ego-lite: A Browser Where AI Agents Share Your Chrome Login Without Stealing Your Tabs
Login state has always been the first and hardest blocker in web automation, whether you're testing internal tools or evaluating AI agents. ego-lite removes that wall by handing agents a real, authenticated session inside your own browser, so they can work alongside you instead of in a separate, sandboxed process that trips risk controls.
ego-lite, from Citro Labs, is a desktop browser built for human-AI agent collaboration. It runs on macOS (Apple Silicon and Intel) and Windows, with Linux on the roadmap. The browser imports your entire Chrome profile, including passwords, cookies, login sessions, extensions, and bookmarks, so agents get a real, trusted session instead of a blank headless context.
Agents execute tasks in separate 'Spaces' — isolated workspaces that don't interfere with your own browsing tabs. You can watch or take over at any time to handle CAPTCHAs or sliders. The built-in ego-browser skill lets you issue natural-language commands like '/ego-browser open the issues page and extract the last 10 titles', and the agent opens its own Space, does the work, and returns results.
The project has 18,373 stars since April, with v2.1.0 released October 16. It's free, MIT-licensed, and stores all data locally. The developers claim it's 47–63% faster than playwright-cli on 31 real-world tasks, though that benchmark isn't independently verified.
The core win is not a new automation syntax but the decision to reuse a human's real, trusted session instead of spinning up clean headless instances that risk controls flag.
Two mental models from the article generalize: login state is a test asset, and letting an agent execute a whole program in one shot costs far less than step-by-step prompts.
If the speed claim holds, it likely stems from reducing LLM tool calls via batch page operations and native snapshots, not from faster browser internals.
The tool positions itself as complementary to Playwright, targeting exploratory and human-in-the-loop testing, where maintaining selectors and session state is the real cost.