Private AI. Native to Chrome. Ready for your framework.

Ship on-device AI
without the browser plumbing.

Production-ready components and composables for Chrome's Prompt API, writing tools, translation, Proofreader, and WebMCP. No inference server. No API key. No deep-reactivity slowdown.

  • MIT licensed
  • TypeScript-first
  • SSR-safe
  • Zero telemetry
Local AI session Ready
Turn this release note into three customer benefits.
Clearer decisions. Structured output keeps every response predictable.

Faster workflows. Streaming local inference responds without a network round trip.

Private by default. Content stays inside the browser's built-in model.
Ask the local model…
Running on-device with Gemini Nano 0 network requests
Prompt API Structured output Long-input handlingWebMCPVue + Nuxt

A production layer, not another model SDK

The native API is powerful.
The lifecycle is the hard part.

Chrome provides the model. Browser AI Kit handles the states and limits that turn a browser experiment into a feature people can depend on.

01

Every browser state has a product state.

Feature detection, downloads, progress, cancellation, quotas, context overflow, failure, and cleanup are reactive and typed.

02

Responsive while the model works.

Native sessions stay outside deep reactivity. Stream rendering is frame-coalesced so typing remains crisp during local inference.

03

Long input is a supported path.

Measure real browser quotas, split at safe boundaries, and merge with API-specific strategies instead of arbitrary character caps.

04

Choose the interface level that fits.

Start with accessible components, move to composables for a fully custom UI, and keep direct access to familiar native operations.

From install to local response in minutes

Small API. Serious defaults.

Drop in a complete chat interface or keep the UI entirely yours. Browser AI Kit stays close to Chrome's vocabulary, so the native documentation and your application code still line up.

Vue quick start
npm install @desource/browser-ai-vue

<script setup lang="ts">
import { PromptApi } from "@desource/browser-ai-vue";
import "@desource/browser-ai-vue/assets/lib.css";
</script>

<template>
  <PromptApi context-strategy="summarize" />
</template>
Production concernNativeKit
Download and readiness UXBuild itIncluded
Measured long inputBuild itIncluded
Prompt context recoveryBuild itIncluded
Nuxt SSR safetyManualIncluded
Accessible starter UINoneIncluded

Calling the native API directly is a good fit for one-off tasks. The kit earns its place when lifecycle, persistence, SSR, or long content matters.

Live in your browser

Eight surfaces.
One consistent experience.

Availability below is detected from this Chrome profile. Open an example to test the real local API—there is no mocked response or hosted fallback behind these demos.

Built for more than one ecosystem

Native to your framework.
Consistent underneath.

Vue, Nuxt, React, Angular, Svelte, and a framework-neutral TypeScript core ship from one repository and share one tested browser-behavior contract.

A privacy boundary users can understand

No inference endpoint hiding in the fine print.

The library does not send prompts, outputs, or telemetry to DeSource Labs. Chrome owns model execution and storage. If your app adds a hosted fallback, that becomes a separate product and consent decision.

Read the privacy and fallback guide →

Open source and ready to build on

Make the browser the AI runtime.

Start with a package guide, run every API against your local Chrome profile, or inspect the shared contracts that keep every framework adapter aligned.