Learn

Learn turns amosblomqvist/learn, Amos Blomqvist's AI learning system from his video How I Use AI to Learn Things, into a local app. It runs his teaching skill and his research and diagram agents on pi, and adds worksheets, quizzes, a lesson outline, a plan map and a side chat. Everything stays on your computer; the server only listens on 127.0.0.1.

Install

Download Learn for your system from Releases, unzip it and open it. On first launch it asks for a notes folder: the tutor reads and writes files there and saves lessons into it. Then add a model provider in Settings → Providers: sign in with a Claude Pro/Max or ChatGPT subscription, or add an API endpoint. Credentials stay in pi's ~/.pi/agent/auth.json.

Builds aren't signed yet, so macOS asks you to allow Learn once in System Settings → Privacy & Security → Open Anyway, and Windows asks you to confirm it. On Windows, also install Git for Windows: the tutor runs its shell commands in Git Bash.

Learn is developed on macOS and runs on Windows and Linux too, though those get less testing.

Optional extras:

• Web research: pi install npm:pi-web-access
• Diagrams: rsvg-convert (librsvg) or ImageMagick. Mermaid diagrams also need Chrome or Chromium and only render when running from source.

Claude subscriptions

extensions/claude-subscription.mjs sends Claude subscription requests through oh-my-pi's Anthropic provider, which mimics Claude Code's requests. Anthropic may not treat this as permitted subscription use; use it at your own risk. API keys and custom endpoints use pi's normal transport.

Development

Run from source

Learn also runs as a local web page. Needs Node 22.19 or newer; npm install brings its own pi and Bun.

git clone https://github.com/j0team/learn
cd learn
npm install
npm run build
LEARN_VAULT=~/notes npm start

Then open http://localhost:4747.

Variable	Default
LEARN_VAULT	current directory	Notes folder the tutor reads and writes
LEARN_PORT	4747	Port
LEARN_SESSIONS	~/.pi/agent/learn-sessions	Where lessons are saved
LEARN_PI	bundled pi under Bun	pi executable
LEARN_PI_ARGS		Extra pi arguments

Working on Learn

LEARN_PORT=4750 npm start                     # API server
LEARN_API=http://127.0.0.1:4750 npm run dev   # UI with hot reload
npm test                                      # server and extension tests
npm run lint
npm run app:dev                               # Electron window from the checkout
npm run app                                   # package the app for this system into release/

The app bundles each system's Bun, so it's built on the system it's for; .github/workflows/build.yml builds all four and attaches them to a release when a v* tag is pushed.

• server/: the Bun server (server.mjs), the pi extension with Learn's tools (learn.ts) and the tutor prompts
• src/: the React UI (Tailwind theme in src/index.css, shared state in src/lib/store.ts)
• skills/ and agents/: the tutor's skills and the subagents they call (a researcher and two diagram makers)
• extensions/: the Claude subscription transport and the diagram render tools
• electron/: the desktop app

Roadmap

• Signed builds

Credits

The teaching approach, the teach skill and the subagents come from amosblomqvist/learn by Amos Blomqvist. Watch his video for the ideas behind it.

License

MIT. See LICENSE.