Local-first study cockpit
Study TUI
A terminal-native AI study tool for real document libraries: PDFs, image-heavy notes, Calibre books, and Zotero papers.
Problem / itch
Students already have messy local material, but most AI study tools force another cloud silo or disposable chat. The itch was a keyboard-first cockpit that loads what you already have and turns it into quizzes, flashcards, notes, exports, and review.
What I built
- Built fuzzy document loading across local folders, Calibre metadata.db, and Zotero, with document-hash-tied progress.
- Added PDF/image parsing, figure-aware reading, BM25 search, multi-document sessions, and a persistent document status bar.
- Built study actions: generated quizzes, flashcards, summaries, SQLite-backed notes, Pomodoro, Anki/Markdown/PDF export, and Manim concept animations.
- Kept it model-provider flexible: local or API models, DuckDuckGo optional web search, context compaction, memory blocks, and tool-calling agents.
Results
Load a chapter, ask for main ideas, quiz weak points, make flashcards from misses, then review document-linked progress later.
Supports PDF/image loading, OCR via EasyOCR, figure-aware reading, BM25 search, multi-document sessions, Calibre, Zotero, web search, and provider switching.
Study actions include AI-generated quizzes, flashcards, summaries, SQLite-backed notes, Pomodoro, Anki .apkg export, Markdown/PDF export, and Manim concept animations.
Package metadata lists Python 3.10+, MIT license, `study` and `study-tui` entrypoints, optional extras for animation, Anki, Zotero, and dev tests.
Status
Shippable local-first study workflow; strongest builder/product proof.