Published: 2026-09-24
Summary

ego lite: a Mac browser your agent drives through a skill, in its own "space"

Chapters / key moments (click to jump — plays here on the page)

FuturMinds installs the ego lite browser app plus its ego-browser skill so Claude Code, Codex or ChatGPT can work in pages you are already signed into, each task in a separate space, stopping before any submit. Mac only for now, with clear security caveats.

Source video

"AI Agents + Ego (lite) = Your Browser on Autopilot" by FuturMinds — Watch on YouTube →

What you install

Two parts: the ego lite browser app for macOS, and a skill called ego-browser that tells the agent how to drive it. Installing the skill alone doesn't replace installing the app.
App: download it from the project's GitHub page, open it, and optionally import your Chrome data so signed-in sessions carry over.
Skill: the README gives an npx command to run from the agent you use (the video runs it in Claude Code; the exact command isn't read out, so copy it from the repository). In a new session, typing /ego then shows the ego-browser skill. In ChatGPT, the video pastes the same skill command and asks ChatGPT to set it up.

How it behaves

Each agent task gets a "space" inside the browser, with its own tabs, so your own browsing isn't cluttered. The button at the top of the app lists all spaces. If you want the result tabs left open, say so in the prompt; otherwise the agent cleans them up.
Spaces on the same profile share cookies and cache. Use a separate profile for a different account.
Login handoff: when an agent hits a login wall (the video uses x.com), it stops and asks. You sign in within that space, then tell it to continue.
Prompts shown: find five AI events in Singapore in the next 30 days and check date, venue, price and registration status, keeping the shortlist pages open; research recurring complaints about AI coding agents on x.com with sources, "don't post or change anything"; and fill a long multi-tab grant-style form from a written brief, stopping before submitting.
The video claims the snapshot it sends back is structured rather than raw HTML and that successful actions are distilled into reusable workflows that run up to 5x faster later. Those are the project's claims, shown with its own benchmarks; we haven't tested them.

Platform limits and alternatives

As of September 2026 the repository lists a Windows beta as "coming soon". The video notes Claude Code and Codex already have official ways to work with browsers, including signed-in sessions, so that part isn't exclusive to ego lite.
For agents on a Linux server (OpenClaw, Hermes), the Mac app and your laptop's logins don't come along. The video says OpenClaw documents Browserless and Hermes supports Browser Use Cloud; check those against each vendor's docs before relying on it.

Caveats

Convenience comes with access. A connected agent acts inside your signed-in accounts and page content may reach its AI provider. The presenter suggests keeping banking, primary email and other sensitive accounts out, and notes the product's own data-collection terms and possible leakage into crash reports.
Start small: ask before anything is submitted or changed, and check the result yourself.

Key takeaways

Give browser tasks their own space and an explicit "stop before submitting".
Tell the agent whether to leave result tabs open.
Sign-in is a human step; don't hand over sensitive accounts.