Model APIs
Responses and conversation state
Use the Responses API and understand input, stream, store, previous_response_id, retrieval and deletion.
Updated
/v1/responses/v1/responses/{response_id}/v1/responses/{response_id}Create a response
POST /v1/responses uses input rather than the Chat Completions messages field. input can be text or an array of input items. This endpoint is intended for Responses clients and requires a compatible model and serving route.
| Parameter | Meaning |
|---|---|
| model | Required: model id visible to the key |
| input | Text or input items; supply a task for a new conversation |
| stream | true requests an event stream |
| store | Requests response storage, subject to deployment policy |
| previous_response_id | Continues an accessible, unexpired stored response |
curl --fail-with-body https://api.token4ai.cloud/v1/responses \
-H "Authorization: Bearer $TOKEN4AI_API_KEY" \
-H "Content-Type: application/json" \
--data '{
"model": "YOUR_MODEL_ID",
"input": "Explain why streaming improves perceived latency.",
"store": false
}'Output items and events
The REST output field is an array of output items; message content may include output_text parts. SDKs may expose an output_text convenience property, which should not be assumed to exist as a top-level REST field. Streaming clients must handle Responses event types and terminal status rather than reuse a Chat Completions delta parser unchanged.
Retrieve, continue and delete
GET /v1/responses/{response_id} retrieves an accessible stored response; DELETE on the same path removes it. Use previous_response_id only when the response was stored. Storage depends on deployment policy, tenant authorization and retention, so it is not a permanent chat database.
If state is unavailable or expired, resubmit the necessary context from your application. Set store:false for requests that should not create conversation state and define your own retention policy. This does not imply zero retention across all operational logs or providers.