Who ego-lite is for#
AI developers automating authenticated web tasks
Developers using Claude Code, Codex, or Cursor who need agents to interact with services requiring a login: CRM back offices, SaaS admin panels, social media accounts, or internal tools behind SSO. ego-lite's shared state means the agent picks up the active session rather than starting from scratch.
Skip if:
Skip if your automation targets only public pages with no authentication requirements. A lightweight headless browser or simple HTTP scraper is simpler and faster for unauthenticated public content.
Teams running parallel lead enrichment or research tasks
Teams that need an AI agent to process a list of leads, competitors, or data sources in parallel across multiple browser sessions. Each target runs in its own Space, so 10 simultaneous tasks do not collide and finish together rather than sequentially.
Skip if:
Skip if your task list is small or if the sources are accessible via public APIs. The parallel Spaces model adds the most value for larger batches of authenticated or dynamic-content targets.
Developers replacing Browserbase or ScraperAPI with a local alternative
Builders who currently pay for a managed browser automation service and want to cut that cost. ego-lite runs on the developer's own machine, uses their own identity, and charges nothing per session. The ego-browser skill replaces the managed API layer for tasks that do not need a cloud runtime.
Skip if:
Skip if your automation runs on a remote server or CI pipeline without a macOS host. ego-lite requires the desktop application on a local Mac and does not currently offer a server-side mode.
The problem it solves#
AI agents that automate browser tasks typically run inside a blank, headless Chromium instance. That means every task begins from a logged-out state: the agent has to navigate login pages, solve CAPTCHAs, handle two-factor authentication, and work around SSO redirects before it can do anything useful. On sites that use Stripe, Salesforce, Intercom, or other embedded widgets, headless approaches often fail entirely because the tools that build page snapshots from JavaScript shims cannot reach cross-origin iframes or shadow DOM elements.
The parallel coordination problem is just as painful: when an agent controls a browser through a framework like Playwright, it takes over the window. You cannot keep working in your own tabs while the agent runs a task. Most automation frameworks were built for scripted pipelines, not for sharing a browser with a human in real time. That leads to a choice between letting the agent interrupt your work session or maintaining a separate second browser just for automation, which defeats the purpose of having an integrated AI agent.
How it solves it#
Shared Chrome state for AI agents
ego-lite imports your bookmarks, cookies, login sessions, extensions, and browser profiles from Chrome in one click. AI agents run inside their own Space and inherit the same logged-in state, so they can reach Gmail, Notion, LinkedIn, Salesforce, and any other service you are already authenticated with. No re-authentication per task.
Parallel Spaces for multitasking
Each agent or task gets its own Space: an isolated workspace inside the same browser window. Claude Code can enrich leads in 10 parallel Spaces while Codex scrapes competitor sites in 5 more, without colliding with each other or with your tabs. Your mouse and focus stay where you left them.
ego-browser skill for any agent
ego-browser is the connection layer between ego-lite and any agent that runs shell commands: Claude Code, Codex, Cursor, Continue, Gemini CLI, Hermes Agent, and custom in-house agents. It exposes browser actions as in-page JavaScript tools. Agents write a snippet composing several steps, and ego-browser runs it in one pass instead of one tool call at a time.
Kernel-level Chromium snapshot engine
Unlike JavaScript shims layered on stock Chrome, ego-lite's semantic snapshot is built into a custom Chromium engine at the kernel level. Snapshots reach into cross-origin iframes, shadow DOM, and third-party SDK widgets like Stripe, Salesforce, and React portals. Ref actions run more reliably and return fewer tokens than raw HTML.
Faster task completion on fewer tokens
ego-lite runs multiple in-page actions as a few lines of JavaScript instead of sequential tool calls. Complex browser tasks finish up to 2.5x faster than agent-browser with substantially fewer tokens, per benchmarks the team published against Vercel's agent-browser on four complex tasks.
Strengths and trade-offs#
Strengths
- Zero cost with local data storageego-lite is free to download with no per-session or per-request pricing. All browsing data, cookies, and login sessions stay on your machine. Unlike Browserbase or ScraperAPI, there is no API key to manage, no per-task billing, and no third party that can see the sessions your agent uses.
- Works with any agent through a single skill installInstall ego-browser once with npx and every agent on that machine can drive the browser. There is no SDK to integrate, no provider lock-in, and no requirement to run a specific LLM. Claude Code, Codex, Cursor, and custom agents all work through the same skill interface.
- No login friction or CAPTCHA interruptionsBecause ego-lite shares your real Chrome identity, the agent inherits your existing sessions. It does not need to log in to each service from scratch, which eliminates CAPTCHA challenges, 2FA prompts, and SSO redirect failures that headless browser automation consistently hits.
- MIT license on the open source repositoryThe ego-lite GitHub repository is MIT licensed, so you can inspect the code, fork it, and build on it commercially with no restrictions. The browser itself is a separate free download. MIT licensing means no AGPL copyleft concern if you embed or extend the skill layer.
Trade-offs
- -macOS only; Windows and Linux not yet availableego-lite runs on macOS (Apple Silicon and Intel). Windows and Linux are listed on the roadmap but not yet available. Teams that work primarily on Linux servers or Windows machines cannot use ego-lite until cross-platform support ships.
