Chat Completions

Create a chat completion

POST /v1/chat/completions

OpenAI-compatible chat inference. Send the standard request body; the gateway maps it to the resolved provider and returns an OpenAI-shaped response with an extra provider field indicating which upstream served the request.

const completion = await client.chat.completions.create({
  model: "anthropic/claude-haiku-4-5",
  messages: [
    { role: "system", content: "You are concise." },
    { role: "user", content: "Explain MoE routing in one sentence." },
  ],
  temperature: 0.7,
});

Streaming

With stream: true the response is a text/event-stream of chat.completion.chunk frames, terminated by data: [DONE]. Set stream_options: { include_usage: true } to receive a final usage-only chunk.

const stream = await client.chat.completions.create({
  model: "openai/gpt-4o-mini",
  messages: [{ role: "user", content: "Count to five." }],
  stream: true,
  stream_options: { include_usage: true },
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

If the upstream fails before any bytes are sent, the gateway returns a normal JSON error with the right status. If it fails mid-stream, a terminal error frame is emitted followed by [DONE], matching how OpenAI clients expect mid-stream errors.

Maniac extensions

These are accepted alongside the OpenAI fields and ignored by stock SDKs.

Map-or-reject policy

OpenAI parameters without an internal analogue are handled predictably:

• Rejected — n > 1 returns 400 invalid_request_error.
• Dropped (request still runs) — logprobs, top_logprobs, input_audio content parts, and json_schema strictness.
• Passed through — unknown top-level params, so new OpenAI knobs keep working.

See the live schema and try requests in the API Reference.