Stop Spawning Headless Browsers: How We Built ego-lite for True Human-Agent Coexistence

On August 15, 2026, we launched ego-lite onto the GitHub Trending charts. It’s a browser terminal application built from the ground up around a simple premise: humans and AI agents should be able to run side by side on the exact same machine, sharing the exact same browser environment without stepping on each other's toes.

<div class="aice-quick-answer" style="background:#f0f7ff;border-left:4px solid #2563eb;padding:14px

Right now, ego-lite is native to macOS (Windows and Linux builds are on our roadmap). You can install it in seconds via three paths: grab the native macOS build, run `npx skills add citrolabs/ego-lite`, or let your preferred agent run the installation for you.

When you boot it up for the first time, ego-lite asks exactly one question: *Do you want to migrate your Chrome data?* Hit yes, and your agents immediately inherit your active logins, cookies, extensions, and bookmarks. No broken authentication loops. No manual 2FA headaches.

Multi-Space Architecture: Zero Interference, Zero Process Sprawl

Most existing agent frameworks spin up completely detached headless instances or rely on clunky external drivers. We took a different route.

ego-lite partitions the browser into isolated Spaces. Each agent gets its own sandboxed Space to execute tasks independently. You can have Codex scraping five competitor landing pages in Space A while Claude Code enriches ten outbound sales leads in Space B. Meanwhile, you’re in the foreground drafting an email. Your active tab, cursor focus, and scroll position remain completely untouched.

Everything is driven through the `ego-browser` skill. When you run a command in your agent's CLI (e.g., `/ego-browser followed @ego_agent on x.com for me`), the agent doesn't execute slow, piecemeal shell commands. Instead, it leverages direct JavaScript APIs exposed natively to the runtime—including `snapshot`, `fill`, `click`, `wait`, `navigate`, and `capture`.

Because the agent can emit complete end-to-end execution scripts in one go rather than waiting on step-by-step roundtrips, complex tasks run up to 2.5x faster. Tool-call volume drops dramatically, and task completion rates jump across the board.

Kernel-Level Snapshots and Local-First Privacy

Visual and structural parsing has always been the bottleneck for browser agents. ego-lite introduces kernel-level DOM and visual snapshotting that reliably parses deeply nested `iframe` architectures—a failure mode that breaks almost every generic wrapper on the market.

Privacy is non-negotiable for us. All snapshot data lives strictly on your local machine. The only telemetry ego-lite records is whether you opted into Chrome data migration during setup. None of your browsing history, page contents, or agent payloads ever touch a cloud server.

Outperforming Walled Gardens and Raw Libraries

The landscape today is split into two extremes:

1. Pure codebases (Browser-Use, Vercel’s agent-browser): These require you to manage detached browser instances from scratch, with zero inherited session state.

2. Closed ecosystems (ChatGPT Atlas, Perplexity Comet): These lock you into proprietary models and walled-garden agents.

ego-lite bridges the gap. We support Claude Code, Codex, Cursor, and any custom agent setup through a unified `ego-browser` interface. It serves as your daily driver browser and your agents' execution engine—free, zero-config, and local-first.

In our internal benchmarks across four complex automation benchmarks, ego-lite outperformed the competition across every run. On one multi-step test, ego-lite finished in just 40% of the time required by `agent-browser` while burning significantly fewer tokens. The more complex the task, the wider that performance gap becomes.

The Real Goal: Shared Sovereignty

We didn't build ego-lite to replace the human in the loop. We built it to turn the browser into a genuinely collaborative canvas.

Automation shouldn’t mean handing over complete control or getting locked out of your screen while an agent takes over your mouse. You should be able to read an article while three agents handle back-office data entry in the background—sharing session state without polluting your workflow. It isn't about who owns the browser; it's about letting humans and AI retain sovereignty over their own work.

ego-lite is fully open under the MIT License, with the standalone browser available as a free download. Read the documentation at [lite.ego.app/document/](https://lite.ego.app/document/) and join the conversation on our Discord, GitHub Discussions, and X.

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top