Fluently

An Anki-style spaced-repetition flashcard app for learning languages. You build decks of cards, study the ones that are due, and rate how well you knew each answer. An SM-2 scheduler decides when to show each card again, so you spend your time on the material you're about to forget.

Live app: fluently.app.space

It's built on the DeepSpace SDK and runs on Cloudflare Workers + Durable Objects. Auth, the database, real-time sync, file storage, and the external-API proxy (LLM, text-to-speech, speech-to-text) all come from the SDK.

Features

• Decks and deck management: create, edit, and organize decks, each with an optional target and native language. Per-deck card counts (new, learning, review, due) and a daily cap on new and review cards.
• SM-2 spaced repetition: an SM-2 scheduler modeled on Anki, with learning and relearning steps, ease-factor adjustment, interval fuzz, leech detection, and per-deck setting overrides on top of global tuning.
• Customizable card types and fields: define card types with named fields (text, image, audio, or HTML), per-field side and required flags, reversible and cloze types, and front/back templates with CSS.
• AI-assisted card generation: generate cards in bulk from a prompt or a word list, with optional generated images and audio. Calls run through the DeepSpace integration proxy.
• Anki .apkg import: import existing Anki decks, including their note types, templates, and media, parsed from the .apkg package.
• Vocabulary library: browse a vocabulary dataset filtered by language, CEFR level, and part of speech, then add selected words as cards. Word lists can also be loaded on demand.
• Pronunciation review: a pronunciation card type that records your audio, transcribes it via speech-to-text, and returns an AI evaluation with a score and feedback. Reference pronunciation audio is sourced from Wiktionary.
• Review and browse modes: a focused review session for due cards with Again/Hard/Good/Easy ratings and live session counters, plus a browse mode for searching and editing the whole collection.
• Theming: built-in preset themes plus user-defined custom themes, with the active theme stored per user.
• Real-time sync: cards, decks, settings, and progress are stored in DeepSpace record collections and sync live across sessions.

Tech stack

• Frontend: Vite + React 19 + Tailwind v4, TypeScript, file-based routing (generouted)
• Backend: a single Cloudflare Worker (Hono) + Durable Objects, via the DeepSpace SDK
• AI and audio: OpenAI for text and image generation, and a speech integration for text-to-speech and speech-to-text, all called through the DeepSpace integration proxy (Vercel AI SDK on the worker)
• Data: DeepSpace record collections with per-collection RBAC

Getting started

Prerequisites

• Node 20+
• A DeepSpace account (free to sign in). The app uses DeepSpace for auth, data, file storage, and the integration proxy, and deploys to <name>.app.space.

Run it locally

npm install
npx deepspace login        # one-time, opens a browser tab
npx deepspace dev          # http://localhost:5173

Sign in, create a deck, and start adding cards.

Testing

npx deepspace test         # smoke, API, and end-to-end tests

The individual suites are available as npx deepspace test smoke, npx deepspace test api, and npx deepspace test e2e. npm run type-check runs tsc --noEmit.

Deploy

npx deepspace deploy       # → <name>.app.space

The deploy subdomain is the name field in wrangler.toml. Rename it there for your own deployment.

Project structure

• worker.ts: the Cloudflare Worker (routes, auth proxy, integration proxy, AI chat, server actions, R2 file storage, and cron)
• src/pages/: file-based routes (home, app shell, catch-all)
• src/components/app/: the app's components (review and browse modes, card and deck editors, card-type and field editors, AI generation, Anki import, word library, pronunciation review, theme and settings panels)
• src/schemas.ts + src/schemas/: record collections and their RBAC (decks, cards, card types, settings, themes, daily progress, media)
• src/utils/: domain logic (SM-2 scheduling, Anki import, card and media storage, the field system, vocabulary dataset loading, themes, and prompts)
• src/types.ts: shared domain types (cards, decks, card types, scheduling, media, themes, settings)

CLAUDE.md in the parent workspace tells AI coding agents to load the DeepSpace SDK skill before working here.

License

MIT, see LICENSE.

---

Built with DeepSpace.