ego lite is a browser purpose-built for AI agents to perform browser automation while allowing human users to work in parallel without interference. Written in JavaScript, it addresses a fundamental limitation in existing browser automation tools: the inability to share a single browser instance cleanly between human and AI workflows. Unlike frameworks such as browser-use and agent-browser that require separate browser instances and struggle with login persistence, ego lite is designed from the ground up as a shared environment where both user and agent operate simultaneously.
The core architecture centers on Spaces, which are parallel workspaces within the same browser instance. Each AI agent or task receives its own isolated Space, enabling multiple agents to multitask concurrently. A user can browse their own tabs in the main browser while Claude Code enriches leads in one Space and Codex scrapes competitor sites in another, all without collision or interference. This parallel execution model eliminates the traditional bottleneck where users and agents fight for control of the same tabs.
ego lite exposes its capabilities through the ego-browser skill, a connection layer that works with any agent CLI including Claude Code, Codex, Cursor, and custom implementations. Rather than relying on command-line interfaces that force agents into sequential decision loops, ego lite wraps browser capabilities as JavaScript functions that agents call directly. This code-based approach rather than CLI-based approach delivers measurable performance gains: complex workflows complete up to 2.5 times faster with higher task success rates and substantially fewer tool calls per task, as demonstrated in benchmarks against Vercel's agent-browser.
The browser inherits Chrome data on first launch, automatically migrating existing logins, cookies, extensions, and bookmarks to eliminate login friction. Users can optionally enable this migration during setup, after which agents immediately access real authentication credentials without additional configuration. The snapshot capability, enhanced through kernel-level customization, produces high-quality page representations that reliably handle challenging cases like deeply nested iframes where competing approaches consistently fail.
Currently available as a macOS application for both Apple Silicon and Intel architectures, ego lite requires zero cost and zero configuration. Installation options include downloading the native app, installing just the ego-browser skill via npx, or allowing an agent to trigger setup on first use. The browser stores all browsing data locally on the user's device, recording only whether Chrome migration was enabled during initial setup.
The roadmap indicates planned expansion to Windows and Linux platforms. A forthcoming feature called experience accumulation will distill successful agent actions into reusable tools and workflows, potentially accelerating similar future tasks by up to 5 times. The project is released under the MIT License, with community engagement occurring through Discord for questions and skill sharing, GitHub Discussions for longer threads and ideas, and X/Twitter for updates and releases. Documentation is available at lite.ego.app/document, covering tutorials, tool references, and integration guides for various agent platforms.
