Reki 歴

A modern spaced repetition desktop app. Tauri 2 + Svelte 5 + Rust.

Built as an Anki replacement that cuts the Python middle layer, ships a ~8 MB installer (vs Anki's ~150 MB), and uses the real FSRS-5 algorithm for scheduling. Supports Markdown, KaTeX math, syntax-highlighted code blocks, inline images, and AI-assisted card generation via Claude Code.

Status

Early development. Core review loop, deck management, AI generation, and SQLite persistence all work. No sync, no mobile, no plugins yet.

Features

• Real FSRS-5 scheduling via the fsrs Rust crate (same algorithm Anki uses)
• Full GFM Markdown — tables, task lists, strikethrough, blockquotes
• KaTeX math with $inline$ and $$display$$ delimiters
• Shiki code highlighting (VS Code TextMate grammars)
• Inline images — drag, paste, or pick; content-addressed storage
• Anki-compatible query language — deck:, tag:, is:due, prop:ivl>=10, re:, nc:, w:, wildcards, boolean operators, grouping
• Customizable keyboard shortcuts for rating
• Three themes — system / light / dark
• AI card generation through your local claude CLI subscription (no API key required)
• SQLite persistence with content-addressed media storage

Tech stack

• Frontend: Svelte 5 + TypeScript + Vite, pure CSS variables (no UI library)
• Backend: Rust via Tauri 2, rusqlite (bundled) for SQLite, fsrs for scheduling
• Rendering: marked + katex + shiki + dompurify
• Testing: cargo test for Rust, vitest for TypeScript

Building

Requires Rust toolchain, Node 22+, and the Linux dev libraries for Tauri (libwebkit2gtk-4.1-dev, libgtk-3-dev, etc.).

npm install
npm run tauri:dev     # dev with hot reload
npm run tauri:build   # production bundles (.deb, .rpm, .AppImage on Linux)

On macOS and Windows, run npm run tauri:build on the target OS to get native installers.

Testing

# Rust tests (db, srs, media)
cd src-tauri && cargo test

# TypeScript tests (markdown rendering)
npm run test

Architecture

Svelte UI (src/)
  ├── lib/components/  reusable components (Markdown, MarkdownEditor, Sidebar)
  ├── lib/pages/       Review, Decks, Browse, Generate, Settings
  ├── lib/stores/      data, theme, router, shortcuts, srs
  ├── lib/markdown.ts  GFM + KaTeX + Shiki pipeline
  ├── lib/media.ts     image upload bridge
  └── lib/query.ts     Anki-compatible search parser

Rust backend (src-tauri/src/)
  ├── db.rs     SQLite schema, migrations, CRUD commands
  ├── srs.rs    FSRS-5 scheduler wrapper
  ├── media.rs  content-addressed image storage
  └── lib.rs    Tauri app setup + command registry

Design docs (docs/superpowers/specs/ and plans/)

The core idea: no Python middle layer. The frontend talks to Rust directly through Tauri commands, cards and media persist to SQLite under the OS app-data directory, and everything that can run in Rust does.

License

MIT. See LICENSE.

Note on the original Anki

This is a ground-up rewrite that does NOT reuse any code from the original Anki project (which is AGPL-3.0). It only borrows the FSRS algorithm (MIT licensed) and the general concept of spaced repetition flashcards. All other code is new.