Napcar

Introduction

Build websites that run AI on the user's machine through Napcar's local models.

Napcar is a Chromium-based browser that lets your website run language-model features on the user's Mac through local models instead of cloud APIs.

Websites provide the interface. The user's machine provides the intelligence.

There are three ways to integrate, in order of preference:

  1. navigator.napcar.ai — the native API with the richest model routing and policy semantics. See JavaScript API.
  2. LanguageModel — a compatibility layer inspired by the Chrome Prompt API. See LanguageModel API.
  3. https://local.napcar.ai/v1 — an OpenAI-compatible endpoint synthesized inside the browser. See OpenAI-compatible endpoint.

All three route through the browser's permission and policy layer. A page can never reach the model backend directly.

Outside Napcar: cloud fallback

Other browsers have no local binding. The @napcar/sdk keeps one call site by running the local model in Napcar and falling back to a cloud provider of your choice (OpenAI by default, or any OpenAI-compatible API) everywhere else — the cloud config is ignored when the local model is available. See Cloud fallback.

Install the SDK (optional)

The @napcar/sdk package gives you typed helpers, graceful detection, and a polyfill fallback so your code degrades cleanly outside Napcar.

npm install @napcar/sdk

Or drop it in via a script tag — it exposes window.Napcar and never modifies navigator or window.LanguageModel:

<script src="https://cdn.jsdelivr.net/npm/@napcar/sdk/dist/napcar.global.js"></script>