ego lite - A Browser Where AI Agents and You Work in Parallel

ego lite is a browser built for AI agents to drive. Its GitHub repository describes it as “the fastest browser for AI agents to run browser automation,” and more usefully as “a browser where you and your AI agents work in parallel.”

That second phrase is the actual point. Browser automation normally forces a choice: either the agent gets its own headless browser with none of your logins, or it drives the browser you are using and fights you for tabs. ego lite is built by Citro Labs around a third option.

Spaces, and why they matter

The mechanism is called a Space — described in the documentation as “parallel workspaces inside the same browser.” Each agent or task gets “its own fully isolated Space,” while your own tabs stay separate.

So a single browser instance hosts your browsing and several agents at once, without collisions. Multiple agents can work simultaneously in different Spaces. That is a genuinely sensible answer to a real annoyance, and it is the feature most worth evaluating the project on.

On first launch the browser inherits your existing Chrome data — “existing logins, cookies, extensions, and bookmarks.” That solves the other half of the problem: an agent operating in a fresh headless browser is logged into nothing, which is precisely why so many automation tasks stall on a sign-in wall.

Code instead of a command loop

The second design decision concerns how the agent controls the browser. Rather than a step-by-step command-and-response loop, ego lite exposes “JavaScript functions the agent calls directly.” The reasoning given in the README:

The agent gets to do what it does best: write code, composing a multi-step task into a single output

This is the same shift seen elsewhere in agent tooling — let the model emit one program rather than fifty round trips. It tends to cut both latency and token consumption, and it is the basis of the project’s performance claims.

Those claims should be read as the project’s own. The README reports that across four complex automation tasks, ego lite “finished each task up to 2.5× faster, with substantially fewer tokens” than Vercel’s agent-browser. That is a vendor benchmark on a self-selected task set, not an independent result — a reasonable signal of intent, not a number to plan against. A further “experience accumulation” feature, claimed to make similar tasks “run up to 5x faster,” is listed as coming soon rather than shipped.

The project also states that kernel-level customisation produces “the highest-quality page snapshots,” including handling of “deeply nested iframes” — the perception side of the same problem, since an agent can only act on what it can see.

Getting it running

ego lite works with “any agent CLI (Claude Code, Codex, Cursor, or a custom one)” through a skill called ego-browser, which the browser installs into every agent’s skills directory on your machine.

The repository can be added to an agent directly:

npx skills add citrolabs/ego-lite

The browser itself is a separate download, published as .dmg builds for Apple Silicon and Intel.

The two things to check before you invest

It is macOS only. The documentation is upfront that “Windows and Linux are on the roadmap,” but today that is a hard gate. If your automation runs on a Linux box or in CI, this is not currently a candidate.

“Open source” needs a qualifier. The repository is MIT licensed, but it does not contain the browser. It holds the ego-browser skill, the agent plugin definitions, specs, and documentation. The README draws the line explicitly: “The contents of this repository are released under the MIT License… The ego lite browser is a separate, free download.”

So the integration layer is open and inspectable; the browser you actually run is a free binary whose source is not in this repository. That is a meaningful distinction if your interest in open source is about auditability or long-term control rather than price — and given that the browser inherits your cookies and logged-in sessions, auditability is not an abstract concern here.

Practical considerations

Credential blast radius. Inheriting your Chrome profile is what makes the tool useful and also what makes it worth thinking about. An agent working in a Space is operating with your live sessions. Consider which accounts you want reachable that way.
Young project, fast-moving. Treat interfaces and behaviour as subject to change, and pin what you depend on.
Zero-cost, zero-configuration is the stated setup story, which lowers the bar for simply trying it on a task you already have.

Verdict

ego lite is one of the more thoughtful takes on agent browser automation: Spaces address a real conflict between human and agent use of the same browser, and calling JavaScript directly is a better fit for how models work than a command loop. Weigh it against two facts rather than the benchmark headline — it is macOS-only today, and the MIT licence covers the integration skill, not the browser binary that inherits your logins. If those are acceptable, the entry cost is a single npx command and a download.