Learnarr

---

Learnarr is a self-hosted companion in the spirit of Sonarr and Radarr. Pick an episode you're about to watch, and Learnarr reads its subtitles, works out how much of the dialogue you already understand, and asks the AI provider you choose for the words worth learning. You go through the suggestions, keep the ones you want, and study them as flashcards built around the real lines you'll hear. Study in the app on your phone, or send the cards to Anki.

Status: early (v0.3). It works end to end, and there will be rough edges. The wiki covers install, setup and how it all works.

Features

• Your library, your subtitles. Browse Plex or Jellyfin and use the subtitle tracks you already have. When there are none, Learnarr downloads one from OpenSubtitles.com with each person's own free account. No media server? Search OpenSubtitles.com for any film or show, or upload a subtitle file of your own. OpenSubtitles.com needs each person's own free account and API key, added in Settings → Account.
• "% you know" for every episode. Learnarr finds each word's dictionary form and counts how much of the dialogue you already know, so you can pick shows at your level.
• Words worth learning, not every word. Your AI provider suggests words for your level. Learn them, mark the ones you already know, or skip them. Or tap any word in the transcript.
• Real sentences. Every card uses the line from the episode where the word came up.
• Spaced repetition that fits a phone, with undo. Push cards to Anki through AnkiConnect, or export a file Anki imports.
• Tap any word for its dictionary form, how common it is and its Wiktionary definitions.
• Notifications by web push or ntfy: a reminder when cards are waiting, and the episode's words when you press play on Plex or Jellyfin.
• Bring your own AI. Anything OpenAI-compatible (OpenAI, Ollama, LM Studio, OpenRouter, Groq, Mistral, Gemini) or Anthropic's Claude. Run a local model and nothing leaves your network.
• For the whole household. The first account is the admin; everyone else gets their own account, languages and cards on the same server.

Install

Learnarr runs as one Docker container with one /config folder. No environment variables are needed: the setup wizard does the rest.

Docker Compose

services:
  learnarr:
    image: ghcr.io/smoothless-io/learnarr:latest
    container_name: learnarr
    ports:
      - "7979:7979"
    volumes:
      - ./config:/config
    environment:
      - PUID=1000
      - PGID=1000
    restart: unless-stopped

docker compose up -d

Docker

docker run -d --name learnarr \
  -p 7979:7979 \
  -v ./config:/config \
  -e PUID=1000 -e PGID=1000 \
  --restart unless-stopped \
  ghcr.io/smoothless-io/learnarr:latest

Then open http://<your-server>:7979. The wizard creates the admin account, connects Plex or Jellyfin and your AI provider, and asks which language you're learning.

Images are built for linux/amd64 and linux/arm64.

Updating

docker compose pull && docker compose up -d

Database changes are applied automatically when the container starts.

Backups

Everything Learnarr keeps is in /config: the database, its encryption keys and downloaded language models. Back up that folder. Without its encryption.key, saved passwords and API keys can't be read.

Configuration

Everything can be set in the app. These environment variables are optional:

Variable	What it does
PUID, PGID	The user and group that own /config (id on the host shows yours). Default 1000.
PUBLIC_URL	Learnarr's address behind an HTTPS reverse proxy, e.g. https://learnarr.example.com. Turns on secure cookies.
MEDIA_SERVER, MEDIA_SERVER_URL, MEDIA_SERVER_TOKEN	Pin the media server (plex or jellyfin) instead of setting it in the app.
AI_PROVIDER, AI_BASE_URL, AI_API_KEY, AI_MODEL	Pin the AI provider (openai-compatible or anthropic).
AUTH_SECRET, ENCRYPTION_KEY	Use your own keys instead of the ones generated in /config on first start.
PORT	The port inside the container. Default 7979.
TRUSTED_ORIGINS	Extra addresses Learnarr is reached at, comma-separated, if sign-in fails with an origin error.
OPENSUBTITLES_API_KEY	An OpenSubtitles.com API key shared with everyone here who hasn't added their own.
UDPIPE_BIN	Where the lemmatiser binary is. Default /opt/udpipe/udpipe in the image.
TZ	Your time zone, for "today" in study limits, streaks and reminders. Default UTC.
NTFY_URL, NTFY_TOKEN	A self-hosted ntfy server (default https://ntfy.sh) and its access token.
REMINDER_HOUR	The hour (in TZ) practice reminders go out. Default 18.

以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