Skip to content
LLMFLUX

API reference · OpenAI

Responses

OpenAI-format inference through your LLMFLUX gateway key.

Endpoint

POST /v1/responses

Send Content-Type: application/json and either Authorization: Bearer YOUR_KEY or x-api-key: YOUR_KEY. Authorization takes precedence. Use a server-side environment variable, never a browser bundle.

Request body

FieldBehavior
modelRequired gateway selector. An enabled canonical ID or configured alias from GET /v1/models.
inputOpenAI-format input, validated by the upstream provider.
streamOnly literal JSON true enables SSE. Otherwise the gateway explicitly sends false upstream.

Send input as a string or a Responses-format input array. A typical result contains output items and usage; OpenAI SDKs expose output_text as a convenience. Tool support and multimodal input depend on the upstream model. The gateway exposes POST only: response retrieval, deletion, conversations and stateful continuation are not independently implemented here.

Compatibility boundary. The gateway resolves the model and forwards the JSON body; it does not translate all model capabilities or validate every provider field. Discovery metadata is descriptive, not a guarantee that every endpoint or tool works for every model.

Example request

Replace YOUR_MODEL_ID with a compatible live ID. SDK examples read LLMFLUX_MODEL and LLMFLUX_API_KEY from your environment.

One gateway key reaches every enabled provider through this path.

curl https://api.llmflux.dev/v1/responses \
  -H "Authorization: Bearer $LLMFLUX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "YOUR_MODEL_ID",
  "input": "Hello from LLMFLUX",
  "stream": false
}'

Response and streaming

Successful non-streaming requests return HTTP 200 and the provider JSON payload, not a new gateway response schema. Token usage is recorded against your account. Set stream: true for SSE streaming.

Errors and limits

Inference requires an active key, any required Telegram verification, a positive account balance and available plan allowance. The per-user inference throttle is 120 requests per minute across keys. Bodies are limited to 25 MB. See error codes for handling 401, 402, 403, 404, 429 and provider failures.