Flyt

Learn Norwegian Bokmål through lessons, reading, and spaced repetition.

Flyt brings learning and practice into one place. Work through a lesson, read Norwegian with a dictionary at hand, and review the concepts and words you save.

Open Flyt · Run locally · Contribute · Privacy

The hosted app is free and uses Google Sign-In. There is no paid tier or expiring trial.

How learning works

1. Learn a concept. Follow a CEFR-organised curriculum with explanations, examples, audio, and interactive exercises. Completing a lesson adds its concepts to your review schedule.
2. Meet it in context. Read curated stories, import articles, or generate stories around vocabulary from your learning activity. Look up unfamiliar words as you read and save them for practice.
3. Review over time. FSRS spaced repetition schedules your next reviews. Cards rotate through variants of a concept, and your review history informs your progress.

You can also:

• Explore the dictionary: look up definitions, inflections, and pronunciation from Ordbokene data, with English glosses supplied by the companion norsk-lemma project.
• Read across the web with FlytLese: the Chromium extension offers word lookup, text translation, vocabulary saving, and article import. Chrome and Edge store listings are not published yet; Firefox and Safari are not supported.
• Ask for help: the chatbot can use your message and current page context. AI generation, chat, and speech features depend on the deployment's provider and service configuration.

Run locally

The development stack needs Python 3.12+, uv, Node 24+, pnpm 12, PostgreSQL 16, and Redis. The exact pnpm version is pinned in package.json; CI and the backend container use Python 3.12. Docker is also required for the isolated end-to-end test runners.

From a checkout of this repository, install the workspace dependencies and copy the environment templates:

pnpm install --frozen-lockfile
cp backend/.env.example backend/.env
cp frontend/.env.example frontend/.env.local

Then follow the local setup guide to configure PostgreSQL and Redis, set real signing and provider-encryption keys, install the locked Python dependencies, and apply migrations. The environment templates need configuration before the API can start. The guide includes a local login without Google OAuth and options for development without an AI provider key.

Once configured, start each process in a separate terminal:

# API
uv run --directory backend uvicorn flyt.main:app --reload

# Background worker for reading imports and generation
uv run --directory backend arq flyt.worker.WorkerSettings

# Web app
pnpm --filter frontend dev

Open http://localhost:5173.

Load learning content

Migrations create the database schema; lesson and dictionary content are imported separately from public releases:

Content	Producer
Lessons and their audio references	norsk-lesson-factory
Dictionary, pronunciation, and vocabulary frequency ranks	norsk-lemma

Use the content import guide for download URLs, preview commands, and frequency-deck creation. Preview lesson imports before applying them: a lesson release replaces the curriculum, removing omitted lessons and their lesson-owned learner state. Lesson audio remains hosted by the content producer.

Contribute

Read CONTRIBUTING.md for setup, environment variables, tests, hooks, and conventions. Backend tests require a disposable database; the testing guide provides an isolated setup. The browser and extension end-to-end runners create and clean up their own Compose stacks.

Pull requests and master run the same CI checks, covering backend, frontend, extension, migrations, generated API types, and security. Run the affected workspace checks before submitting a change.

Repository layout

Directory	Responsibility
frontend/	React, TypeScript, Vite, TanStack Router/Query, and Tailwind CSS web app
backend/	FastAPI API, SQLAlchemy models, Alembic migrations, FSRS scheduling, and arq workers
extension/	FlytLese Chromium Manifest V3 extension
packages/ui/, packages/lexicon/	UI and dictionary components shared by the web app and extension
morph-svc/	Optional morphology fallback service
stt-svc/	Optional speech transcription service behind the backend
knowledge/	Current product, domain, architecture, and operations documentation

PostgreSQL stores durable content and learner state; Redis supports transient coordination and generation. The API and background worker run as separate processes. Start with system architecture and architecture decisions for the engineering boundaries.

Self-hosting and privacy

Backend and frontend container images publish to GHCR after CI passes on master. Deployment is managed outside this repository. The operations guide covers migrations, workers, readiness checks, and recovery; the optional morphology and speech services have their own deployment guides.

The hosted app's privacy policy explains account and learning data, external AI processing, error monitoring, and data requests. Story generation can send selected vocabulary and story instructions to a configured provider. Chat requests can send your message, recent history, and page context to the selected provider. You can delete your account in the app.

For deployment details, see authentication and API boundaries and provider credential operations.

Attribution and licenses

Flyt's application source code is licensed under the MIT License. Imported dictionary and frequency data retain their own licenses and attribution requirements.

Dictionary data — CC BY 4.0

Dictionary data is derived from Bokmålsordboka / Nynorskordboka:

Bokmålsordboka/Nynorskordboka, Universitetet i Bergen og Språkrådet, ordbøkene.no, CC BY 4.0.

License · Ordbøkene open data · Citation guide

The norsk-lemma pipeline fetches, enriches, and packages this data. English glosses are LLM-generated and may contain errors; they are a learning aid, not an authoritative translation of the source dictionaries.

Vocabulary frequency data — CC BY-SA 4.0

Frequency ranking for the word-pack decks is derived from the Norwegian Kelly list, used under CC BY-SA 4.0:

Norwegian Kelly list, UiO Text Laboratory (tekstlab.uio.no/kelly), CC BY-SA 4.0.

Speech models and runtime

The optional CPU speech service uses NB-Whisper Medium under Apache-2.0 and faster-whisper / CTranslate2 under MIT. See the speech service notices for model provenance and distributed license text. Browser clients use the authenticated backend proxy; the frontend does not ship a model runtime.