Plain-AyI
← All projects
Training & fine-tuning

autoresearch

Lets an AI coding agent run small AI-training experiments on its own overnight and keep what works.

Fast-rising

What it does

autoresearch gives an AI coding agent (such as Claude or Codex) a small but real setup for training a language model, the kind of AI behind chatbots. The agent changes the training code, trains for five minutes, checks whether the result got better, and then keeps or throws away the change. It repeats this again and again.

Because each run is limited to five minutes, the makers say you can expect about 12 experiments an hour, or around 100 while you sleep. In the morning you get a log of what was tried and, hopefully, a better model. Results are measured with one score, where lower is better.

The person doesn't edit the training code directly. Instead, they write plain instructions in a text file that guides the agent, and improve those instructions over time. The training code is a simplified, single-graphics-card version of the maker's nanochat project.

Who it's for

AI researchers and hobbyists who want to experiment with letting AI agents run machine-learning research automatically. It suits people comfortable with Python and the command line.

Can I use it without coding?

Developers only. You need to run commands, set up Python tools, and work with an AI coding agent inside the code folder. New to this? Start here.

Can I run it?

Cost
Free, but needs an AI accountThe software is free, but it uses an online AI service (like OpenAI or Anthropic), which usually charges for use.
Runs on
Linux
Computer
Works best on a powerful computer or one with a graphics card (GPU)

Try it

Using Claude Code, Cursor or another AI coding assistant? Paste the prompt and it will walk you through the setup.New to this?

Before you install

  • Only download from the official GitHub page. Copies with similar names exist for many popular projects.
  • Plain-AyI is an independent guide and doesn't host or review the software it lists. Please check each project's official page before installing.
  • License (MIT, stated in its README): Very permissive: free to use, change and share, including for business.
  • More on staying safe

Where people are talking about it