What is ego(lite)
What ego(lite) is, and what it does better than browser extensions, cloud browsers, built-in agent browsers, browser CLIs, and AI browsers.
ego(lite) is a Chromium browser for macOS and Windows that you and your AI agents use together. You browse in it the same way you browse in Chrome. Your agents use the same browser to do web tasks for you, such as research, filling in forms, collecting data from websites, and testing web apps.
Agents control ego(lite) through the ego-browser Skill. The Agent needs to run local commands on the same computer. Claude Code, Codex, Cursor, Continue, Gemini CLI, Hermes Agent, OpenClaw, OpenCode, and custom Agents can use this route when their execution environment permits it. A web chat does not gain control of your local browser merely because you name the Skill.
Complete your first task
Install ego(lite), finish onboarding, load the Skill in your local Agent, and ask it to open a page and return a specific result. For an internal tool without an API, start with a read-only request such as reading one authorized record or downloading a report. Confirm the account and organization before allowing a change.
Use ego-browser to open our authorized internal report page. Confirm the account and organization shown on the page. Read the requested report and return its URL and visible fields. Do not modify any records.The browser keeps the authorized website session. The Agent decides what to do and receives page observations through your chosen model. Keeping the browser on your computer does not mean that page content stays out of that model's context. If a service already provides a suitable API or connector, that can be a simpler way to obtain its data.
Start with the macOS guide or Windows guide, then connect Claude Code or Codex.
How ego(lite) works
Each feature below has its own page under "Features".
Space
A Space gives an Agent its own task pages while your personal pages stay in place. You can inspect its work or take over when a step needs you. Our macOS focus check sampled another application in front during five browser actions; it did not test continuous keyboard focus or every permission prompt.
A Space is not a new window, a second Chrome profile, or a headless browser. All Spaces run inside the same ego(lite) browser, and agents can use the logins you already have there.
You can run many Spaces at the same time, usually one per task. How many depends on your computer's memory and CPU. You can open any Space to watch the agent, stop the task, or take over.
→ Space
Snapshot
A Snapshot is the text version of a page that the Agent reads. ego(lite) lists the elements on the page and gives each one a short reference, such as @3. Its size depends on the page and observation range. Use the viewport, full document, or a frame subtree according to what the task needs.
Calling page.snapshot() on a login page can return a structure like this:
Page: Example - Log in
URL: https://example.com/login
@1 [heading] "Log in"
@2 [form]
@3 [input type="email"] placeholder="Email"
@4 [input type="password"] placeholder="Password"
@5 [button type="submit"] "Continue"
@6 [link] "Forgot password?"The Agent can act on observed references, such as page.fill('@3', '...') and page.click('@5'). Read the Snapshot again after the page changes. Temporary references and selectors both need to match the current interface.
ego(lite) builds Snapshots inside its modified Chromium engine. The Snapshot guide shows a tested cross-origin iframe and open shadow root, along with a canvas case that needs a screenshot. A successful controlled test does not establish compatibility with every third-party widget.
→ Snapshot
ego-browser
ego-browser is the command-line tool that agents use to control ego(lite). The agent writes a short Node.js script, and ego-browser runs the whole script in one call:
ego-browser nodejs <<'EOF'
const task = await taskSpace('read example page');
const page = task.page('p1');
await page.goto('https://example.com');
console.log(await page.snapshot());
console.log({ url: await page.url(), title: await page.title() });
await task.finish({ keep: [] });
EOFThis invocation opens a page, reads it, returns its URL and title, and finishes the task. An Agent can combine known actions in one script and observe where the next decision requires new information. Compare complete tasks before drawing conclusions about speed or model cost.
Automatic Experience accumulation (limited beta)
Automatic Experience accumulation remains in limited beta. Do not assume that every successful task becomes an automatically saved workflow. You can save and rerun your own script today; that workflow still needs maintained selectors, valid account state, and result verification.
→ Skills
What ego(lite) does better than other browser automation tools
Most tools that let AI agents use a browser are one of five types. The table gives the short answer. The sections after it explain what ego(lite) does better than each type.
