Lathe

// Overview

Lathe describes itself as an experiment in using LLMs to teach you rather than think for you. You prompt a coding agent with a topic, its skills write a hands-on, multi-part technical tutorial, and then you work through that tutorial by hand in a local reader built for the purpose — typing the code out yourself instead of watching an agent produce it. Every tutorial records its sources, which model wrote it, and the prompt that set its voice.

It has two halves. The LLM skills (`/lathe`, `/lathe-extend`, `/lathe-verify`, `/lathe-ask`, `/lathe-tag`, `/lathe-work`) run inside an interactive coding-agent session; the Go CLI stores tutorials under `~/.lathe/tutorials/`, serves the rendered output at `http://localhost:4242`, and owns all durable state. The binary never calls a model itself — all model work stays in whatever agent session and subscription you already use. The skills are bundled into the binary and install into Claude Code, Cursor, Codex, Gemini CLI, opencode, Cline, Windsurf or Antigravity with `lathe skills install`.

The reader is where the project spends its effort: search and filtering over a tutorial library, hover table-of-contents navigation, and side-notes written throughout to prompt deeper thinking. Its Ask, Verify and Add-a-part buttons work two ways — run `/lathe-work` once and a worker loop drives them directly in your agent session, with answers rendering in place; with no worker connected, each button hands you the exact slash command to paste. The author is candid that Lathe is itself vibecoded, that its tutorials are LLM output rather than a substitute for human-written material, and recommends reaching for the largest thinking model you have.

// What it does

• Generate a single-part or multi-part technical tutorial from any prompt, inside the coding agent you already use
• A purpose-built local reader at localhost:4242 with search, filtering, table-of-contents navigation and inline side-notes
• Ask, Verify and Extend a tutorial from the reader — live through a `/lathe-work` worker loop, or by pasting the command it hands you
• Agent-agnostic skills bundled into the binary: Claude Code, Cursor, Codex, Gemini CLI, opencode, Cline, Windsurf and Antigravity targets
• The CLI never calls a model, so tutorials run on your existing agent subscription — including a local OpenAI-compatible endpoint such as Ollama
• Every tutorial documents its sources, the model that wrote it, and the voice prompt behind it

// Getting started

Lathe is a single self-contained binary. You need `lathe` on your $PATH and a supported coding agent for the skills to run in.

brew install devenjarvis/tap/lathe

# or
curl -sSf https://raw.githubusercontent.com/devenjarvis/lathe/main/install.sh | sh

# or
go install github.com/devenjarvis/lathe@latest

lathe skills install                   # ./.claude/skills/<name>/SKILL.md
lathe skills install --user            # ~/.claude/skills/<name>/SKILL.md
lathe skills install --agent cursor    # Cursor slash commands
lathe skills install --agent all       # every supported target
lathe skills list

/lathe build a 3D Slicer in Erlang

lathe serve

/lathe-work

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Get from zero to one in an obscure or very young domain where no human-written tutorial exists yet
• Learn by typing the code yourself rather than having an agent write the project for you
• Ask a question about the part you are reading and have the answer land in the reader, without leaving the tutorial
• Extend a tutorial you finished with another part, or have the agent verify that its code actually compiles and runs

// How Lathe compares

Lathe alongside other open-source learning & tutoring tools AI/TLDR tracks, ranked by GitHub stars.