The selection
A radar I run at arsentev.ai/github selects and reviews open-source AI projects from GitHub, Hugging Face and Hacker News, one page per project. Between 11 June and 12 September 2026 it selected 490 projects: 422 from GitHub, 63 from Hugging Face, 5 from Hacker News. By kind that's 427 repositories, 48 Hugging Face models, 15 Hugging Face Spaces.
By month of selection: June 70, July 134, August 204, September (up to the 12th) 82.
Median GitHub stars at the moment of selection was 493. Top languages across the repositories: Python 135, TypeScript 104, JavaScript 51, Rust 19, Swift 16. On Hugging Face the most common tasks were text-generation 16, image-text-to-text 16, image-to-video 7, text-to-speech 6.
One snapshot so far
There is one snapshot of this data, taken on 12 September 2026, the last day of selection. Depending on the project, the gap between when it was selected and the snapshot ranges from 0 days to about three months.
On that snapshot, of the 427 GitHub repositories: 416 were still available, 10 deleted (the API returns 404), 1 archived.
Of the available repositories, 19% have no licence or a licence GitHub doesn't recognise. The breakdown: MIT 218, Apache-2.0 93, NOASSERTION (GitHub can't tell) 43, no licence file 36, AGPL-3.0 16, GPL-3.0 8.
71.6% had a push in the last 30 days.
The sample is curated, not random. It's what the radar picked, so it says nothing about GitHub as a whole.
Tracking going forward
From 12 September a script takes a daily snapshot of every selected project. For GitHub repositories that's stars, forks, open issues, licence, last push, archived status. For Hugging Face models and Spaces it's likes, downloads, last modified. Removed repositories stay in the data with their status attached, so the set doesn't drift toward only the ones that are still around.
The dataset is open, under CC BY 4.0, on Zenodo, with three files: items.csv, snapshots.csv and stats.json. Every number above comes from the build script running on those CSV files. A new version comes out on the 1st of each month, and the series DOI always points to the latest one: https://doi.org/10.5281/zenodo.22730450. The numbers in this piece are from the 2026-09-13 edition: https://doi.org/10.5281/zenodo.22730451.
Three checks before you build on one
Before picking up a repository from a list like this, or from any trending list, these are the three things I'd check myself:
- does the repository still open
- is the licence field on GitHub set to a known licence, not blank and not NOASSERTION
- was there a recent push
None of these take more than a minute, and all three are visible right on the repository page, before you clone anything or read a line of code.
Evgenii Arsentev, PhD. The dataset and the radar pages are at arsentev.ai/github.
Top comments (0)