- -Reusable skills feature is not yet availableThe README lists experience accumulation (reusable skills that distill successful actions for faster future runs) as a coming-soon feature. Tasks cannot currently benefit from learned patterns, so the agent rebuilds context from scratch on each run.
- -106 open issues as of August 2026The repository has 106 open issues. Shipped in April 2026, the project is actively iterating and some edge-case behaviours may not yet be resolved. Check current issue status before integrating ego-lite into critical workflows.
ego-lite vs alternatives#
ego-lite vs Browserbase
Browserbase is a managed cloud service that provides AI agents with isolated browser sessions through a proprietary API. ego-lite is the free, local alternative: the agent uses your actual Chrome identity rather than a clean cloud instance.
The key difference is session state. Browserbase starts each session from scratch, so agents must handle authentication, CAPTCHA, and SSO in every run. ego-lite inherits your existing logins from the first use, so those authentication steps never occur. For tasks that require access to authenticated services you use daily, that single difference eliminates a significant source of task failures.
Browserbase is the better fit when you need browser automation to run on a remote server or in a CI pipeline without a macOS host. ego-lite runs locally on your Mac and does not offer a cloud-hosted mode.
| Feature | ego-lite | Browserbase |
|---|---|---|
| License | MIT (repo) | Proprietary |
| Cost | Free | Paid (per session) |
| Session state | Inherits your Chrome identity | Fresh session per run |
| Self-hosting | Local Mac app | Managed cloud |
| Login friction | None | Must handle each run |
| Platform | macOS | Cloud (any host) |
ego-lite vs Bright Data
Bright Data is a commercial web data platform providing proxies, browser APIs, and scraping tools at scale. Its core value is access to residential and datacenter proxies for scraping public data without IP blocks. ego-lite does not provide proxy services and is not a data collection platform for large-scale public web scraping.
For tasks requiring authenticated access to services you already use, ego-lite is the more direct fit. Bright Data excels when the goal is scraping public pages at scale from many IP addresses, which is a different problem from sharing your personal browser state with an agent.
| Feature | ego-lite | Bright Data |
|---|---|---|
| License | MIT (repo) | Proprietary |
| Cost | Free | Paid (per GB / per request) |
| Primary use | Agent browser with auth state | Proxy network and scraping API |
| Scale | Single machine, parallel Spaces | Cloud-scale distributed requests |
ego-lite vs ScraperAPI
ScraperAPI is a paid HTTP proxy API for scraping public websites. It handles JavaScript rendering and IP rotation for fetching public page content. ego-lite is an agent browser for interacting with authenticated services: clicking, filling, navigating, and reading behind a login.
If your scraping target is a public page with no authentication, ScraperAPI is a simpler tool: send an HTTP request, get back the rendered HTML. ego-lite adds value when the agent needs to be logged in, perform actions across multiple pages, or run parallel sessions in isolated workspaces.
| Feature | ego-lite | ScraperAPI |
|---|---|---|
| License | MIT (repo) | Proprietary |
| Cost | Free | Paid (per request) |
| Authentication | Full (inherits your sessions) | None |
| Interaction | Click, fill, navigate, read | HTTP fetch only |
| Platform | macOS desktop | Cloud API |
Install and self-host#
Install ego-lite by cloning the repository; the ego-browser skill source and setup references are included. Download the macOS browser app from the releases page.
```bash
git clone https://github.com/citrolabs/ego-lite
```What it's built on#
- Languages
- JavaScriptTypeScript
- Tooling
- esbuildRollup
FAQ#
Is ego-lite free to use?
Yes. ego-lite is a free macOS download with no subscription, no per-session pricing, and no API key required. The GitHub repository is MIT licensed. The only cost is the machine it runs on, which is typically the developer's existing computer.
Which AI agents work with ego-lite?
Any agent that can run a shell command works with ego-lite through the ego-browser skill. Confirmed compatible agents include Claude Code, Codex, Cursor, Continue, Gemini CLI, Hermes Agent, and OpenClaw. Custom in-house agents that execute shell tools also work without any additional SDK integration.
Does ego-lite work on Windows or Linux?
Not yet. ego-lite currently runs on macOS (both Apple Silicon and Intel). Windows and Linux support are on the roadmap. The ego-browser skill can be installed via npx on any platform, but the browser application itself requires macOS.
How does ego-lite handle my login sessions and private data?
All browsing data stays on your computer. ego-lite does not upload your cookies, passwords, browsing history, or agent session data to any external server. The only thing recorded during setup is whether you chose to migrate Chrome data.
How is ego-lite different from Playwright or Puppeteer?
Playwright and Puppeteer are browser automation libraries that drive pages through CSS selectors against a separate browser instance. ego-lite is a real daily-use browser that your agent shares with you. The ego-browser skill exposes in-page JavaScript tools (snapshot, fill, click, navigate, capture) rather than selector-based commands, and runs multiple steps in one pass. There is no separate browser to install or configure.
Similar open-source tools#
PyScrappy
AI-native web scraping toolkit + MCP server
Agent-Reach
Give agents local web and social-source access
Firecrawl
Turn any website into clean markdown or structured JSON for LLMs
dograh
Open source voice agent builder with telephony and BYOK AI support
whishper
Local speech-to-text transcription and subtitling with a web UI
r3
Minimalist Forth language for games and systems programming