Citrolabs Releases ego-lite: Browser Designed for Parallel Human-AI Web Automation

New Browser Architecture Separates Human and Agent Workspaces

Citrolabs has released ego-lite, a desktop browser built specifically to let users and AI agents operate simultaneously without conflict. Unlike existing automation tools that drive a separate browser instance, ego-lite creates isolated "Spaces" within a single browser—one for the human, one for each agent—so neither takes control away from the other.

The project targets developers using coding agents such as Codex or Claude Code. According to the project's GitHub repository, the browser runs on macOS and Windows, with Linux support listed on the roadmap.

How It Works

• Installation: Download the browser; on first launch it offers to migrate Chrome data (logins, cookies, extensions, bookmarks).
• Agent integration: The installer adds an `ego-browser` skill to the agent's local skills directory.
• Invocation: In the agent CLI, type `/ego-browser` followed by a natural-language task description.
• Execution: The agent opens the target page in its own Space, reads a structured Snapshot of the page, performs actions, and reports back—while the user's tabs remain untouched.

Reported Performance Claims

The developers state they benchmarked ego-browser against four other browser-automation tools on Real-World Bench, a suite of 31 complex tasks on live websites. All tools were driven by the same agent (`pi`) using the same model (`gpt-5.6-sol`, thinking effort max) and graded by the same judge against per-task rubrics. The project reports:

• Speed: 47%–63% faster task completion than the other tools tested.
• Success rate: Higher full-success rate (every requirement met).
• Token efficiency: Fewer tokens consumed per task.

Note: These figures come from the project's own benchmarking description. Independent reproduction has not been verified at publication time.

Privacy and Data Handling

The project states that all browsing data, cookies, and local browser state remain on the user's device. Telemetry is described as deliberately narrow—limited to product signals such as whether ego-lite is set as the default browser. The browser itself is a free download; the repository code is released under the MIT License.

Getting Started and Verification

1. Download: Visit the GitHub repository at `github.com/citrolabs/ego-lite` for the latest release assets.
2. Documentation: Tutorials, tool reference, and integration guides are published at `lite.ego.app/document/`.
3. Community support: Discord (setup help, skill sharing) and GitHub Discussions (ideas, longer threads) are linked from the repository.
4. Verify claims: Developers can run the Real-World Bench suite themselves or review the benchmark methodology described in the project documentation.

Practical Implications

• Parallel workflow: Users no longer need to surrender their browser session to an agent.
• State continuity: Agents inherit existing logins and cookies, reducing re-authentication friction.
• Cost control: Fewer tokens per task may lower API costs for teams running large-scale automation.
• Platform limits: Linux users must wait for the roadmap item; no timeline is published.

As with any early-stage developer tool, teams should evaluate stability, security posture, and compatibility with their existing agent stack before adopting in production.