qfeuilla

qfeuilla / Webfurl

Public

A browser use AI agent that uses a compressed unfoldable representation of HTML for high token efficiency

36
2
100% credibility
Found Mar 02, 2026 at 25 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Rust
AI Summary

WebFurl is an open-source Rust tool that powers an interactive AI web browsing agent using recursively compressed semantic trees of web pages for efficient LLM context management and real browser control.

How It Works

1
🔍 Discover WebFurl

You hear about WebFurl, a clever companion that helps an AI assistant browse websites smartly without getting overwhelmed by page details.

2
⚙️ Prepare your setup

Download the files and create a simple settings note where you paste your AI service connection details to link the smart helper.

3
🚀 Launch everything

Click the start button – it prepares the tools, opens a web browser window you can see, and brings up a chat prompt ready to go.

4
🌐 Visit a website

Type a web address like your favorite vacation site, and watch the assistant squeeze the whole page into a neat outline of key sections.

5
💬 Chat naturally

Ask questions like 'find cheap stays with good reviews' – the assistant zooms into relevant parts, clicks buttons, and fills forms for you.

6
🔍 Explore deeper

Use simple commands to expand sections, search smartly, take screenshots, or describe pictures as needed, all while keeping things efficient.

✅ Get perfect results

You quickly discover exactly what you need, like the best deals or info, with the AI staying sharp and the browser handling real actions smoothly.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 25 to 36 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is Webfurl?

Webfurl is a Rust-based browser agent that launches a real Chrome instance via DevTools Protocol and compresses full web pages—often 200k+ tokens of HTML—into a recursive semantic tree of 20-50 tokens at the top level. It solves LLM context explosion in browser agents by caching compressed subtrees across users in MongoDB, unfolding only query-relevant branches via embeddings, and linking summaries to live DOM selectors for clicks, form fills, and navigation. Developers get an interactive CLI: hit `/url `, chat naturally like "find cheap listings," and use commands such as `/search`, `/unfold`, or `/screenshot`.

Why is it gaining traction?

Unlike flat summarizers, Webfurl's tree lets agents zoom into sections without reloading the whole page, with query-driven pre-unfolding ensuring LLMs see focused views—static navs stay cached, dynamic prices recompress cheaply. Token savings hit 90%+ on sites like Airbnb, plus vision model support for images and automatic CAPTCHA detection. The cross-user cache and one-command `start.sh` (with Docker MongoDB) hook browser agent llm builders tired of verbose prompts in github browser automation flows.

Who should use this?

AI researchers prototyping browser agent chrome tools or browser agent api integrations for n8n workflows. LLM devs building autonomous web scrapers needing browser agent parser efficiency for lookups on e-commerce or search sites. Indie hackers automating browser agent switcher tasks like form testing or data extraction without selenium bloat.

Verdict

Grab it if you're hacking browser agents—solid README, quickstart, and live stats make evaluation dead simple despite 19 stars and 1.0% credibility signaling early maturity. Polish tests and add REST API for production, but it's already usable for agent experiments today.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.