ego lite
A macOS browser that lets AI agents automate logged-in web tasks in isolated workspaces.
Per-dimension scores and reasoning
Evidence shows: README claims local data storage and only records whether Chrome migration was opted into, but no code-level verification; installation asks about Chrome data migration but does not explain data usage and deletion; no data flow diagram or detailed explanation; no specific measures for sensitive data handling; dependency security only via npm ci and tests in CI, no dependency audit or vulnerability scan; external effects include adding skill to agents directory and possibly migrating Chrome data, but permission scope not explicit; no rollback mechanism; source attribution via MIT license and CitroLabs copyright, but publisher unverified. Deductions: lack of code-level evidence, opaque data flow, no rollback.
Evidence shows: README and CI workflow are consistent, tests exist but only cover publish workflow; dependency availability not explicit, but CI uses npm ci and package-lock.json; failure messages not provided in docs, but CI has error handling. Deductions: dependency availability unverified, failure messages not provided.
Evidence shows: target audience clearly AI agent users, scenarios include browser automation and parallel tasks; capability boundaries not explicit, but macOS support mentioned, Windows/Linux on roadmap; trigger precision via /ego-browser command and skill description, but not detailed; environment fit only macOS, and relies on Chrome data migration. Deductions: unclear capability boundaries, insufficient trigger precision, limited environment fit.
Evidence shows: README structure clear, includes quick start, features, comparison, benchmarks, docs links; install notes detailed with multiple methods; naming stable, but no version history; examples and FAQ have example commands but no FAQ; known limitations only mention macOS support, not others; license MIT, but no changelog; maintenance responsibility not explicit, but CI and release workflows exist. Deductions: no changelog, incomplete known limitations, unclear maintenance responsibility.
Evidence shows: output usability via skill and docs, but no actual output examples; marginal value via comparison and benchmarks, but no reproducible data; cost-benefit claims faster and fewer tokens, but no detailed cost analysis. Deductions: lack of verifiable benchmark data.
Evidence shows: README claims not traceable to tests or data; cross-source verification only via external links (docs, Discord), but no independent verification; facts and inferences not clearly separated, e.g., benchmark data source not stated. Deductions: claims lack traceability, no independent verification, facts and inferences conflated.
- Publisher identity unverified; treat claims with caution.
- Data flow opaque, especially regarding Chrome data migration usage and deletion.
- No dependency audit or vulnerability scan evidence.
- Benchmark data lacks reproducible methodology.
What does this agent do, and when should you use it?
ego lite is a macOS browser for people and AI agents to work in parallel through a skill called ego-browser. It gives each agent or task an isolated Space, so an agent can work in the background without taking over the user's tabs. The skill exposes browser capabilities as in-page JavaScript tools, allowing an agent to compose a multi-step web task into one script execution. Its documented tools are snapshot, fill, click, wait, navigate, and capture, and the agent reports back after completing the task. The repository contents are MIT licensed, while the ego lite browser itself is distributed separately as a free macOS download.
A user installs the ego lite app or adds the ego-browser skill with npx skills add citrolabs/ego-lite. When an agent first runs a browser task, the skill guides installation of the app; on first launch, the user can choose to migrate Chrome data, including existing logins, cookies, extensions, and bookmarks. In an agent CLI, the user enters /ego-browser followed by a natural-language request. The agent opens the page in its own Space, reads a Snapshot, and uses in-page JavaScript tools such as snapshot, fill, click, wait, navigate, and capture. The agent can run a JavaScript snippet containing multiple page actions in one pass, then report the result.
- A researcher using Codex needs to gather information from logged-in sites without losing control of their current browser tabs.
- A sales operator using Claude Code wants to enrich multiple leads in separate Spaces at the same time.
- An operator with existing Chrome sessions wants an agent to follow an account, fill a form, or navigate a signed-in website.
- A user working with pages that contain nested iframes wants the agent to inspect a Snapshot before interacting with the page.
- A developer using a custom agent CLI wants to connect browser automation through the ego-browser skill.
What are this agent's strengths and limitations?
- Each agent or task gets an isolated Space, enabling parallel browser work without disrupting the user's tabs.
- Its documented in-page JavaScript tools—snapshot, fill, click, wait, navigate, and capture—can combine multi-step actions into one execution.
- Optional first-launch Chrome migration carries over existing logins, cookies, extensions, and bookmarks.
- The repository explicitly positions ego-browser as a connection layer for Codex, Claude Code, Cursor, and custom agent CLIs.
- macOS is the only explicitly supported operating system; Windows and Linux are listed on the roadmap.
- The core workflow requires the separately installed ego lite browser app, not merely the repository skill.
- Chrome migration brings existing login state, cookies, extensions, and bookmarks into the browser, which requires consideration before adoption.
- The browser is distributed as a separate download; the MIT license explicitly covers the repository contents.
How do you install or deploy this agent?
Run npx skills add citrolabs/ego-lite to install only the ego-browser skill. Alternatively, install the macOS ego lite app from the Apple Silicon or Intel DMG downloads provided by the repository. The first browser task run by an agent guides app installation, and first launch offers Chrome-data migration. The documentation does not specify a separate API key or account credential requirement.
How do you use this agent?
In a supported agent CLI, enter /ego-browser followed by a task description, for example: ego-browser follow @ego_agent on x.com for me. The agent opens the page in its own Space, reads a Snapshot, performs the page actions, and reports back. The README explicitly names Claude Code, Codex, Cursor, and custom agent CLIs as integrations through ego-browser.
How does this agent compare with similar options?
The repository compares ego lite with Browser-Use, Vercel's agent-browser, ChatGPT Atlas, and Perplexity Comet. Its stated position is that it combines a browser with shareable login state and isolated Spaces, unlike browser automation frameworks, while remaining controllable by external agents unlike built-in-agent browsers. The README also reports benchmark results against Vercel agent-browser on four complex tasks; those performance claims come from the repository's own comparison.