Browser AI is happening. Chrome shipped Gemini Nano. Brave has Leo with BYOM. Edge has Copilot. Safari has... Apple Intelligence somewhere.
Every browser is making different choices. Understanding these choices matters for developers building AI features and users trying to use them.
Chrome: Gemini First
What shipped (Chrome 131):
// Chrome's built-in AI APIs
const session = await ai.languageModel.create();
const response = await session.prompt("Hello");
const summarizer = await ai.summarizer.create();
const summary = await summarizer.summarize(text);
Model: Gemini Nano (on-device) User choice: None—you get Gemini Availability: Chrome-only, specific hardware requirements
Technical details:
- Runs locally via Chrome's model runtime
- ~2GB download (one-time)
- Requires specific CPU/GPU capabilities
- Not available on all devices
The Google strategy:
- Make Gemini ubiquitous
- Establish API patterns (before standards exist)
- Capture developer mindshare
- Reduce incentive to use alternatives
Concerns:
- No provider choice
- Chrome-only (not a web standard)
- Google controls updates
- Data collection unclear
Brave: BYOM (Bring Your Own Model)
What shipped (Brave Leo):
Settings → Leo AI:
├── Claude (default)
├── Claude Instant
├── Llama 3
├── Mixtral
├── Ollama (local)
└── Custom endpoint
Models: Multiple cloud + local User choice: Full control Availability: Brave browser only
Technical details:
- Sidebar assistant (not API for websites)
- Reverse proxy for privacy (strips identifying info)
- Ollama integration for local models
- Custom endpoint support
The Brave strategy:
- Differentiate on privacy and choice
- Don't lock users to one provider
- Support local models natively
- Build on existing "privacy browser" brand
What's missing:
- No standard API for websites
- No permission model (can't request from websites)
- Brave-specific (not portable)
Edge: Copilot Integration
What shipped:
Copilot sidebar integrated into Edge:
- Chat assistant
- Page summarization
- Content generation
- Powered by GPT-4/GPT-4o
Model: OpenAI (via Microsoft) User choice: Limited (Copilot or nothing) Availability: Edge, Windows
Technical details:
- Microsoft account integration
- Syncs with Windows Copilot
- Enterprise management options
- Uses Microsoft's API infrastructure
The Microsoft strategy:
- Integrate Copilot everywhere
- Drive Microsoft account adoption
- Enterprise AI adoption
- Compete with Google on AI-in-browser
Concerns:
- Locked to Microsoft ecosystem
- Requires Microsoft account for full features
- Enterprise focus may limit consumer features
Firefox: Waiting
Current state: No built-in AI features (as of late 2024)
What exists:
- WebGPU support (for third-party AI)
- Translation features (ML-powered)
- Experimental AI features in Nightly
Mozilla's stated approach:
- "We're exploring AI thoughtfully"
- Privacy-first considerations
- Open to partnerships and standards
The opportunity:
- User-controlled AI (choose provider)
- Privacy-first implementation
- Open standards leadership
- Differentiation from Google
The risk:
- Being left behind
- Users switch to "AI-enabled" browsers
- Perception of being outdated
Safari: Apple Intelligence
Current state (macOS Sequoia, iOS 18):
Apple Intelligence features:
- Writing tools
- Notification summaries
- Image generation (limited)
- Siri improvements
For Safari specifically:
- Limited browser-specific features
- No API for websites
- Tied to Apple devices
Model: Apple's on-device + optional ChatGPT User choice: Apple AI or (opt-in) ChatGPT Availability: Apple Silicon devices only
Apple's strategy:
- On-device privacy
- Tight ecosystem integration
- Don't expose raw AI API to websites
- Control the experience
Concerns for web developers:
- No standard API
- Apple-only features
- Can't rely on Safari AI from web apps
Arc: Browse for Me
What shipped:
"Browse for me" feature:
- AI summarizes search results
- Answers questions from web content
- Integrated into address bar
Current status: Arc company pivoting focus, future unclear
