ericjang

ericjang / autogo

Public

Autoresearch for Go

10
1
100% credibility
Found Apr 30, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

AutoGo is a research codebase for training strong Go AIs from scratch using self-play, neural networks, and automated experimentation workflows.

How It Works

1
🔍 Discover AutoGo

You find this fun project that lets you build and train your own smart Go-playing buddy from scratch.

2
🚀 Start your playground

Open it in a ready-to-use space where everything is set up for you, no complicated installs needed.

3
🤝 Connect smart helpers

Link up AI thinkers to help guide experiments and make decisions automatically.

4
⚡ Watch AI learn Go

Your assistant runs smart games and training rounds on its own, getting stronger with each step while you watch the progress charts glow.

5
🎮 Play against your AI

Challenge your new Go player online and see how it performs against pros.

🏆 Strong Go master ready

Celebrate having your own powerful Go AI that beats top programs – all built with easy automation!

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is autogo?

Autogo is a Python framework for training strong Go bots from scratch via self-play on 9x9 and 19x9 boards, hitting 77% win rates against KataGo baselines. It automates the full research loop—data collection, hyperparameter tuning, and throughput optimization—using AI agents like Claude via simple CLI skills like `autoresearch` for metric optimization and `experiment` for quick tests. Developers get a dev container setup for local editing, SSH-dispatched Docker jobs across GPU clusters, and NFS-shared checkpoints for seamless scaling.

Why is it gaining traction?

Unlike raw AlphaGo repos, autogo prioritizes agent-driven autoresearch over manual scripting, letting Claude interpret results and iterate experiments in cheap, fast-signal domains like Go. Users notice 10x MCTS speedups from batched C++ inference, easy cluster adds via `cluster.py add user@host`, and scaling law insights transferable to LLMs or robotics. The web demo to play the AI hooks tinkerers instantly.

Who should use this?

RL researchers testing self-play scaling or recursive improvement; AI engineers prototyping distributed training stacks without Ray/Kubernetes overhead; game AI devs or robotics sim folks needing quick MCTS+NN baselines with automated tuning.

Verdict

Grab it if you're into autoresearch GitHub experiments like Karpathy/Claude-style workflows—solid for Go but generalizes well. With 10 stars and 1.0% credibility, it's early (thin tests, cluster-focused docs), so fork and validate locally before production scaling.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.