Live & ShippedAutomation
Moteur anti-détection

Owl Browser

Self-hosted anti-bot browser engine for automation at scale

Présentation en français

Owl Browser est un environnement Chromium durci dédié à l'automatisation web et aux audits de sécurité, avec isolation des empreintes matérielles et pilotage via le protocole MCP.

Owl Browser is a self-hosted browser automation engine engineered to execute complex web workflows without getting blocked. It features 27 C++ override modules and 31 Chromium patches directly in the Blink core, spoofing navigator, canvas, WebGL, WebRTC, fonts, and audio at the engine level rather than through fragile JavaScript prototype hooks. Owl Browser includes OwlMark, a structured handle-addressable rendering view that cuts model token consumption by 10x compared to screenshot-based scraping. It also provides embedded vision models for local challenge verification and supports up to 256 concurrent contexts.

Spécification de déploiementPriorité matériel
Engine CoreChromium with custom Blink C++ patchset
Supported SDKsPython (owl-browser-sdk), Node.js (@olib-ai/owl-browser), REST, MCP
Context Cold-StartUnder 12ms per isolated session
Maximum ConcurrencyUp to 256 isolated contexts per process
Norme Zéro Télémétrie :Aucun traceur marketing, aucune télémétrie distante, et pleine propriété des données.
182+
Automation Tools
Covers clicks, inputs, network interception, and storage.
256
Parallel Contexts
Isolated sessions with independent cookie jars and cold-starts under 12ms.
10x
Token Reduction
OwlMark structured view cuts context tokens 10x vs screenshots.
38.5k
Median Tokens
Standardized MCP benchmark vs 73,420 tokens for Playwright MCP.
Spécifications vérifiées à partir du code source
Stealth Engine01

Blink Engine C++ Patches

27 C++ modules and 31 Chromium core patches spoof hardware signatures directly in C++, evading fingerprint detection systems like Cloudflare and Akamai.

Token Efficiency02

OwlMark Handle Addressing

Transforms pages into semantic, handle-addressable text maps. Models interact with elements directly by handle ID without transmitting massive image screenshots.

Local Vision AI03

Embedded Vision Verification

Includes an on-device Qwen3-VL-2B vision model via llama.cpp that solves visual verification challenges locally without external solve APIs.

Agent Native04

Built-in Natural Language Loop

The native browser_nla tool executes autonomous observe-act loops inside the browser. Achieved 18/18 correct on benchmark tasks at 4,844 median tokens.

Core Engineering Capabilities

Detailed feature checklist for Owl Browser, designed and audited for stability in real operational environments.

27 C++ override modules and 31 Blink patches for deep fingerprint stealth

OwlMark semantic handle view saving 10x context tokens compared to screenshots

Embedded Qwen3-VL-2B vision model for local interactive challenge resolution

Native MCP server with 79 tools across agent, automation, and webdev profiles

Drop-in compatibility with existing Playwright and Puppeteer codebases

Support for up to 256 concurrent contexts with cold-start under 12 milliseconds

Ingénierie sur mesure

Besoin de Owl Browser adapté pour votre entreprise ?

Chaque brique technologique peut être adaptée et déployée directement dans vos infrastructures privées avec pleine propriété du code.