Alibaba has released PageAgent, an open-source JavaScript GUI agent that can operate any web page through natural language. Adding a single <script> tag lets developers embed an AI copilot into an existing site.
Open Source · Alibaba
PageAgent: one <script> tag drops an AI copilot into any web page
An MIT-licensed, client-side JavaScript GUI agent that operates any website through natural language — reading the DOM directly as text, with no screenshots or vision models required.
1
Line to integrate (v1.11.0)
One instruction vs. the manual way
A legacy ERP/CRM task that once took ~20 clicks collapses into a single natural-language command.
~20 clicks
Manual navigation
→
1 command
Natural language
How it reads the page
DOM as text
No screenshots
→
→
Human review
Confirm before act
→
Page action
In-page, no server
Reading the DOM directly means lower token cost and faster runs than vision-based browser agents. Works with GPT, Claude, Qwen, Gemini, DeepSeek, Grok, or local models via Ollama / LM Studio.
Where it shines
One-line AI copilot for any SaaS or legacy ERP/CRM
Token savings and speed from DOM-first reading
Multi-tab control via optional Chrome extension
Accessibility, voice commands, auto form-filling
Limits to plan for
Client-side API keys need a backend proxy in production
Prompt-injection risk — verify sensitive ops server-side
No Shadow DOM, canvas, or cross-origin iframes
A UX layer, not a Playwright-style testing replacement
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…