Model: Cloud AI (unspecified) User choice: None Availability: Arc browser
What it showed:
- AI-first browsing is viable
- Users want proactive AI
- Integration can be seamless
The Standards Gap
┌─────────────────────────────────────────────────────────────┐
│ Browser AI Approaches │
├──────────┬──────────────┬────────────┬──────────────────────┤
│ Browser │ Built-in AI │ User Choice│ Standard API │
├──────────┼──────────────┼────────────┼──────────────────────┤
│ Chrome │ Gemini Nano │ None │ Chrome-only (ai.*) │
│ Brave │ Leo (multi) │ Full │ None (sidebar only) │
│ Edge │ Copilot │ Limited │ None │
│ Firefox │ None │ N/A │ None │
│ Safari │ Apple Intel. │ Limited │ None │
│ Arc │ Yes │ None │ None │
└──────────┴──────────────┴────────────┴──────────────────────┘
What's missing:
- No standard API (navigator.llm)
- No cross-browser compatibility
- No user choice in most browsers
- Web developers can't rely on any approach
For Web Developers
Current Options
Option 1: Target Chrome only
if (window.ai?.languageModel) {
// Chrome's API
const session = await ai.languageModel.create();
}
Risk: Only works in Chrome
Option 2: Use extensions
if ('llm' in navigator) {
// WebLLM or similar extension
const response = await navigator.llm.prompt(input);
}
Risk: Users must install extension
Option 3: Backend AI
const response = await fetch('/api/ai', {
method: 'POST',
body: JSON.stringify({ prompt: input })
});
Risk: You pay for AI, no user choice
Recommended Approach
async function getAIResponse(prompt) {
// Prefer user's browser AI (most user-friendly)
if ('llm' in navigator) {
return await navigator.llm.prompt(prompt);
}
// Try Chrome's API (Chrome users without extension)
if (window.ai?.languageModel) {
const session = await ai.languageModel.create();
return await session.prompt(prompt);
}
// Fallback: your backend
const res = await fetch('/api/ai', {
method: 'POST',
body: JSON.stringify({ prompt })
});
return res.json().then(d => d.response);
}
This prioritizes user choice, then Chrome's built-in, then your backend.
What Users Should Know
If You Want Choice
Best: Brave Leo with Ollama
- Install Brave browser
- Set up Ollama (local AI)
- Configure in Leo settings
- Full control, privacy
Alternative: WebLLM extension + any browser
- Install WebLLM extension
- Configure your AI provider
- Works on any site that supports it
If You Want Simplicity
Easiest: Chrome with Gemini
- Just use Chrome 131+
- AI features "just work"
- No setup required
- Limited to Google's AI
If You Want Privacy
Best: Local models via Ollama + Brave/Extension
- Data never leaves device
- No cloud dependency
- Full control
What Needs to Happen
For Users to Win
- Open standard (
navigator.llm) adopted by browsers - User choice in all browsers (not locked to vendor)
- Local model support built into browsers
- Permission model users understand (like camera/location)
For Developers to Win
- Single API that works across browsers
- Graceful degradation when AI isn't available
- No vendor lock-in in their code
- Cost model that works (user-powered ideal)
For the Web to Win
- Standards process (W3C/WHATWG)
- Multi-stakeholder buy-in (not just Google)
- Privacy by default (local options)
- Competition on quality (not lock-in)
Timeline Estimate
2024: Current state (fragmented) 2025: Standards discussions begin, Firefox experiments 2026: Initial cross-browser implementations 2027: Stable, standardized browser AI
This mirrors WebGPU's timeline (~4 years from experiments to stable).
Conclusion
Browser AI is inevitable. The question is what form it takes:
Vendor-controlled: Each browser ships its own AI, no interoperability, users locked in
Open standards: navigator.llm or similar, user choice, cross-browser compatibility
The 2024 landscape is clearly vendor-controlled. Chrome has Gemini. Edge has Copilot. Brave is the exception with BYOM.
For the web to remain open, we need standards. Otherwise, "browser AI" becomes "Google AI" for 65% of users.
The next few years will determine the outcome.
This landscape analysis will be updated as browsers evolve. Last updated: November 2024.