POST /v1/chat/completions

The main inference endpoint. Same request and response shape as OpenAI; inference.club proxies the request to one of your online providers and streams the response back.

Request

curl https://api.inference.club/v1/chat/completions \
  -H "Authorization: Bearer $INFERENCE_CLUB_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3-8b",
    "messages": [
      {"role": "system", "content": "You are concise."},
      {"role": "user", "content": "Say hello."}
    ],
    "temperature": 0.2
  }'

The body is forwarded to the provider's LLM server unchanged, so any OpenAI-compatible field your local server understands (temperature, top_p, max_tokens, tools, response_format, …) just works.

model is required and must match a model name advertised by one of your online providers. Lookup is exact-match.

Buffered response

{
  "id": "chatcmpl-abc123",
  "object": "chat.completion",
  "created": 1729960000,
  "model": "qwen3-8b",
  "choices": [
    {
      "index": 0,
      "message": {"role": "assistant", "content": "Hello."},
      "finish_reason": "stop"
    }
  ]
}

Streaming

Add "stream": true to the request body. The response becomes a Server-Sent Events stream of OpenAI-format delta chunks:

data: {"choices":[{"index":0,"delta":{"role":"assistant"}}]}

data: {"choices":[{"index":0,"delta":{"content":"Hello"}}]}

data: {"choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: [DONE]

inference.club passes the upstream stream through with no buffering, so chunks arrive as fast as your provider's local server can produce them.

Error responses

The 404 response uses inference.club's error envelope:

{
  "error": {
    "message": "No online provider serving model 'foo' for this user.",
    "type": "no_provider"
  }
}