karpathy/nanochat

From the README

nanochat nanochat is the simplest experimental harness for training LLMs. It is designed to run on a single GPU node, the code is minimal/hackable, and it covers all major LLM stages including tokenization, pretraining, finetuning, evaluation, and inference.

Setup prompt

Paste this into Cursor or Claude Code. It only includes install lines found in the README, and it asks the agent to review the code first.

Set up the GitHub repository karpathy/nanochat (The best ChatGPT that $100 can buy.).
Repository: https://github.com/karpathy/nanochat

Official install commands found in the README:
uv sync --extra gpu    # Use for CUDA (A100/H100/etc.)
uv sync --extra cpu    # (or) Use for CPU-only / MPS
uv sync --extra gpu --group dev

Clone the repository and follow those steps. Do not invent package names or extra install commands.
Review the code before running anything.

Buzz by window

Same mentions, three half-lives. This is not a star history. Star gain appears after a second daily snapshot.

• Today1.30
• Week2.02
• Month2.49

Spotted on

• X @Yance16zNov 3, 2026
• X @SanketShipsOct 30, 2026

Related

• pbakaus/impeccable

  The design language that makes your AI harness better at design.

• thedotmack/claude-mem

  Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context…

• Nutlope/hallmark

  Anti-AI-slop design skill for Claude Code, Cursor, and Codex.

• anomalyco/opencode

  The open source coding agent.