Autoresearch
An open-source project that lets AI agents autonomously run LLM training experiments and keep the best model changes.
🛡️ AgentReady threat assessment
MAESTRO 7-layer threat model + OWASP AIVSS risk score for Autoresearch, derived from its capabilities.
These scores are auto-generated from public information (the agent's own listing, docs, and repository) using the canonical OWASP AIVSS formula and the MAESTRO framework — an estimate for guidance, not a penetration test, audit, or certification. See the scoring methodology — every score is re-derived by the same automated method as an agent's public evidence changes.
Overview
Autoresearch is an open-source project by Andrej Karpathy that lets AI agents run autonomous machine learning research loops on a small but real LLM training setup. The repository is designed so an agent edits the main training file, launches a fixed 5-minute experiment, evaluates whether the result improved, and then keeps or discards the change before repeating the cycle. Its README describes the setup as a lightweight autonomous research organization driven by instructions in a program.md file rather than traditional manual code iteration. The project is built around a simplified single-GPU nanochat training workflow and is aimed at developers and researchers exploring automated model improvement, agent-driven experimentation, and compact research loops on their own hardware. :contentReference[oaicite:0]{index=0}
Key features and capabilities
- open-source
- LLM training
- autonomous research
- machine learning
- single GPU
- nanochat
- experiment loops
- PyTorch
- model optimization
- research automation
Use cases
- Running autonomous overnight experiments to improve small language model training setups.
- Testing agent-driven code changes in a controlled single-file training workflow.
- Exploring automated research loops for architecture, optimizer, and hyperparameter changes.
- Studying how AI systems can manage iterative machine learning experimentation with minimal human intervention.