July 24, 2026AgentsSkillsOpen Source

ego-lite Gives Your Agent Its Own Browser So It Stops Grabbing Yours

Anyone who has watched an agent drive a browser knows the pain. It takes over your screen, moves your mouse, and you sit there frozen until it is done. ego-lite from CitroLabs fixes this in the most obvious way once you see it: give the agent its own space in the same browser and let you keep working in yours.

It is a macOS browser, MIT licensed, and the interesting design choice is how the agent acts. Instead of clicking one thing at a time through a command loop, the agent picks up a skill called ego-browser and calls it as JavaScript functions, composing a whole multi-step task at once. It opens the page in its own Space, reads a snapshot, acts, reports back. The browser inherits your Chrome logins and cookies on first launch, so the agent is not locked out of everything, and it hands vision models clean page snapshots instead of raw pixels. CitroLabs claims 2.5 times faster on complex tasks with far fewer tokens, and it works with any CLI, Claude Code, Codex, Cursor.

It is trending on GitHub for a reason. The browser has been the ugliest surface for agents from day one. Everything else got clean APIs and MCP, and the web stayed a pile of pixels and hijacked cursors. Treating the browser as a place where multiple agents each get an isolated room, and exposing it as a skill an agent composes against instead of a puppet it jerks around, is the right shape. https://github.com/citrolabs/ego-lite
← Previous
Big Tech Parked $1.65 Trillion of AI Debt Where You Can't See It
Next β†’
Alibaba's Open Code Reviewer Mixes Rules With Agents on Purpose
← Back to all articles

Comments

Loading...
>_