Chrome Built-in AI Models in the Browser

Chrome's Gemini Nano - zero-download AI for summarization, translation, and text generation, built directly into the browser.

Overview

The Chrome Built-in AI family is available through the @localmode/chrome-ai provider, with models that are 0 MB to download (they ship with or are downloaded by Chrome itself). The package supports three tasks: summarization, translation, and text generation (via the Prompt API), and all three use the same LocalMode interfaces as every other provider.

Running Chrome Built-in AI locally in the browser eliminates API costs, removes network latency, and keeps all user data on-device. After Gemini Nano is available on the device, inference is instant and works offline. Each API targets a different task - choose based on your application's requirements and verify Chrome version compatibility.

Architecture and History

Chrome Built-in AI is a unique provider in LocalMode's ecosystem: it uses Google's Gemini Nano model that ships directly with Chrome, requiring zero model downloads from your application. The @localmode/chrome-ai package provides summarization, translation, and text generation (Prompt API) capabilities that work instantly - no waiting for model downloads, no storage quota consumed by your app, no user bandwidth used.

The Summarizer API condenses long text into concise summaries with configurable output length, type, and format. The Translator API handles text translation between supported language pairs. The Prompt API (window.LanguageModel) exposes Gemini Nano as a full LanguageModel, enabling generateText(), streamText(), and generateObject() - the same functions that work with WebLLM or wllama. All three APIs use the same LocalMode interfaces as other providers.

The Summarizer, Translator, and Language Detector APIs became stable in Chrome 138. The Prompt API (text generation via window.LanguageModel) became stable for web pages in Chrome 148; in Chrome 127–137 it was available only via origin trial for extensions.

The key trade-off: Chrome Built-in AI only works in Chromium-based desktop browsers (Chrome 138+ or Edge 138+). It does not work in Firefox, Safari, or mobile Chrome. A simple try/catch pattern handles this gracefully - try Chrome AI first (instant, zero-cost), and on failure fall back to @localmode/transformers models (broader support, requires download). This hybrid approach gives users the best possible experience on their browser.

Variant Comparison

The following table lists every Chrome Built-in AI variant available through @localmode/chrome-ai, with their real model IDs as used in the provider.

Model ID	Interface	Size	Speed	Quality	Chrome Version
`chrome-ai:gemini-nano-summarizer`	`SummarizationModel`	0 MB (built-in)	Fast	Good	138+
`chrome-ai:gemini-nano-translator`	`TranslationModel`	0 MB (built-in)	Fast	Good	138+
`chrome-ai:gemini-nano`	`LanguageModel` (Prompt API)	0 MB (built-in)	Fast	Good	148+ (web)

Note: "0 MB" means your application downloads nothing - Gemini Nano is managed by Chrome itself. The initial Gemini Nano download (~1.5 GB) is handled by Chrome in the background, not by your application.

Size Distribution

Size Range	Count
Under 200MB	3	variants

How to choose a variant: All three variants use the same underlying Gemini Nano model. Pick based on task: gemini-nano-summarizer for condensing text, gemini-nano-translator for language translation, and gemini-nano (Prompt API) for open-ended text generation. Check Chrome version requirements - the Prompt API requires Chrome 148+ for web pages, while Summarizer and Translator work from Chrome 138+.

Provider-Specific Code Examples

All Chrome Built-in AI variants use the same LocalMode interfaces from @localmode/core. Switching between providers requires changing only the import and model factory - no application logic changes.

Summarization

import { summarize } from '@localmode/core';
import { chromeAI } from '@localmode/chrome-ai';

const { summaries } = await summarize({
  model: chromeAI.summarizer({ type: 'key-points', length: 'short' }),
  texts: ['Long article text here...'],
});
console.log(summaries[0]);

Translation

import { translate } from '@localmode/core';
import { chromeAI } from '@localmode/chrome-ai';

const { translations } = await translate({
  model: chromeAI.translator({ sourceLanguage: 'en', targetLanguage: 'de' }),
  texts: ['Hello, world!'],
});
console.log(translations[0]);

Text Generation (Prompt API - Chrome 148+)

import { generateText } from '@localmode/core';
import { chromeAI } from '@localmode/chrome-ai';

const { text } = await generateText({
  model: chromeAI.languageModel({ systemPrompt: 'You are concise.' }),
  prompt: 'Explain quantum tunnelling in one sentence.',
});
console.log(text);

Fallback Pattern

For maximum browser compatibility, attempt Chrome AI first and fall back to @localmode/transformers for browsers that don't support it:

import { summarize } from '@localmode/core';
import { chromeAI } from '@localmode/chrome-ai';
import { transformers } from '@localmode/transformers';

// Try Chrome AI (instant, zero-download), fall back to Transformers.js
let model;
try {
  model = chromeAI.summarizer();
} catch (error) {
  console.warn('Chrome AI not available, using Transformers.js fallback:', error);
  model = transformers.summarizer('Xenova/distilbart-cnn-6-6');
}

When to Use Chrome Built-in AI

Chrome Built-in AI models are a strong choice when:

You need zero-download AI - No model files are served or cached by your application; Gemini Nano is managed entirely by Chrome.
Your users are on Chrome/Edge desktop - All three APIs require Chrome 138+ (or Edge 138+) on desktop. Firefox, Safari, and mobile Chrome are not supported.
Summarization or translation are your primary tasks - The Summarizer and Translator APIs are stable from Chrome 138 and optimized for those tasks.
Text generation on Chrome 148+ - The Prompt API (LanguageModel) became stable for web pages in Chrome 148 and provides full generateText() / streamText() support.

Hardware and OS Requirements

Chrome Built-in AI has specific device requirements because Gemini Nano runs on-device:

Requirement	Detail
Browser	Chrome 138+ or Edge 138+ (desktop only)
OS	Windows 10/11, macOS 13+ (Ventura), Linux, ChromeOS (Chromebook Plus)
Disk space	At least 22 GB free on the volume containing your Chrome profile
GPU	More than 4 GB VRAM, or
CPU	16 GB RAM + 4 or more cores (CPU-only path)

Mobile (Android, iOS) and Incognito mode are not supported.

HuggingFace Model Cards

Chrome Built-in AI models are not hosted on HuggingFace - they are distributed and managed by Google via Chrome. See the Chrome Built-in AI developer documentation for model details.

Chrome AI Overview - full provider guide including feature detection and fallback patterns
Chrome AI Language Model - Prompt API usage, streaming, and structured output
Chrome AI Summarization - Summarizer API with type/format/length options
Chrome AI Translation - Translator API

Methodology

Facts on this page were verified against two primary sources: (1) the @localmode/chrome-ai package source code at packages/chrome-ai/src/ (model IDs, interfaces, and Chrome version references in error messages and comments), and (2) the official Chrome for Developers documentation fetched directly (API availability by Chrome version, hardware/OS requirements, Prompt API stable release version). The Prompt API stable-for-web version (Chrome 148) was confirmed via the Chrome I/O 2026 announcement page. Hardware requirements (22 GB disk, >4 GB VRAM or 16 GB RAM + 4 cores) are sourced directly from the Chrome "Get started" guide as of May 2026.