Responses

POST /v1/responses

POSThttps://api.topbilling.in/v1/responsesscope chat:write

The OpenAI Responses API, for Codex CLI and client.responses.create in the official SDKs. It is stateless: send the whole conversation in input each time.

Authentication

Send your key as Authorization: Bearer tb_… or x-api-key: tb_…. See Authentication.

Parameters

modelstringrequired

A Top Billing model id from GET /v1/models.

inputstring | arrayrequired

A string, or items: message (roles user, developer, system, assistant; input_text and input_image parts), function_call and function_call_output.

instructionsstring

System instructions.

toolsarray

Function tools, flat: { type: "function", name, description, parameters }. Hosted tools are ignored.

tool_choicestring | object

auto, none, required or a named function.

max_output_tokensinteger

Upper bound on generated tokens; sizes the reservation.

temperature / top_pnumber

Sampling.

reasoning.effortstring

A hint for reasoning models; never rejected.

text.formatobject

json_schema for structured output, on models that support it.

streamboolean

Server-sent events: response.created … response.completed.

storeboolean

Accepted; responses are never stored, so it is reported as false.

Rejected with 400 unsupported_parameter: previous_response_id, conversation, background, prompt.

Request

shell

curl https://api.topbilling.in/v1/responses \
  -H "Authorization: Bearer $TB_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "<model-id>", "input": "Hello"}'

Errors

The same as Chat, except that an empty wallet or a spend cap answers 429 insufficient_quota, which is what OpenAI returns and what Codex stops on. A failure mid-stream arrives as a response.failed event.