OpenAI-compatible endpoint

Napcar synthesizes an OpenAI-compatible endpoint inside the browser. No DNS lookup or network request happens — the request is intercepted and served by the local model backend.

const response = await fetch("https://local.napcar.ai/v1/chat/completions", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    "X-Napcar-LLM-Handoff": "required",
  },
  body: JSON.stringify({
    model: "napcar-default",
    messages: [{ role: "user", content: "Summarize this text locally." }],
    stream: false,
  }),
});

const completion = await response.json();
console.log(completion.choices[0].message.content);

Endpoints

• GET /v1/models
• POST /v1/chat/completions (streaming + non-streaming)
• POST /v1/embeddings
• POST /v1/responses (best-effort subset)

Streaming

With stream: true, the response is Server-Sent Events compatible with standard Chat Completions clients, terminated by data: [DONE].

data: {"id":"chatcmpl_napcar_…","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"Napcar"},"finish_reason":null}]}

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

data: [DONE]

Behavior

• The endpoint requires the origin's local-AI permission, like the native API.
• Authorization headers are ignored and never logged.
• Model names are generic aliases by default; the exact model is hidden.
• Errors use the OpenAI error shape with a napcar_local_ai_error type and a Napcar error code, mapped to the appropriate HTTP status.