camofox-browser - Open source AI Agent anti-detection browser - AiBoss

What is camofox-browser?

Camofox Browser is an open-source AI Agent anti-detection browser server built on Camoufox (a heavily modified C++ fingerprint version of Firefox). It provides AI with structured webpage snapshots (reduced in size by 90%+) via a REST API, generating stable element references (such as e1, e2) for direct AI interaction. Camofox Browser incorporates C++-level fingerprint forgery, bypassing mainstream anti-scraping verification methods like Cloudflare; it supports cookie import, session persistence, and 15+ search macros.

The main functions of camofox-browse

C++ level anti-detection browsing: The underlying implementation is a modified version of Firefox called Camoufox. In the C++ implementation layer, it forges fingerprint information such as hardware concurrency, WebGL renderer, AudioContext, screen geometry, and WebRTC, which cannot be detected by the JavaScript layer.
Structured webpage snapshot: Extracting the original HTML into a numbered accessibility snapshot (reducing size by over 90%), AI can directly... e1,e2 Wait for stable references and page interactions.
Cookie import and session persistence: It supports importing Netscape format cookie files to achieve login-free access. User session cookies and localStorage are automatically persisted to the local machine and the login status is maintained after restart.
Search macro shortcuts: Built-in search macros for 15+ mainstream platforms (such as...) @google_search,@reddit_search,@youtube_searchA single command directly reaches the target page, and the Reddit macro returns JSON data.
Proxy and GeoIP are automatically matched: It supports residential proxies and Backconnect rotating proxies, and automatically sets the browser's time zone, language, and geographic coordinates based on the proxy's outgoing IP address to maintain fingerprint consistency.
YouTube subtitle extraction: Integrates yt-dlp to quickly extract subtitles from any YouTube video without requiring an API key and supports multi-language filtering.
VNC Visual Interactive Login: Manually complete interactive login for complex websites in a browser using noVNC, and export the stored state for subsequent automated reuse by the Agent.
Session tracing and debugging: Supports enabling Playwright tracking per session, generating screenshots, DOM snapshots, and network requests. .zip Documents are available for future review.
Lightweight deployment and elastic scaling: It uses approximately 40MB of memory when idle, supports one-click deployment with Docker, Fly.io, and Railway, and features lazy loading of the browser engine and automatic shutdown when idle.

The technical principle of camofox-browse

C++ layer fingerprint spoofing: Camoufox modifies Firefox source code. navigator.hardwareConcurrencyProperties such as WebGL rendering strings, AudioContext fingerprints, screen resolution, and WebRTC are forged by anti-scraping systems before the JavaScript is executed.
Accessibility Tree Snapshot: Playwright's accessibility tree is used to extract the semantic structure of the page, generating a textual snapshot with stable references, which is then passed to the LLM instead of the original DOM.
Playwright browser context isolation: Each user corresponds to an independent BrowserContextCookies, localStorage, and cache are completely isolated to avoid session interference.
Lazy loading and idle garbage collection: The Camoufox engine (approximately 300MB) is started on the first request, and the process is automatically shut down after 5 minutes of inactive sessions to reduce persistent resource consumption.
Structured JSON logs: All requests and responses are output as a single-line JSON object, containing... reqId It facilitates log aggregation and tracing in a distributed environment.

How to use camofox-browse

Install dependencies: implement npm install @askjo/camofox-browserThe first installation will automatically download the Camoufox engine, which is about 300MB.
Local startup: Run after cloning the repository npm install && npm startThe service listens by default. http://localhost:9377.
Configure access keys (production environment): Generate a 32-bit hexadecimal key and set environment variables. CAMOFOX_ACCESS_KEYAll requests must include Authorization: Bearer <key>.
Importing cookies to achieve login-free access: Place the Netscape format cookie file exported by the browser into ~/.camofox/cookies/ The directory allows you to inject sessions by calling the import tool via API or the OpenClaw plugin.
Create a tab and get a snapshot: Towards /tabs Send a POST request to create a tab, then call... /tabs/:id/snapshot Get the belt e1,e2 A summary of the structured pages referenced.
Perform page interaction: pass /tabs/:id/click or /tabs/:id/type Pass in the element reference number and text content to enable click and input operations.
Navigate using search macros: Call /tabs/:id/navigate And transmitted "macro": "@google_search" and "query": "关键词"Directly access the search results page.
Docker Deployment: Execute in the project root directory make upThe Makefile will automatically detect the CPU architecture and pre-download the binary files, and the build will start in about 30 seconds.

