A new player called ego lite just dropped on Hacker News with a bold claim: browser automation should be code-first, not CLI-first. The team at CitroLabs built their browser from the ground up to let AI agents write JavaScript directly against web pages rather than shell out sequences of command-line calls. The pitch is simple—less overhead, fewer round trips, and tasks that complete up to 2.5× faster than traditional approaches.

Why JavaScript Beats CLI for Browser Agents

The core innovation here is architectural. Existing tools like browser-use and Vercel's agent-browser treat the browser as something to be driven externally through a separate process. They log actions, parse output, then issue more commands in a tedious back-and-forth loop. Ego lite flips this by exposing its capabilities as JavaScript functions—snapshot, fill, click, wait, navigate, capture—that agents call directly from within the page context. An agent composes a multi-step workflow into a single programmatic execution instead of getting stuck in "call two commands, look at result, call two more" paralysis.

Spaces: Parallel Workspaces That Don't Collide

One of the biggest pain points with existing automation frameworks is tab management—you and your agent end up fighting over the same browser window. Ego lite solves this with isolated Spaces, fully sandboxed workspaces where each agent runs independently while you keep browsing normally on the main thread. The demo videos show Claude Code enriching 10 leads in 10 parallel Spaces simultaneously while Codex scrapes competitor sites in five more—all without touching your cursor or stealing your tabs.

Kernel-Level Snapshots and Experience Accumulation

The team claims kernel-level customization gives ego lite the strongest page snapshot capability on the market, reliably handling tough cases like deeply nested iframes where other approaches consistently break down. They're also teasing an experience accumulation feature that distills every successful agent action into reusable skills and workflows—similar tasks reportedly run up to 5× faster once the system learns your patterns.

The Comparison Table Tells a Story

Ego lite checked their feature matrix against Browser-Use, Vercel's agent-browser, ChatGPT Atlas, and Perplexity Comet. They hit checkmarks in every cell where competitors came up empty—parallel multitasking, reusable skills, Chrome data inheritance without login friction, local-only storage, and daily-use browser capability. On benchmarks against Vercel's agent-browser across four complex automation tasks, ego lite finished each one faster with substantially fewer tokens, and the advantage widened on harder workflows.

Getting Started

The installation is straightforward for macOS users today—download the app or run a curl install script that drops in the browser, the ego-browser helper, and integrates skills into your agent CLI of choice. Windows and Linux support are on the roadmap. On first launch, you can migrate existing Chrome data including logins, cookies, extensions, and bookmarks so your agent inherits real credentials immediately.

Key Takeaways

  • Ego lite exposes browser automation as JavaScript APIs rather than CLI commands, eliminating round-trip overhead
  • Isolated Spaces let multiple agents run in parallel without interfering with your browsing session
  • Benchmarks show 2.5× speed improvement over Vercel's agent-browser on complex tasks with fewer tokens
  • Chrome data migration gives agents instant access to real logins and sessions
  • Currently macOS only, with Windows and Linux support planned

The Bottom Line

This is the right architectural bet—JavaScript-first browser automation aligns perfectly with how LLM agents actually reason about tasks. When you stop forcing code generators to think in shell commands and let them write actual code against a DOM API, you're fighting less friction at every layer. Ego lite's Spaces concept alone makes this worth watching for anyone running automated workflows through Claude Code or Codex.