BREAKING
Alibaba open-sources PageAgent
0
k
GitHub stars
0
k
forks
0
%
TypeScript
How it works: DOM dehydration
1
Scan DOM elements
↓
2
Build FlatDomTree
↓
3
Feed text to LLM
↓
4
Fire click/input/scroll
Inside-out vs external automation
PageAgent
in-page
●
Reads structured DOM
●
Client-side, no backend
●
One script tag or npm
Selenium / Playwright
external
●
Drives browser outside
●
Screenshots or vision
●
Needs separate runtime
Any OpenAI-compatible endpoint
Free under MIT, LLM costs vary
AI NEWS BLITZ
Alibaba has released PageAgent, an in-page JavaScript AI agent, under the MIT license.