Everyone "learns with ChatGPT" now — and a week later all that is left is a chat transcript you will never reopen. Chats answer questions; they do not teach. Real learning needs structure, practice, feedback and repetition. That bugged me, so I built a tool that turns "I want to understand X" into an actual course.
It is Learn (Almost) Anything — a free, open-source desktop app for macOS and Windows. You type a topic and an agent interviews you, researches the material, designs a curriculum, writes illustrated lessons, quizzes you, grades your homework and schedules flashcard reviews. And the key part: the engine is the very Claude Code or Codex CLI you already pay a subscription for.
Why not just a chat
It is the difference between asking a smart friend questions all evening and having that friend sit down and teach you properly. Most AI tools stop at "here is some text" — this one closes the loop:
- Lessons that look like lessons — articles with real sourced images, Mermaid diagrams and interactive widgets. Every draft goes through an editor and fact-check pass.
- Comprehension tests — they check understanding, not verbatim recall, and interleave concepts from earlier modules.
- Real homework — essays, diagrams, GitHub repos and autograded tasks that actually run your code and make you retry until it passes.
- Spaced repetition — flashcards are pulled from every lesson, so the material sticks after you have read it.
For big goals ("become a data engineer", "academic painting from zero") there are roadmaps: they lay out stages and skills, run a diagnostic, and spawn courses from each skill.
Your subscription is the engine
The app is free and runs no paid backend. Every model call goes through an agent CLI already installed on your machine: Claude Pro/Max → Claude Code, ChatGPT/Codex plan → Codex. Courses, progress and media live locally — no server of ours sees your content.
Publish your best courses to the public catalog, install others, and translate any course into another language in full. Teams can run their own catalog inside their infrastructure with one Docker command.
Where to get it
I wrote up the full breakdown with all the screenshots on the project page. Grab a build (macOS .dmg / Windows .msi) from the GitHub releases. The source is open: github.com/legostin/learn-almost-anything.
Built on Tauri 2, React 19 and a Node sidecar using the Claude Agent SDK and Codex SDK. If you have been meaning to really understand something instead of "ask and forget" — give it a try.