Back to all agents
#13 by GitHub stars · Research

autoresearch

Andrej Karpathy

An experimental agent loop that changes training code, runs short experiments, and keeps improvements to a small language model.

View source on GitHub 93,897 stars checked 2026-08-15
Access
Open source
Setup
Technical
Pricing
Free software; substantial GPU costs
License
MIT
Maintenance
Limited recent activity
Last repository push
2026-03-26
Our verdict

Compelling but highly specialized

It is a clear, minimal example of an agent conducting measurable machine-learning experiments autonomously.

Who it fits

ML researchers with suitable NVIDIA hardware who want a controlled autonomous experimentation baseline.

Who should skip it

General research, consumer tasks, or users without a strong GPU and ML engineering experience.

What you need

  • NVIDIA GPU tested on H100
  • Python 3.10+ and uv
  • ML experiment supervision