The core advantages of camofox-browse

Bypassing mainstream anti-scraping and CAPTCHA measures: Since the fingerprint is forged at the C++ layer, it can reliably pass bot detection on platforms such as Google and Cloudflare, avoiding frequent triggering of human verification.
Tokens are extremely efficient: Structured snapshots reduce the size by more than 90% compared to the original HTML, significantly reducing the LLM context window footprint and API call costs.
Element references are stable and reliable: After minor adjustments to the page layout,e1,e2 References can remain valid through the accessibility tree, reducing the probability of the Agent failing to operate due to DOM changes.
Out-of-the-box search capabilities: 15+ search macros cover mainstream information sources, eliminating the need for an agent to construct search URLs and parse pages; sites like Reddit directly return structured JSON.
Production-level safety model: Cookie import is disabled by default (required) CAMOFOX_API_KEYIt supports global Access Key authentication, path traversal protection, runtime injection of all sensitive data, and zero hard-coded keys.

The project address for camofox-browse

GitHub repositoryhttps://github.com/jo-inc/camofox-browser

Comparison of similar products to camofox-browse

Dimension	Camofox Browser	Puppeteer / Playwright	Selenium WebDriver
Anti-detection capability	C++ layer fingerprint forgery, natively bypassing Cloudflare and Google verification.	The stealth plugin needs to be installed separately; the plugin itself may become the new fingerprint.	There is no native anti-detection feature; it requires the use of an external fingerprint browser or proxy pool.
AI interaction efficiency	Structured accessibility snapshot, size reduced by 90%+, with stable element references.	To receive the complete HTML, you need to parse it yourself or take a screenshot and send it to the AI for recognition.	Returning to the complete HTML results in high parsing costs and lacks a native element numbering mechanism.
Deployment resource consumption	Approximately 40MB of free space, lazy loading + automatic shutdown.	The Chromium process is resident and consumes a lot of memory.	A persistent browser instance consumes a lot of resources.
Cookie/Session Management	Built-in Netscape Cookie import, session persistence, and VNC interactive login	Cookie injection and context management code needs to be written manually.	Profile and Cookie injection need to be configured manually.
Search shortcuts	Built-in search macros for 15+ platforms; Reddit directly returns JSON.	There are no built-in search macros; you need to construct the URL and parsing logic yourself.	No built-in search macro
Proxy Integration	Native support for residential agents and Backconnect, automatically synchronizing timezone/language/coordinates.	Proxy parameters need to be manually configured; there is no automatic GeoIP matching.	Proxy needs to be manually configured; there is no automatic GeoIP matching.
Applicable Scenarios	AI Agent Anti-Detection Browsing, Cloud Automation, Production-Grade Data Acquisition	Front-end testing, general web page automation, web crawler development	Traditional end-to-end testing, complex enterprise-level testing frameworks

Application scenarios of camofox-browse

AI Agent Web Automation: Provides a programmable browser backend for LLM-based agents to perform tasks such as data capture, form filling, and information retrieval.
E-commerce and competitor monitoring: Bypassing platform anti-scraping mechanisms, it automatically captures product prices, inventory, and review information, and supports account-level data collection even when logged in.
Social media automation: After importing cookies, automatically manage accounts on Twitter, LinkedIn, Instagram, etc., and perform content publishing, interaction, and data analysis.
Academic and public opinion research: Use search macros to quickly retrieve information from Google, Wikipedia, and Reddit, and extract YouTube video subtitles for content analysis.
Cloud-based automated testing: Deployed in a VPS or container environment, it serves as a headless browser service to support end-to-end testing and webpage monitoring in the CI/CD process.