Authentication

POST /v1/chat/completions requires Authorization: Bearer <API_KEY>. Gateway base URL: https://api.misar.dev.

The gateway accepts the standard OpenAI chat completions request and returns an OpenAI-shaped response, so the stock OpenAI SDK works with only a changed base URL. Under the hood, requests are routed through the cascading MoE engine.

POST/v1/chat/completions

Generate a chat completion. Set stream: true for an incremental text/event-stream; otherwise a single JSON object is returned.

Request body

modelstringbody

Model id to route to. Defaults to the gateway's configured DEFAULT_MODEL (e.g. qwen3:8b) when omitted.

messagesArray<{role, content, name?}>bodyrequired

Conversation messages. content is a string, or an array of content parts (text, image_url, file) for multimodal input. image_url accepts a data: URI or a remote URL; file accepts { filename, data (base64), mime_type }.

max_tokensintegerbody

Maximum tokens to generate. Defaults to the gateway OPENAI_MAX_TOKENS (4096).

temperaturenumberbody

Accepted for compatibility. Routing does not force a temperature — each provider applies its own default.

streambooleanbodydefault: false

When true, responses stream as OpenAI SSE chunks ending with data: [DONE].

toolsArray<object>body

OpenAI-format tool/function definitions.

tool_choicestring | objectbody

Controls tool selection, forwarded to the router.

reasoning_effortstringbody

Reasoning-effort knob: none, minimal, low, medium, high, or xhigh. When unset, the effort is auto-derived from the last user message (floored at medium). A nested reasoning: { "effort": "..." } object is also accepted; the top-level field wins.

Response fields (non-streaming)

idstring

Completion id, e.g. chatcmpl-....

objectstring

chat.completion (non-streaming) or chat.completion.chunk (streaming).

modelstring

The model actually used to serve the request.

choicesArray<{index, message, finish_reason}>

The generated choice(s). message is { role: "assistant", content }.

usage{ prompt_tokens, completion_tokens, total_tokens }

Token accounting for the request.

Request

curl https://api.misar.dev/v1/chat/completions \
  -H "Authorization: Bearer $MISARCODER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3:8b",
    "messages": [{ "role": "user", "content": "Write a haiku about caching." }],
    "reasoning_effort": "low"
  }'

200 — Completion

{
  "id": "chatcmpl-8f2a1c9d4e6b7a3f9c1d2e5b",
  "object": "chat.completion",
  "created": 1753900000,
  "model": "qwen3:8b",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "Bytes rest in warm store..." },
      "finish_reason": "stop"
    }
  ],
  "usage": { "prompt_tokens": 14, "completion_tokens": 17, "total_tokens": 31 }
}

Streaming chunks

data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"Bytes"},"finish_reason":null}]}

data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{},"finish_reason":"stop"}]}

data: [DONE]

Tool calls

When the model calls a tool, streaming chunks carry delta.tool_calls (with the function name, then streamed arguments fragments) and the stream ends with finish_reason: "tool_calls".

Status codes

200 — Completion returned (or stream opened).
401 — Missing or invalid API key.
502 — Upstream model error.
503 — Model router not initialized.
504 — Request timed out.

Chat Completions

Tool calls

Status codes

On this page