Karpathy Autoresearch — AI Agent Framework: Live Stats & TrendScore

Live GitHub stats, community sentiment, and trend data for Karpathy Autoresearch. TrendingBots tracks star velocity, fork activity, and what developers are saying — updated from real data sources.

GitHub data synced: Mar 26, 2026 • Sentiment updated: Apr 12, 2026

GitHub Statistics

• GitHub Stars: 79,088
• Forks: 11,536
• Open Issues: 185
• Primary Language: Python
• Weekly Star Velocity: +1532 stars/week

Community Sentiment

Community Buzz: Karpathy's autoresearch is just automated overfitting no? (HackerNews) and I just spent yesterday applying Kaparthy's autoresearch on an ML problem (HackerNews)

Pros & Cons

What People Love

novel research discoveries, automated hyperparameter sweeps (HackerNews), increased efficiency (HackerNews)

Common Complaints

context anxiety, overfitting

Biggest Positive: novel research

Biggest Negative: overfitting risk

Why Karpathy Autoresearch Stands Out

autoresearch offers a unique approach to autonomous research by leveraging AI agents to iteratively experiment and refine LLM training models, addressing the challenges of manual hyperparameter tuning and experiment management. This is achieved through its single-file modification design, fixed time budget, and self-contained architecture. By automating the research process, autoresearch enables researchers to focus on high-level decisions and explore new research directions. The project's emphasis on comparability across experiments and platforms also ensures that results are reproducible and reliable.

Built With

Build a research agent that runs experiments on nanochat training — autoresearch simplifies single-GPU setup and experiment iteration via AI agent, Build a multi-agent orchestration framework — autoresearch automatically chains search, extraction, and synthesis agents with minimal human intervention, Build a literature review tool that reads 50 papers — autoresearch enables DeerFlow chain automation for synthesis and aggregation of research results, Build a GPU compute cluster for training LLMs — autoresearch streamlines single-GPU setup and experiment iteration via AI agent, Build an open-source self-contained research framework — autoresearch reduces external dependencies and complexity via fixed time budget and self-contained design

Getting Started

1. 1. Install uv project manager (if you don't already have it): `curl -LsSf https://astral.sh/uv/install.sh | sh`
2. 2. Install dependencies: `uv sync`
3. 3. Download data and train tokenizer (one-time, ~2 min): `uv run prepare.py`
4. 4. Manually run a single training experiment (~5 min): `uv run train.py`
5. 5. Try running the agent with a prompt like `Hi have a look at program.md and let's kick off a new experiment! let's do the setup first.` to verify it works

About

AI agents running research on single-GPU nanochat training automatically

Category & Tags

Category: research

Market Context

Autoresearch is competitive with other AI solutions