A shared browser for AI agents: parallel automation, shared logins

ego (lite) is a browser built from the ground up for AI agents to run browser automation. It lets you and your agents work in the same browser simultaneously: your tabs stay yours, while each agent gets an isolated Space for its own tasks. It is free, requires zero configuration, and exposes your real logged-in Chrome state to agents through a skill called ego-browser. Today it runs on macOS.

Browser automation frameworks like Browser-Use or Vercel's agent-browser are libraries: they drive a separate browser, logins rarely carry over cleanly, and you end up fighting with the agent over the same tabs. AI browsers like ChatGPT Atlas or Perplexity Comet ship a built-in agent, so only that agent can use the browser. ego lite is one browser designed for sharing: your real profile, your real logins, and any agent you bring (Claude Code, Codex, Cursor, etc.).

Rather than exposing a CLI where the agent alternates between issuing commands and reading output, ego lite wraps browser capabilities as JavaScript functions the agent calls directly. The agent composes a multi-step task into a single script, which keeps tool calls down and lets complex workflows finish faster. The README reports up to 2.5× faster runs and higher task success rates compared with the conventional CLI approach on complex tasks.

Each agent gets its own fully isolated Space inside the same browser. Multiple agents can run at once — for example, Claude Code enriching 10 leads while Codex scrapes 5 competitor sites — without colliding or stealing your tabs. You can see which Space is active, take it over, or stop it. The browser's kernel-level customization also produces high-quality page snapshots, including for deeply nested iframes where other tools typically fail.

ego-browser is the glue between any agent CLI and the ego lite browser. It exposes the page as a set of in-page JavaScript tools — snapshot, fill, click, wait, navigate, capture — and the agent writes a JavaScript snippet that calls these tools. ego-browser then runs the snippet on the page in one pass. The repository ships plugins for Claude and Codex (.claude / .codex skill directories), plus an agent-skills spec under spec/.

Install the macOS app (Apple Silicon or Intel), or add just the skill with npx skills add citrolabs/ego-lite. You can also paste a setup prompt into your agent and let it install for you. On first launch, ego lite asks whether to migrate your Chrome data; say yes and the agent inherits your existing logins, cookies, extensions, and bookmarks. A typical task in your agent CLI looks like:

ego-browser follow @ego_agent on x.com for me

The agent picks up the skill, opens the page in its own Space, reads a Snapshot, acts on the page, and reports back — while your own tabs stay untouched.

The README's comparison table positions ego lite against Browser-Use, Vercel agent-browser, ChatGPT Atlas, and Perplexity Comet. The differentiators are parallel multitasking, reusable skills, inheriting Chrome data, separate workspace in the same browser, compressed semantic input, external-agent control, local data, no login friction, daily-use browser, and free. Against Vercel's agent-browser on four complex tasks, ego lite finished each up to 2.5× faster with substantially fewer tokens. The comparison chart lives in docs/assets/ego-vs-agent-benchmark.png.

ego lite currently runs only on macOS; Windows and Linux are on the roadmap. Your browsing data stays on your device — ego lite only records whether you opted into Chrome migration during setup. The 'experience accumulation' feature, which distills successful actions into reusable tools to make later tasks up to 5× faster, is marked as coming soon. The browser app itself is a separate free download; the repository code is MIT-licensed.