ego lite gives AI agents isolated browser Spaces so automation leaves your tabs alone

A new browser called ego lite has reached Hacker News's front page via a Show HN post, positioning itself as a browser built for AI agents rather than humans alone. Its central pitch: agents run web automation in isolated, parallel workspaces while you keep browsing in the same app, with no cost and no configuration required.

A separate Space for every agent

According to the project page shared on Hacker News, every agent that connects to ego lite gets its own fully isolated workspace, called a Space. You browse in the foreground; agents operate in the background. Multiple agents can run simultaneously — the page's example has Claude Code enriching ten sales leads while Codex scrapes five competitor sites at the same time — and none of them touch your tabs. That arrangement of one browser with separate workspaces is the feature the team leans on hardest, because the usual alternatives are either a second browser install dedicated to automation or an agent commandeering the one you are actively using.

Agents write code instead of issuing commands

The more technical claim concerns how agents drive the browser. Instead of stepping through a CLI loop — one command, one observation, repeat — ego lite has agents write JavaScript functions directly. The team claims complex tasks finish up to 2.5× faster this way, with higher success rates and fewer tool calls, which would also mean fewer model round-trips per task. The page also highlights its page-snapshot mechanism: through customisation the makers describe as happening at the kernel level, it claims to cope with awkward cases such as deeply nested iframes where other snapshot approaches break down.

Any agent CLI can drive it

An accompanying connector called ego-browser links agent CLIs to the browser; Claude Code, Codex and Cursor are named. It exposes the browser as a set of JavaScript tools — snapshot, fill, click, wait, navigate, capture — and installs as a skill with npx skills add citrolabs/ego-lite. During setup, ego lite offers to migrate your existing Chrome data, so agents inherit your logins, cookies, extensions and bookmarks. That removes the familiar pain of agents hitting login walls, though it also hands authenticated sessions to automated software, which is worth pausing on.

Self-reported comparisons

The project publishes a feature table against Browser-Use, Vercel's agent-browser, ChatGPT Atlas and Perplexity Comet, in which ego lite is the only product credited with parallel multitasking, reusable skills, a shared browser with separate workspaces, locally stored data and a zero price. The page also references benchmark results versus Vercel's agent-browser on complex automation tasks, though no figures appear in the text surfaced through the feed. Everything in the table is the vendor grading its own homework, so treat it as a positioning statement rather than a verified verdict.

Availability

For now it is a macOS download, with the page offering builds per architecture, and Windows and Linux listed as coming soon. Usage is free, according to the project.

Why it matters

Browser automation is becoming the default way AI agents act on the web, and current tooling mostly forces a choice: a headless or separate browser that lacks your logins, or an agent loose inside the browser you are using. ego lite's isolation model is a plausible answer to both problems, and its code-over-CLI approach targets a real cost lever, since fewer tool calls means fewer tokens and faster task completion. The open questions are just as clear: the speed claims are self-reported, benchmark details are not published in the launch material, and inherited Chrome sessions concentrate a lot of access in whatever agent you connect. With OpenAI's Atlas and Perplexity's Comet pushing their own agent-first browsers, an agent-agnostic, locally stored alternative tests whether this category stays open or gets absorbed by the major platforms.