Compared with browser extensions
Examples: Claude in Chrome, the Codex Chrome extension, the Kimi Chrome extension.
With an extension, the agent sends commands to an extension installed in your Chrome, and the extension carries them out in your Chrome window. This works well for getting one thing done quickly on the page you are looking at, such as filling in a form. For a multi-step task, when you want to keep doing your own work while it runs, ego(lite) does better in these ways:
- No extra browser extension to install or pair. Install ego(lite), finish onboarding, and load the ego-browser Skill in your local Agent. Browser control uses that local runtime instead of an extension paired with your Chrome window.
- The agent does not take over your tabs. An extension works in the same window as you, so while you work, the agent may use the tab you are on or grab window focus. In ego(lite), each agent task runs in its own Space with its own tabs. The Agent operates those task pages while your personal pages stay in place. You can open the Space to inspect its work or take over when a step needs you.
- Separate pages for parallel tasks. Each task can use its own Space and output files. Our local macOS check ran two runtime invocations in different Spaces and verified different records during overlapping intervals. This does not establish maximum concurrency or account isolation. Extensions can also provide task windows or tab controls; compare the actual workflow.
Detailed comparisons: ego(lite) vs Claude in Chrome, ego(lite) vs Codex Chrome extension, ego(lite) vs Kimi Chrome extension.
Compared with cloud browsers
Example: Browserbase.
A cloud browser runs on the provider's servers, and your agent connects to it over the internet. This suits products that run a large number of browsers on servers, such as 100 or more at once, and tasks that must keep running while your computer is off. For tasks on your own accounts, run from your own computer, ego(lite) does better in these ways:
- Your logins stay on your computer. To use your own accounts in a cloud browser, you have to give the provider your passwords, cookies, or a saved signed-in session. ego(lite) runs on your computer. Your passwords, cookies, login sessions, and the data you imported from Chrome stay on your device instead of going to a hosted browser service.
- No charge for browser time. Cloud browsers usually charge per session or per minute. The ego(lite) app is free. Your model and network costs still apply.
- Fewer CAPTCHAs caused by unfamiliar IP addresses. A cloud browser sends traffic from the provider's IP addresses. Other customers share these addresses, and they may be in a different country from you, so some websites show more CAPTCHAs or security checks. ego(lite) uses your own internet connection and follows your VPN or proxy settings, so websites see the same kind of session they see when you browse yourself. Websites can still ask for a CAPTCHA or another security check, and ego(lite) does not bypass them.
Detailed comparison: ego(lite) vs Browserbase.
Compared with built-in agent browsers
Examples: the Codex browser, the browser built into Cursor.
Some AI apps include their own browser. It usually opens in a panel inside the app, and only that app's agent can use it. It needs no setup, and its separate login state keeps the agent away from your real accounts. When you want the agent to work with your real accounts in your usual browser, ego(lite) does better in these ways:
- Your logins are already there. A built-in browser often has its own login state, so you have to sign in to websites again inside it. With ego(lite), you can import your Chrome profile in one click during setup, including tabs, bookmarks, passwords, extensions, cookies, login sessions, and profiles.
- It is a complete browser. A built-in browser has fewer features than a full browser. ego(lite) is a complete Chromium browser with tabs, tab groups, windows, profiles, bookmarks, saved passwords, extensions, downloads, and incognito mode. You can use it as your regular browser while your agents work in the same browser.
- Several tasks can run at the same time. A built-in browser may limit how many tasks run at once. In ego(lite), each task runs in its own Space, and several Spaces can run at the same time.
- It works with any agent. A built-in browser only works inside its own app. Claude Code, Codex, Cursor, Gemini CLI, OpenClaw, and other agents all use ego(lite) through
ego-browser. You can change agents without changing browsers.
Detailed comparison: ego(lite) vs Codex browser.
Compared with browser CLIs
Examples: playwright-cli, Browser Harness, agent-browser, chrome-devtools-cli.
