karpathy/autoresearch
★ 98,166 stars
Covered in week 2026-W11 · the 2026-03 report
AI agents running research on single-GPU nanochat training automatically
What autoresearch is
Seeing as there seems to be a lot of interest in tinkering with autoresearch on much smaller compute platforms than an H100, a few extra words. If you're going to try running autoresearch on smaller computers (Macbooks etc.), I'd recommend one of the forks below. On top of this, here are some recommendations for how to tune the defaults for much smaller models for aspiring forks
From the repository README · read it on GitHub →
Transcript
What if AI could do research while you sleep? Andrej Karpathy's autoresearch lets AI agents run autonomous experiments on LLM training overnight. The agent modifies train.py, trains for five minutes, checks if the result improved, and keeps iterating. You just program the markdown instructions and wake up to a log of around a hundred experiments. It's single GPU, single file, and completely self-contained. Star the repo at github.com/karpathy/autoresearch