apfel - Top 1% on GitHub (actually top 0.003%)

So this one I didn’t see coming.

While preparing my OpenClaw workshop in spring 2026, I was looking for a way to call a small local model fast, offline, no infrastructure, from a script. And then I noticed (Daumen mal Pi, after sinking 170 hours into local AI experiments) that every modern Mac running macOS Tahoe already has Apple Intelligence sitting right there on the Neural Engine. Yeah, you need to enable it. But after that … it is just there.

Just nobody talks to it from the command line. So I built a CLI that does.

I called it apfel.

From one post to 5,000+ stars

I pushed it. I communicated it. I posted about it once. And then it just kept going, going, going.

In May 2026 it had 5,316+ stars on GitHub. Developers from Nvidia, Google, Apple, PlayStation, … (and a couple of names I can’t mention) were following the repo. I got invited to a podcast.

And it kept going: on 11 November 2026, the GitHub API showed 6,597 stars for Arthur-Ficial/apfel.

What “top 1% of GitHub” actually means

“Top 1% of GitHub” is one of those phrases people throw around. So I went and checked. Real numbers, queried from the GitHub Search API in May 2026:

• Total public repos on GitHub (Octoverse 2025): 395,000,000
• Repos with >= 1 star: 29,558,539 (only ~7.5% of repos ever get a single star)
• Repos with >= 5 stars: 4,782,533
• Repos with >= 100 stars: 457,528
• Repos with >= 1,000 stars: 61,752
• Repos with >= 5,000 stars: 11,802

1% of 395 million public repos is the top 3,950,000 repos. That bar is so low that any repo with ~5 stars or more clears it. So yes, “top 1%” is technically true once you have any meaningful star count.

Where apfel actually sat

With 5,316 stars, apfel was inside the 11,802 club: repos with 5,000+ stars. That is roughly the 11,000th most-starred public repo on the platform.

The math:

• vs all 395M public repos: 11,000 / 395,000,000 = top 0.003%
• to be clear: that is 300x better than the “top 1%” claim
• “top 1% of GitHub” is the conservative version. The actual position was top 0.003%

Either way: one of ~12,000 repos out of ~395 million ever crosses 5,000 stars. apfel joined that club. I am still surprised.

(Source: GitHub Search API, queried in May 2026; Octoverse 2025: 395M public repos, 630M total including private.)

Install

brew install apfel

That’s it. No API key. No model download. No config. The model is already on your Mac, apfel just lets you talk to it. Free. MIT.

It is pipe-friendly and scriptable, and it also runs as an OpenAI-compatible HTTP server on localhost:11434, so you can drop it in anywhere your tools expect OpenAI.

• apfel.franzai.com - the landing page.
• github.com/Arthur-Ficial/apfel - the repo. Please star it. It helps.