Summary

ego-lite addresses the challenge of seamless collaboration between human users and AI agents within a web browser. Unlike traditional browser automation frameworks that require separate browser instances and struggle with credential management, ego-lite provides a unified browser environment where users and AI agents can operate concurrently without conflict. This allows AI agents to leverage existing user logins and session data, enabling them to perform complex tasks on authenticated websites efficiently. The primary benefit is accelerated task completion for AI agents, reduced token consumption, and an uninterrupted browsing experience for the user.

How It Works

ego-lite's core innovation lies in its architecture, which exposes browser functionalities as JavaScript functions directly callable by AI agents. This code-centric approach enables agents to compose multi-step workflows into single, efficient outputs, bypassing the iterative command-response cycles common in other automation tools. The browser features "Spaces," which are isolated, parallel workspaces for each AI agent or task. This separation ensures that agent activities do not interfere with the user's primary browsing session. This design results in significantly faster execution of complex tasks (up to 2.5x improvement over agent-browser), higher success rates, and fewer tool calls.

Quick Start & Requirements

• Platform: macOS only; Windows and Linux are on the roadmap.
• Installation:
  • Download the macOS application directly.
  • Alternatively, install the ego-browser skill using npx skills add citrolabs/ego-lite.
  • Agents can be instructed to set up ego-lite by referencing its GitHub repository and installation documentation.
• Onboarding: The first launch prompts users to migrate data from Chrome, including logins, cookies, extensions, and bookmarks, for agent reuse.
• Running Tasks: Activate the ego-browser skill within an agent CLI (e.g., /ego-browser) followed by a plain-language description of the desired task.
• Demo: A demo video is available [link omitted due to irrelevance].
• Documentation: Detailed tutorials and guides are available at lite.ego.app/document/.

Highlighted Details

• Code-based Capabilities: Agent interactions are via JavaScript functions, leading to faster runs and fewer tokens compared to CLI-based automation.
• Parallel Spaces: Each agent operates in its own isolated workspace, allowing multiple agents or tasks to run concurrently without interference.
• High-Fidelity Snapshots: Utilizes kernel-level customization for superior webpage snapshots, crucial for AI models, reliably handling complex elements like nested iframes.
• Universal Agent Control: The ego-browser skill acts as a bridge, exposing browser actions (snapshot, fill, click, navigate) as in-page JavaScript tools for any compatible agent.
• Experience Accumulation (Coming Soon): A planned feature to distill successful agent actions into reusable tools, aiming to accelerate similar future tasks significantly.

Maintenance & Community

• Community: Active community support is available via Discord, GitHub Discussions, and X/Twitter updates.
• Roadmap: Future development includes support for Windows and Linux operating systems.

Licensing & Compatibility

• License: The repository's contents are released under the MIT License.
• Browser: The ego-lite browser application itself is a separate, free download.
• Compatibility: Designed to integrate with various agent CLIs, including Claude Code, Codex, and Cursor, by exposing browser functionality as in-page JavaScript tools.

Limitations & Caveats

• Platform Support: Currently limited to macOS; Windows and Linux support are pending.
• Feature Maturity: The "experience accumulation" feature is marked as "coming soon" and is not yet implemented.
• Data Privacy: While browsing data remains local, the system logs only the user's opt-in status for Chrome data migration during setup.