marspa73

marspa73 / atarijam

Public

🕹️ Is this the smallest language model in the world? I just managed to squeeze real artificial intelligence into 30 kilobytes, running on a 1979 Atari 800. A fully generative, deterministic language model, powered by a neural network and built to run on 8‑bit hardware.

23
2
100% credibility
Found Apr 05, 2026 at 23 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Assembly
AI Summary

A set of ultra-compact generative language models trained to run as interactive chat programs on 1979 Atari 800 and 130XE computers, with sizes from 1.6KB to 76KB brains.

How It Works

1
🔍 Discover JAM

You stumble upon this exciting project about the world's tiniest AI brains that chat back on a 1979 Atari computer.

2
📥 Grab the files

Download the folder with ready-made AI personalities and simple build tools to your computer.

3
🧠 Build your AI

Run the quick build to create fun, tiny chat programs in moments, or use the included ready ones.

4
💾 Load into Atari

Put the program file into an Atari emulator or your real vintage machine to get it running.

5
Pick a personality
🐛
Tiny Pico

Ultra-small and quirky for quick fun.

📦
Compact JAM

Balanced size with strong chat skills.

🚀
Advanced XE

Bigger brain for smarter talks.

👶
Kid JAM

Playful child-like responses.

6
💬 Start chatting

Type simple questions or prompts and watch the AI generate replies right there on screen.

🎉 Magic on old hardware

Feel the wonder of a real thinking machine alive in kilobytes on 40-year-old tech.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 23 to 23 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 atarijam?

atarijam squeezes generative language models into kilobytes of Assembly code that runs natively on a 1979 Atari 800, delivering real artificial intelligence on 8-bit hardware. It offers six variants—from a 1.6KB picojam proof to 76KB bank-switched models—all deterministic, neural-powered, and capable of chat-like responses without lookups. Users get prebuilt XEX binaries for emulators, ATR disks for real Ataris, and Python build scripts to retrain with PyTorch on custom data.

Why is it gaining traction?

It claims the title of smallest language model on GitHub, packing small language models into absurdly tiny footprints that actually work on vintage 6502 chips. The hook is firing up a 30KB LLM on Atari hardware via a one-line build_all.py, blending retro nostalgia with extreme quantization feats no modern framework matches. Pretrained weights and verifiable builds lower the barrier for skeptics.

Who should use this?

Retro hackers loading SD cards into Atari 800s for AI chats, embedded devs benchmarking tinyML limits on 8-bit MCUs, or conference speakers demoing "smallest LLM GitHub" stunts. Perfect for training kid-like personalities or POKEY audio experiments without buying new silicon.

Verdict

Worth forking for atarijam fans—MIT-licensed, docs solid, builds reliable with cc65—but 15 stars and 1.0% credibility score mean it's a clever prototype, not production-ready. Test on Altirra emulator first.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.