// Agent profile
Page Agent is an open-source JavaScript GUI agent from Alibaba that runs inside web pages to control interfaces with natural language. Unlike traditional browser automation tools that drive the browser from an external process, Page Agent operates entirely client-side — it reads the live DOM as structured text via a technique called DOM dehydration, sends it to any text-capable LLM, and executes planned actions as native browser events. No screenshots, no multimodal models, no browser extensions required for single-page tasks.
Page Agent takes the opposite approach from traditional browser automation. Instead of driving a browser from an external process with Selenium or Playwright, Page Agent lives inside the webpage as client-side JavaScript. When it receives a natural language command, it scans the DOM and converts every interactive element into a compact text representation — a process called DOM dehydration. This text map goes to an LLM, which plans the action, and Page Agent executes it as a native browser event. No screenshots, no multimodal vision models, no special permissions. With 24,000+ GitHub stars and 1,100+ daily stars at its peak, Page Agent became one of the fastest-growing open-source projects of mid-2026.
The DOM dehydration technique is Page Agent’s core innovation. By converting page structure to text rather than sending screenshots, it works with any text-capable LLM — not just expensive multimodal models. Text tokens cost 10-20x less than image tokens, making the approach significantly cheaper to operate. Because the agent runs in-page, it inherits the user’s cookies, session, and authentication automatically. The optional Chrome Extension adds multi-tab support, and the beta MCP server lets external AI orchestrators like Claude Code or Cursor use Page Agent as a tool.
Product teams embed Page Agent as an in-app AI copilot that guides users through complex workflows with natural language. Developers use it to automate form-heavy internal tools — turning 20-click admin workflows into a single sentence. The MCP server integration makes it a building block for larger agent systems that need browser interaction capabilities. QA teams use it for exploratory testing where semantic understanding (“click the submit button”) is more resilient than brittle CSS selectors.
LLM-driven planning means Page Agent achieves 70-85% success rates rather than deterministic execution — it’s not suited for mission-critical flows like payment processing. Client-side deployment exposes LLM API keys unless a proxy is configured. Single-page scope by default; multi-tab workflows require the Chrome Extension.
Web developers who want to embed AI assistance directly into their product UI. Teams automating browser-based workflows who want something more resilient than selector-based scripts. Builders of multi-agent systems who need a browser interaction tool that integrates via MCP.
Builder.io's open-source framework for building agent-native applications — shared actions, SQL-backed state, identity, tools, skills, jobs, observability, and UI surfaces that all work together.
CLI orchestration tool that installs and configures platform connectors for 16+ social and web sources — Twitter/X, Reddit, YouTube, GitHub, Bilibili, and more. Zero API fees.
Persistent memory layer for AI coding agents — benchmark-backed (95.2% on LongMemEval-S), 92% fewer tokens per session vs full-context pasting, zero manual memory.add() calls.