OpenClaw with Firefox

OpenClaw is an AI agent framework that controls browsers via CDP. By default, it uses Chrome — which is trivially detectable by anti-bot systems. With foxbridge, OpenClaw can use Camoufox (anti-detect Firefox) instead, with zero code changes.

Why Switch from Chrome?

AI agents browsing the web with Chrome face immediate problems:

Detection: Anti-bot systems (Cloudflare, DataDome, Akamai) fingerprint Chrome’s automation markers. Headless Chrome is detected in milliseconds.
No stealth: Chrome has no fingerprint spoofing. Every agent context looks identical — same WebGL hash, same canvas fingerprint, same user agent.
Memory: Chrome uses 50-80MB per browser context. At scale (50+ agents), this becomes prohibitive.

Camoufox solves all three: undetectable fingerprints, per-context identity, and 10-15MB per context.

Setup

1. Install Foxbridge


go install github.com/VulpineOS/foxbridge/cmd/foxbridge@latest

2. Start Foxbridge with Camoufox


foxbridge --binary /path/to/camoufox --port 9222 --headless

3. Configure OpenClaw

In your openclaw.json:


{
  "browser": {
    "cdpUrl": "ws://127.0.0.1:9222",
    "enabled": true,
    "headless": true
  }
}

That’s it. OpenClaw now browses through Camoufox. Every page visit, click, and form fill happens in an anti-detect Firefox context.

Multiple Agents with Different Fingerprints

Each browser context created by OpenClaw gets a unique Camoufox fingerprint:


// Agent 1 — unique UA, WebGL, canvas fingerprint
const ctx1 = await browser.createBrowserContext();
 
// Agent 2 — completely different identity
const ctx2 = await browser.createBrowserContext();

The fingerprints come from BrowserForge — a Bayesian network trained on real browser populations. Each context gets:

Unique user agent string (OS-matched)
Unique screen resolution
Unique WebGL renderer/vendor (from a database of 147 GL parameters)
Unique canvas and audio fingerprint seeds
OS-matched font list

At Scale with VulpineOS

For production deployments with many agents, VulpineOS provides a complete management layer:

Embedded foxbridge — one Camoufox process, one CDP server, hundreds of agents
Context pool — pre-warmed contexts for instant agent startup
Per-agent identity — fingerprints, proxies, cookies managed automatically
Injection protection — strips hidden DOM text that could hijack agent instructions
Action Lock — freezes pages during agent thinking to prevent mutation
Cost tracking — per-agent token budgets with alerts
Web panel — manage everything from a browser dashboard


# VulpineOS handles foxbridge automatically
./vulpineos
# OpenClaw agents connect via the embedded CDP server on :9222

Comparison

	OpenClaw + Chrome	OpenClaw + Foxbridge + Camoufox
Bot detection	Detected immediately	Undetectable
Fingerprints	All contexts identical	Unique per context
Memory per context	~50-80MB	~10-15MB
Setup	None (default)	`foxbridge --binary camoufox --port 9222`
Injection protection	None	VulpineOS AX filter