A narrow, specific setup: autoresearch runs AI agents that carry out research on single-GPU nanochat training automatically. The repo is on GitHub. It fits people running small-scale LLM training experiments who want agents driving the loop.