Responses and context

Responses supports protocol continuations with state retained by Netter. It differs from portal history and stateless Chat Completions.

Create a response

curl --fail-with-body https://api.llm.netter.io/v1/responses \
  -H "Authorization: Bearer $NETTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"PUBLIC_MODEL_NAME","input":"Summarize the benefits of a work queue.","max_output_tokens":1024,"store":true}'

Choose a model with an active Responses mapping. Keep its returned identifier if you need retrieval or a context continuation.

Continue, retrieve and delete

Action	Operation
Continue	New POST /v1/responses with previous_response_id in the allowed context
Retrieve	GET /v1/responses/{id}
Delete	DELETE /v1/responses/{id}

Resources remain isolated by owner and tenant. Another user's identifier does not grant access. Continuations reconstruct the required complete input; Netter sends store=false upstream in this workflow.

Retention and cost

Stored state is encrypted with a default 30-day lifetime. Deletion removes the Netter-held resource. Each continuation keeps its own context and mapping; deleting one response does not automatically remove every other resource in the chain.

A continuation can reprocess reconstructed context. It is not free provider memory. Each request keeps its credit reservation and measured usage accounting.

Models and opaque artifacts

Maintain context-compatible models and parameters. Switchyard does not reclassify a bound continuation as a separate new conversation. If the target is revoked or a binding configuration changes, start a new compatible context.

Preserve signed and encrypted reasoning blocks without rewriting them. Protected artifacts may expire or become invalid after an internal key rotation.

Three forms of history

• Portal chat: conversations saved after completed answers.
• Chat Completions: clients submit the required context each time.
• Stored Responses: clients can reference authorized resources and continuations.

None provides automatic memory across independent conversations. An optional S3 archive is a separate export flow, not model context automatically reused later.