Inference

Responses

Item-based output and explicit terminal states

Use Responses when your integration benefits from typed output items, structured events and an explicit completed or incomplete terminal state.

Create a response

Send input, model and supported response controls to POST /v1/responses. The route must preserve every required feature across its eligible providers.

curl https://api.octoryn.dev/v1/responses \
  -H "Authorization: Bearer $OCTORYN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"policy/frontier","input":"Return the top three risks."}'

Streaming events

Responses emits named events rather than Chat Completions chunks. Handle output-item deltas, response.completed, response.incomplete and response.failed distinctly.


Structured output

When a route declares JSON Schema support, request response_format and validate the final object again at the application boundary.


Do not infer success from HTTP 200 alone

A started stream may finish incomplete. Inspect the terminal event, incomplete_details and request correlation metadata before committing downstream work.

NextEmbeddings & rerank