Run AI agents in isolated browser spaces while keeping your own tabs active
AI agents like Codex or Claude Code can use your browser's login state and cookies to perform web tasks while you continue surfing in your own tabs. The agents operate in independent spaces within the same browser, so they do not interfere with your screen. This tool requires no separate browser installation and migrates your existing Chrome data for immediate use.
A browser shared by humans and agents
Existing browser automation tools act only as a bridge for agents to control a browser, which often requires launching a separate browser and fails to transfer login information reliably. This tool is designed from the start for use by both humans and AI. Agents perform tasks in their own independent 'spaces' while you use the browser freely in your own space. As a result, agents do not touch your tabs or take over your mouse pointer. On first run, migrating Chrome data transfers existing logins, cookies, extensions, and bookmarks without requiring additional configuration.
Faster and cheaper through code
The way agents manipulate the browser differs from traditional command-based approaches. This tool provides browser functions as JavaScript functions, allowing agents to write multiple steps as a single code block. This reduces the repetitive process of executing commands one by one and checking results, making complex tasks finish much faster. According to the README, this approach yields a higher task success rate and significantly reduces the number of tool calls and costs per task. Additionally, internal browser engine optimizations improve the quality of snapshots that convert pages to text, enabling stable handling of web pages with complex structures.
How multiple agents work simultaneously
Multiple AI agents can perform different tasks at the same time. For example, while Claude Code collects 10 lead information items, Codex can crawl 5 competitor sites in parallel. Since each agent moves only within its own space, they do not conflict with each other. Through a connection layer called ego-browser, various agent CLIs such as Claude Code, Codex, and Cursor can control this browser. Agents only need to call the necessary tools, such as page snapshots, input, click, wait, navigate, and capture.
Can you try it right now
It is currently available on macOS, with Windows in closed beta and Linux on the roadmap. You can install it by downloading the macOS app, running the npx skills add citrolabs/ego-lite command, or pasting an installation instruction into an agent. Data is stored only on your device, and the information collected is limited to minimal product signals, such as whether default browser settings are enabled. It is released under the MIT license, and the browser app itself is free to download. It is suitable for people who frequently perform complex web automation tasks and want to delegate them to AI while safely maintaining their login state.
