Use the Responses API

Set BASE_URL to your HTTPS API origin without a trailing slash or /v1, and pass API_KEY from your credential store. OpenAI-compatible SDKs use ${BASE_URL}/v1; Anthropic-compatible SDKs use BASE_URL.

Responses is the main OpenAI-compatible inference surface. You need an API key, an available model ID, and credits. Generation can consume credits.

Send a request

printf 'Authorization: Bearer %s\n' "$API_KEY" \
  | curl -q -H @- -i "$BASE_URL/v1/responses" \
    -H "Content-Type: application/json" \
    -d '{
  "model": "gpt-5.6-sol",
  "input": "Say hello in one sentence.",
  "max_output_tokens": 64,
  "stream": false,
  "store": true
}'

A successful non-stream response has an id, model, status, output and usage. The generated words vary. Keep the HTTP request-id header for support without sending input text. API Activity shows recent run outcomes; its row ID differs from that header. A stored response can be retrieved with GET /v1/responses/{id} using the same organization's key. If you set store to false and do not bind a conversation, retrieval is unavailable.

Read the result

This is an example of a successful response shape. Your ID, text, and token counts will differ:

{
  "id": "resp_018f3f407b447cc4b8ba3f62656c299c",
  "object": "response",
  "created_at": 1790244000,
  "status": "completed",
  "incomplete_details": null,
  "model": "gpt-5.6-sol",
  "output": [
    {
      "id": "msg_018f3f407b447cc4b8ba3f62656c299c",
      "type": "message",
      "role": "assistant",
      "status": "completed",
      "content": [{ "type": "output_text", "text": "Done.", "annotations": [] }]
    }
  ],
  "output_text": "Done.",
  "usage": {
    "input_tokens": 12,
    "input_tokens_details": { "cached_tokens": 0 },
    "output_tokens": 5,
    "output_tokens_details": { "reasoning_tokens": 0 },
    "total_tokens": 17
  }
}

Check status first. For a simple text response, read output_text; for tools, citations, or multiple blocks, process each output item by type. usage describes protocol tokens. The finalized credit debit is on API Activity; do not derive a money price from total_tokens alone. With status: incomplete, inspect incomplete_details.reason and do not present partial text as a complete answer.

Retrieve the stored response

Use the id from the completed JSON response. The cURL example prompts for it; set UNLIMGPT_RESPONSE_ID in the environment for TypeScript, Python, or Go. This read does not start another generation:

printf 'Response id: '
read -r UNLIMGPT_RESPONSE_ID
printf 'Authorization: Bearer %s\n' "$API_KEY" \
  | curl -q -H @- -i "$BASE_URL/v1/responses/$UNLIMGPT_RESPONSE_ID"

HTTP 200 returns the stored response with its id, status, output, and usage; check that the id matches the one you entered. An unknown, deleted, or inaccessible response returns 404. The request above uses store: true; with store: false, a stateless response cannot be retrieved.

Storage and retry

The body id identifies a stored response or explicit cancellation; the HTTP request-id header is for support. GET /v1/responses/{id} requires a key from the same organization. Continuing with previous_response_id is not supported yet: include the required history in a new input. After a network interruption, retry the logical request with its original Idempotency-Key as described in safe retries.

For incremental output, see streaming.