How to Let AI Agents Share Your Browser Login: A Practical Guide to ego-lite

How to Let AI Agents Share Your Browser Login: A Practical Guide to ego-lite

As a backend developer, I recently hit a very real pain point:

My company asked me to compile the listing status of a batch of competitors in a backend admin system. Each time, it was the same cycle: log in → check data → take a screenshot. Doing this a dozen times manually felt like a waste. Naturally, I thought about asking an AI agent to handle this grunt work. But here’s the catch: AI agents don’t have my login sessions. I could either hand over my credentials (a huge security red flag), or manually log in first and then hand off control (cumbersome). Frameworks like browser-use often require spinning up a separate browser instance, and passing cookies around usually breaks things.

That’s when I discovered ego-lite on GitHub (now boasting 2.5k+ stars). It’s built specifically to solve this problem: a browser "shared" by you and your AI agent. You keep using your normal tabs, while your AI agent works in its isolated Space using the exact same login sessions. Zero interference, zero complex setup.

In this tutorial, I’ll walk you through installing and configuring ego-lite from scratch in under 10 minutes, and demonstrate a real-world scenario: having an AI agent automatically log into an admin system, scrape data, and compile a report for you.

Prerequisites

• Operating System: Currently macOS only (Apple Silicon & Intel supported). Windows and Linux are on the roadmap.
• AI Agent CLI: You’ll need an agent that can call external skills, such as Claude Code, Codex, or Cursor.
• Google Chrome (Optional but recommended): If you’re already logged into the target websites in Chrome, migrating your data will let the agent inherit those sessions instantly.

Step 1: Install ego-lite

ego-lite consists of a desktop browser application + an ego-browser skill package. There are three ways to install it, but I highly recommend the first one for the smoothest experience:

Option 1: Download the macOS Installer (Recommended)

Choose the version that matches your chipset:

• Apple Silicon (M1/M2/M3/M4): Download .dmg
• Intel: Download .dmg

Double-click the .dmg file to install. This one-step process automatically registers the ego-browser skill in your agent’s skill directory, making it immediately callable.

Option 2: Install the Skill via npx

npx skills add citrolabs/ego-lite

This command only installs the ego-browser skill, not the browser app itself. The first time your agent tries to run a browser task, it will prompt you to install the ego-lite app.

Option 3: Let Your Agent Install It for You

Simply paste this into your agent CLI:

Set up ego lite for me: https://github.com/citrolabs/ego-lite

Read `skills/ego-browser/references/install.md` and follow the steps to install ego lite.

Why Option 1? It’s a one-and-done setup that skips the extra pop-up prompts during the agent’s first run, getting you straight to work.

Step 2: Enable Chrome Data Migration (Crucial Step)

When you launch ego-lite for the first time after installation, it will ask you a single question: Do you want to migrate your Chrome data?

Select "Yes." Here’s why: your Chrome browser holds all the cookies, login sessions, extensions, and bookmarks you rely on. Once migrated, your AI agent can use ego-browser to directly access your authenticated sessions, completely bypassing manual logins.

Official Privacy Note: All browsing data remains strictly on your local machine. ego-lite only records whether you opted into Chrome migration. Your data privacy is fully guaranteed.

Step 3: Run Your First Browser Task

Once everything is set up, type the following into your agent CLI:

/ego-browser Follow @ego_agent on x.com for me

What’s happening here?

1. /ego-browser is the designated skill trigger prefix. The agent recognizes it and loads its browser automation capabilities.
2. You can write the rest of the prompt in plain natural language. The agent automatically breaks it down: navigate to page → take a snapshot → locate the "Follow" button → click → verify result.
3. The entire operation runs inside the agent’s isolated Space. Your own browser tabs remain completely untouched.
4. The agent uses ego-lite's exposed JS toolset: snapshot, fill, click, wait, navigate, and capture. It bundles these actions into a single execution, running up to 2.5x faster than traditional CLI step-by-step commands.

Real-World Scenario: Auto-Login & Dashboard Data Scraping

Let’s tackle a highly practical use case. Suppose you have an internal management system (e.g., an e-commerce backend) where you need to compile today’s newly listed product statuses.

Goal

Have the AI agent: Log in to the backend → Navigate to the product management page → Extract product names and statuses for today’s new listings → Return a structured list.

Execution Steps

1. Verify Chrome is Logged In
Since you migrated Chrome data earlier, ego-lite has already inherited your session. No need for the agent to handle passwords.

2. Issue the Command to Your Agent

/ego-browser Open https://admin.example.com/products, find all products listed today, and organize their names and listing statuses into a Markdown table.

3. Wait for Execution
The agent opens the page in its isolated Space, automatically leverages your existing session to enter the backend, scrapes the product info, and returns structured data. Meanwhile, you can continue coding or reading docs in your own browser without any interference.

4. Monitor Space Status
The ego-lite UI clearly shows which Spaces are active. You can take over or terminate any task on the fly.

Why This Beats Traditional Workflows

• No Credential Sharing: The agent uses your real cookies. You never need to paste login credentials into a prompt.
• True Parallelism: You can have Claude Code scraping one site in one Space, while Codex crawls five competitor sites in another Space. They never fight over tabs.
• Drastic Token Savings: By writing JS to call ego-lite's built-in tools, the agent completes multi-step operations in one shot. It avoids the costly "run command → read output → run next command" loop, saving up to 2.5x on token consumption for complex tasks.

FAQ & Troubleshooting

1. macOS Only? Yes. As of this writing, Windows and Linux support are on the roadmap. Windows users can temporarily use a macOS VM or wait for the official release.
2. Will Migrating Chrome Data Delete My Bookmarks? No. The migration is a one-way copy into ego-lite. Your original Chrome profile remains completely intact.
3. What If the Agent Gets Stuck? ego-lite lets you take over any Space at any time. Running Spaces are visible in the browser UI, where you can pause or stop them instantly.
4. ego-browser Skill Isn't Working? Verify that you installed via .dmg or correctly registered it via npx skills add. You can also ask your agent to re-read skills/ego-browser/references/install.md to verify its status.
5. Is It Secure? Everything runs locally. ego-lite only logs your opt-in status for Chrome migration and never uploads your browsing data. It’s MIT-licensed, fully open-source, and auditable.

Summary & Next Steps

Here’s a quick recap of what we covered:

1. Download and install ego-lite (the .dmg is recommended).
2. Migrate Chrome data on first launch to give agents instant access to your sessions.
3. Use the /ego-browser prefix in your agent CLI to issue browser tasks in plain English.
4. Let agents execute automation in isolated Spaces while you keep working uninterrupted.

What truly stands out about ego-lite is its core philosophy: instead of giving agents a throwaway browser to work in, it lets them share your actual browser, each in their own isolated space. This fundamentally solves session isolation and multi-tasking bottlenecks.

What to do next: If you’re already using Claude Code or Codex, try the /ego-browser command to automate form-filling or data extraction. Once comfortable, experiment with running multiple Spaces in parallel—having different agents scrape different sites simultaneously will genuinely double your efficiency.

Check out the project here: citrolabs/ego-lite. It’s MIT-licensed and completely free. Join the community on Discord for support and discussions.