Chat Completions

For file inputs, upload through the Files API, then reference the returned file_id. The gateway converts the input; provider acceptance is determined upstream.

Use {"type":"file","file":{"file_id":"..."}} inside messages[].content. Images become image_url, videos the InOneAPI video_url extension, audio input_audio, and documents file.file_data. Keep ordinary text as type: text; content order is preserved.

Use POST /v1/chat/completions for OpenAI-compatible chat, tool calling, JSON output, and streaming. InOneAPI keeps the public request shape stable while routing by model to the selected provider mapping.

Endpoint

https://api.inoneapi.com/v1/chat/completions

# -H "X-APP-NAME: Your App Name"
# -H "X-APP-URL: https://your-app.example.com"
curl "https://api.inoneapi.com/v1/chat/completions" \
  -H "Authorization: Bearer $INONEAPI_API_KEY" \
  -H "Content-Type: application/json" \
  --data '{
  "model": "YOUR_MODEL_ID",
  "messages": [
    {
      "role": "user",
      "content": "Hello"
    }
  ]
}'

Parameters

Field	Type	Required	Details
model	string	yes	Public InOneAPI model ID. The gateway maps it to an enabled upstream model.
messages	array	yes	Ordered conversation messages. Roles normally include `system`, `user`, `assistant`, and `tool`.
temperature	number	no	Sampling randomness. Provider ranges differ; use 0-2 unless the model documents a narrower range.
top_p	number	no	Nucleus sampling. Usually set either `temperature` or `top_p`, not both.
max_tokens	number	no	Maximum generated tokens. Some providers use `max_completion_tokens`; mappings normalize where supported.
stream	boolean	no	When true, returns server-sent events with chat completion chunks.
tools	array	no	OpenAI-compatible tool definitions. Tool support depends on the selected model and mapping.
tool_choice	string/object	no	Controls automatic, required, or named tool calls.
response_format	object	no	Use for JSON object or JSON schema output when supported.
stop	string/array	no	Stop sequences. Support and maximum count are provider-specific.
user	string	no	End-user identifier for abuse tracing. Do not put secrets or personal data in this field.

Message content

Text-only messages can use a string content. Multimodal models may accept an array of content parts such as text and image_url; unsupported parts are rejected or transformed according to the model mapping.

{
  "role": "user",
  "content": [
    { "type": "text", "text": "Describe this image." },
    { "type": "image_url", "image_url": { "url": "https://example.com/image.png" } }
  ]
}

Response

Non-streaming responses follow the OpenAI chat completion shape. Usage fields are calculated locally when possible and aligned to provider token rules configured for the model.

{
  "id": "chatcmpl_...",
  "object": "chat.completion",
  "created": 1760000000,
  "model": "YOUR_MODEL_ID",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "TTFT is the time until the first token is received." },
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 18,
    "completion_tokens": 13,
    "total_tokens": 31
  }
}

Streaming

With stream: true, parse each data: line as a JSON chunk until data: [DONE]. The first chunk is the best client-side signal for TTFT, while the response header X-Gateway-Trace-ID links the request to gateway diagnostics.

Errors

Use 400 responses to fix invalid JSON, unsupported parameters, or mapping issues. Use 401/403 for key and permission checks. Use 429 with Retry-After for rate or budget pressure. Use 5xx and the trace ID when contacting support.

Do not retry streaming requests blindly after partial output.
Keep API keys server-side.
Log X-Gateway-Trace-ID, status, model, and timing, not prompt or completion content.

Response metadata

The response includes usage counters and a gateway trace ID for diagnostics.