Show HN: Lathe – Use LLMs to learn a new domain, not skip past it
Points and comments are a snapshot, not live.
Lathe generates hands-on technical tutorials from LLM prompts, then you work through them yourself in a local UI.
Lathe is a Go CLI tool paired with LLM skills that generate multi-part technical tutorials on demand. Rather than having an LLM write code for you, the tool creates structured learning materials you work through yourself. The system stores tutorials in ~/.lathe/tutorials/, serves them via a web UI on localhost:4242, and supports iterative refinement through skills like /lathe-extend, /lathe-verify, and /lathe-ask. The author built it to recreate hands-on learning moments while using LLMs to break into domains lacking human-written resources, such as 3D slicer software or embedded development in Zig. Two writing voices ship by default: plainspoken (precise, no persona) and companion (warm, wry). The author acknowledges tutorials will hallucinate but argues the risk is lower when you type code yourself and actively catch errors. Installation is a single binary; skills integrate with Claude Code, Cursor, and Codex.
What commenters are saying
Commenters found the concept promising as a way to use LLMs for learning rather than code generation. One noted it eases the barrier to entry on unfamiliar projects. A commenter questioned why this improves on a hand-written prompt and mentioned ChatGPT Study mode as an alternative. Another user described a similar pattern at work: pairing a custom CLI app with LLM skills for deterministic, repeatable tasks like executive summaries of backlogs. One person asked whether Lathe could search for and supplement human-written tutorials first rather than generating from scratch. Questions emerged about broader LLM support (ollama, Gemini) beyond Claude/Cursor/Codex.
The author confirmed daily use and stability on Claude Code/macOS, acknowledged the tool's "vibecoded" nature, and expressed openness to supporting additional LLM providers in v0.4.0.