Skip to main content

​
Generation endpoints

The gateway returns each protocol’s native JSON or stream events, without the console API’s { data, requestId } envelope. The model must declare the protocol you use. See API keys for headers.

​
Requests and responses

Use Content-Type: application/json and a platform model ID. Set a reasonable explicit output limit for generation requests. System One does not use generation-length parameters. GET /v1/models returns an OpenAI-style model list filtered by key permissions.

​
Streams and failures

Each protocol has its own stream format: Responses uses typed SSE events, Chat Completions uses delta chunks, and Messages and Gemini use their own event or candidate formats. Do not reuse a parser across protocols without adapting it. An interrupted connection does not prove the request was never executed. Keep response IDs and logs, then confirm execution state and usage before deciding what to do next. See Capabilities and limits and Troubleshooting.
This guide describes user generation APIs. Background execution, WebSockets, token counting, and compact require compatible channels. The documentation does not automatically execute API requests.