A browser CLI is a command-line tool that an AI agent runs to control a browser. Most are built on browser automation frameworks such as Playwright or Puppeteer, or directly on the Chrome DevTools Protocol. They work well when you want a fresh browser for every run, for example in scripts or CI. When an agent does tasks on real websites, ego(lite) does better in these ways:
- The Agent can use an authorized browser session. In ego(lite), import supported state or sign in directly, then confirm the website account. Other tools also offer persistent sessions, extensions, or existing-browser connections. Compare the actual configuration. Websites can expire a session or ask for verification under any route.
- The whole task takes fewer model turns. With these tools, the agent usually works one step at a time: it sends a command or a small group of actions, waits for the result, reads the page again, and then decides the next step. Each step costs a model turn. In ego(lite), the agent can run several actions in one call: open a page, read it, click, fill in a form, and report the result. Together with compact Snapshots, this means the whole task takes fewer model turns, so it finishes sooner and uses fewer tokens.
- You can see what the agent is doing. These tools often run the browser in headless mode, with no visible window. In ego(lite), you can open the agent's Space at any time to watch it, pause the task, or take over a step that needs you, such as a QR-code login.
In the Real-World Bench, we ran ego(lite) and the four browser CLIs above on the same 31 tasks on live websites, with the same model and the same scoring. ego(lite) completed 93.5% of the tasks with a perfect score, the highest rate of the five tools. The published results include the execution interface, model cost, and task times. Use those historical conditions when interpreting the comparison. They are separate from the new local documentation fixtures and do not measure Playwright MCP as a distinct entrant.
Detailed comparisons: ego(lite) vs Browser Harness, ego(lite) vs agent-browser, ego(lite) vs Playwright, ego(lite) vs Playwright MCP, ego(lite) vs Puppeteer.
Compared with AI browsers
Examples: Perplexity Comet, Dia.
An AI browser is a browser with the browser company's own AI assistant in a sidebar. The assistant runs on the company's AI service and works inside the browser. This works well for quick help with the page you are reading, such as a summary or a translation. When you want an agent to do real work across websites and your own computer, ego(lite) does better in these ways:
- Your agent can also work with your computer. The assistant in an AI browser usually cannot read your local files, run commands, or open your code repository. In ego(lite), the agent is your own agent, such as Claude Code or Codex, so it can use the browser and your computer in the same task. For example, it can review the UI of a web app you are building: open the page on your local dev server, find a layout problem, change the code in your repository, reload the page, and check the fix. An AI browser's assistant can look at the page, but it cannot change your code.
- The rules come from the agent you choose. The assistant in an AI browser follows the safety rules its company sets, and these rules are often stricter. For example, it may refuse a task that collects data from many pages. ego(lite) does not decide which tasks your agent may run. That depends on the agent you chose and the limits you set for it.
- You choose which model reads your pages. The assistant in an AI browser sends the pages it works on to the company's AI service. With ego(lite), page content goes to the model your agent uses. You choose that model, and it can be a local model that runs on your computer.
- No paid plan for agent features. AI browsers often limit their agent features by plan, and running tasks in the background or several at once can need a higher plan. The ego(lite) app is free, and each task runs in its own Space, so several tasks can run at the same time. You keep paying for your agent and model as you do today.
Privacy
ego(lite) runs on your computer. The passwords, cookies, history, bookmarks, extensions, and profiles you import from Chrome are stored there, as they are in Chrome, and they stay under your control.
When an agent works on a page, the agent reads the page content and sends it to the model it uses. ego(lite) runs the browser and passes commands and results between the browser and your agent.
For more on how data is handled, see the Privacy Policy.
You do not need an account or an email address to use ego(lite).
Platforms and price
ego(lite) is available for macOS and Windows. Use the platform-specific setup guide. This page does not announce Linux availability.
ego(lite) is free to download and use. There is no subscription, no per-task fee, and no charge for browser time. You use your own agent and your own model, and you pay for them as you do today.
Further reading
Why we built ego(lite) and how we designed it: A browser for you and your agents to work in parallel.