Ego Lite - A browser designed specifically for AI agents, enabling humans and AI to work together.

What is Ego Lite?

Ego Lite is a Chromium browser developed by Citro Labs specifically designed for AI Agents. It addresses pain points of traditional browser automation tools (such as Playwright and Puppeteer) when driving the browser, such as lost login state, high resource consumption, and interference with normal user browsing. It enables collaborative work between human users and the AI Agent. By exposing a Node.js runtime via the CLI, the Agent can silently drive the browser in an independent "task space," performing operations such as page loading, clicking, API probing, and screenshot verification, while inheriting the user's login state without interfering with the normal browsing window.

Main features of Ego Lite

• Semantic snapshot extraction:pass snapshotText() Get the semantic structure of page text, buttons, links, and clickable elements.
• Element interaction operations:support click,fillInput With these instructions, the Agent can accurately click buttons and fill out forms.
• Low-level script execution:pass js() Execute JavaScript within the page context to directly read DOM state and front-end data.
• Browser protocol call:pass cdp() It uses the Chrome DevTools Protocol to intercept network requests and perform deep debugging.
• Screenshot visual verification:captureScreenshot Supports visual workflows for validation in non-standard DOM scenarios such as rich text, maps, and whiteboards.
• Network request crawling:browserFetch It can directly initiate or intercept internal browser requests to obtain the actual API return data from the front end.
• Task space isolationEach task has its own task space(Tab groups) reuse login states and do not interfere with each other.

Ego Lite's technical principles

• Chromium-based Agent Driver LayerEgo Lite builds a CLI layer for agents on top of the Chromium kernel. During installation, the system exposes a Node.js runtime, and the agent communicates through... ego-browser nodejs Execute the pre-defined script, which encapsulates... snapshotText,click,js,cdp,browserFetch Helper functions, etc.
• Three-tier workflow architecture:
  • Semantic WorkflowPrioritize snapshotText() Extract semantic information from interactive elements on a webpage, suitable for standard webpage structures.
  • Visual WorkflowWhen the DOM is inconsistent with the actual interface (such as Canvas, SVG, map), revert to screenshot + coordinates + keyboard action verification.
  • underlying workflow:pass js() Inject code into the page context, or through... cdp() It directly calls the Chrome DevTools Protocol to read the front-end state, detect hidden API interfaces, and parse the front-end bundle.
• Task Space Isolation MechanismEach task space is an independent browsing context with its own tabs, and by default inherits the current user's cookies and login status. The agent can operate within sites that require login and will not preempt control of the browser window currently being used by the user.

Follow us on WeChat and reply with "open source",join inAI open source project discussion group

How to use Ego Lite

• Download and installVisit the official website https://lite.ego.app/ to download the Ego Lite app. After installation, migrate your Chrome data (bookmarks, extensions, cookies, login state) with one click.
• Install Agent Skill PackRun in the Agent tool npx skills add citrolabs/ego-liteAutomatic installation ego-browser CLI skills.
• Restart AgentAfter installation, restart the Agent to ensure the skill pack is loaded correctly.
• Send browser taskEnter in the Agent dialog @ego lite Or enter directly ego-browser The initial instruction describes the webpage to be browsed and the target operation.
• Agent silent executionThe agent automatically creates a task space, opens the target page, reads a semantic snapshot or executes JavaScript, and closes the task space and returns the results after obtaining the data.
• Manual intervention (optional)When a task requires manual confirmation, the Agent can pause and relinquish control, allowing execution to resume after user confirmation.

Ego Lite's core advantages

• Do not interfere with the user windowThe agent runs in a separate task space and will not take over the user's current browser tab or trigger foreground pop-ups.
• Inherit login statusNo additional account and password configuration is required; the agent can directly access sites that require login (such as email or internal systems).
• Three-tier workflow coverageSemantics + Visuals + Underlying Protocols: Adapting to all scenarios from standard web pages to complex rich media pages.
• Front-end status is readable:pass js() It can directly read the internal state of frameworks such as React/Vue to explore real API interfaces, breaking through the limitations of traditional web crawlers.
• On-site verification capabilityThe Agent can perform real operations and read real responses, combining model judgments with browser-based on-site calibration to reduce illusions.

Ego Lite project address

• Project official websitehttps://lite.ego.app/
• GitHub repositoryhttps://github.com/citrolabs/ego-lite

Comparison of Ego Lite with similar competing products

Application scenarios of Ego Lite

• Information aggregation and monitoringThe agent periodically accesses multiple subscribed sites to extract the latest article titles, links, and summaries, generating a daily report.
• Competitor dynamics trackingAutomatically access competitors' official websites, app stores, and social media to capture information such as version updates, pricing changes, and user reviews.
• Internal system automationWhile inheriting the login status, the Agent operates the enterprise's internal backend to query data and fill in reports.
• Web application acceptance testingIn the Vibe Coding scenario, the agent automatically opens the web application under development, executes the function click stream, and takes screenshots to verify the UI status.
• In-depth data researchWhen accessing academic databases or paid content sites that require login, the agent reads the page structure, probes the API, and downloads or organizes materials.