What's moving in open AI right now — GitHub's fastest-growing repos, Hugging Face's trending models, and the star race across 20 open-source agent frameworks. Public data, credited to its authors, cached kindly, with honest fallbacks.
Source: github.com/trending (weekly + monthly windows), parsed server-side by our Cloudflare Pages Function and edge-cached about 1 hour. GitHub publishes no official trending JSON API, so we read the public page and always link every repo straight to its home. Each card is labeled with the window its star gain measures. All repos belong to their authors — we aggregate and credit, never copy.
Source: the public Hugging Face models API (sort=trendingScore), fetched directly by your browser — this request never touches our servers. Every model links to its Hugging Face page and belongs to its authors. ABUZ8 is not affiliated with Hugging Face; our own account currently has no public models — publishing sovereign models is Roadmap.
Source: the public GitHub REST API, aggregated by our Cloudflare Pages Function and edge-cached about 30 minutes, for a curated set of 20 open-source agentic projects. We track them because they matter to anyone building agents locally; the code and the stars belong to their authors and communities. Want to run agents yourself? Start at our Agents hub and Models guide.
The open preprint server run by Cornell. New machine-learning, language, and AI papers land here daily, free to read.
Daily community-voted trending AI papers, usually with linked models, datasets, and demos alongside each paper.
hf.co/papers →Our free learning hub — plain-language guides to local models, agents, and the sovereign AI stack.
education.html →440 posts on running AI locally: model guides, tool walkthroughs, agent build notes, and honest write-ups.
Browse the blog →How to choose and run local models on your own hardware — quantization, VRAM budgets, and runtimes.
models.html →Our deeper board pages — more GitHub trending views and the Hugging Face ecosystem hub, both credited feeds.
No fabricated numbers, no fake urgency, no invisible tracking. Here is exactly where every number on this page comes from and what happens when a feed goes down.
GitHub's public trending pages, the public GitHub REST API, and the public Hugging Face API. We never scrape private data and never claim anyone's work as ours — every card links to its author.
Two Cloudflare Pages Functions do the work: /api/github-trending parses GitHub's weekly and monthly trending pages (GitHub has no official trending JSON API), and /api/github-agents pulls live star counts for 20 curated open-source agent projects.
Trending is cached about 1 hour, agent stars about 30 minutes, at Cloudflare's edge. That keeps the boards fresh without hammering anyone's servers.
If the trending parse fails, you get a real frozen snapshot captured 2026-05-24, labeled "cached snapshot." If GitHub rate-limits the star board, affected rows fall back to last-known cached counts and the board says so. If Hugging Face is unreachable, you get a labeled feed-offline note — never an invented number, never an empty broken board.
Each tab loads its feed once, the first time you open it. Nothing auto-refreshes in the background. Reload the page when you want fresh data.