
Pass a tools array and only tool-capable models remain as candidates. If none qualify, the request stops with 400.
Image input
Mix image_url parts into content and candidates narrow to vision-capable models.
Cancelled streams and billing
Closing the connection mid-stream does not necessarily stop upstream generation. The gateway keeps reading the response to the end, within a time and output limit, and bills on the final usage the upstream reports. If a limit is hit first, it bills an estimate of what was generated up to that point.
service_tier
The service_tier parameter is accepted and ignored. Every request runs at the standard tier and is billed at standard prices.