Lector
Show HN: Evaluating Local LLMs as language translators for my app
Last verified:
What is Lector?
Lector is a self-hosted language learning reader that combines reading, vocabulary mining, cloze practice, AI-assisted translation, writing correction, and an LLM tutor in one app. It is designed for people who want to learn languages by reading real content while keeping their data on hardware they control.
The reading side supports EPUB files, web articles, and pasted text. While reading, you can click words for translations, track word states like new, learning, and known, and see familiar vocabulary highlighted in context. The app uses a built-in dictionary for common words and can fall back to an AI model for uncommon words or phrases.
Lector also includes cloze practice built from sentence pairs, with multiple-choice and typing modes plus spaced repetition scheduling. Its current language-pack documentation highlights Afrikaans to English support, with Spanish to English and German to English marked as coming soon. The sentence bank is sourced from Tatoeba and ordered by word frequency so common vocabulary comes first.
Beyond reading and drills, Lector adds AI-powered writing feedback, an always-available tutor chat, statistics like vocabulary growth and streaks, optional text-to-speech, and direct AnkiConnect integration for sending cards to Anki Desktop. It is aimed at self-hosters, language learners who like active reading, and people who prefer local control, offline use, and no subscription lock-in.
Lector pricing
Pricing model: Freemium
Lector is described as free with no monthly subscription, no premium tiers, and no feature gates. The website says the app itself is free to self-host, while optional AI translation can use Anthropic API key billing on a pay-as-you-go basis or an Anthropic Pro/Team OAuth token tied to an existing Claude subscription. Optional Google Cloud Text-to-Speech also implies separate usage-based cost if enabled.
Lector pros
- Self-hosted on your own hardware
- No cloud account required
- No telemetry or analytics
- Offline reading works without internet
- EPUB import supported
- Web article import supported
- Raw text paste supported
- Click-to-translate while reading
- Tracks word states
- Highlights known vocabulary in text
- Built-in dictionary for common words
- LLM translations for uncommon phrases
- Cloze practice included
- Multiple-choice cloze mode
- Typing cloze mode
- Spaced repetition scheduling
- Direct AnkiConnect integration
- Writing journal with grammar correction
- LLM tutor chat for questions
- Progress stats and heatmaps
Lector cons
- Requires self-hosting setup
- Docker knowledge helpful
- Best experience depends on optional AI provider
- AI translation needs internet when using cloud models
- Local model quality varies by model choice
- Only Afrikaans-English pack is currently available
- Spanish and German packs are still coming soon
- AnkiConnect requires Anki Desktop on the same machine as the browser
- Tailscale or reverse proxy setup may be needed for remote access
- No managed hosted service is described
- Paid Anthropic usage can add ongoing AI costs
Frequently asked questions about Lector
What is Lector used for?
Lector is used for language learning through reading, vocabulary mining, cloze practice, writing feedback, and AI tutoring in a single self-hosted app. It is meant to help you read real material, learn words in context, drill sentences with spaced repetition, and review your progress over time.
What content can I import into Lector?
You can import EPUB files directly, paste web article URLs so Lector can extract the text with Readability, or paste raw text for quick reading practice. That makes it flexible for books, articles, and short passages.
How does translation work while reading?
You can click a word to see its translation. Lector uses a built-in dictionary for common vocabulary and can use an AI provider for context-aware translations of uncommon words and phrases when configured.
Does Lector work offline?
Yes, the reader, cloze practice, and vocabulary tracking work without internet access. The main feature that needs a connection is AI translation when you use a cloud provider.
How is cloze practice organized?
Lector ships with about 2900 Afrikaans-English sentence pairs from Tatoeba. The practice system supports multiple-choice and typing modes, and it schedules reviews with spaced repetition across five mastery levels.
Can I use my own AI model?
Yes, Lector supports local or self-hosted OpenAI-compatible servers such as Ollama, LM Studio, vLLM, or another compatible endpoint. You can set an endpoint, an optional API key, and a model name in the AI settings.
Can Lector send cards to Anki?
Yes, Lector connects directly to AnkiConnect on your local machine so you can push vocabulary cards into Anki Desktop. The setup requires the AnkiConnect add-on and a matching CORS allowlist entry for your Lector origin.
How is Lector deployed?
Lector is distributed as a Docker image and the recommended deployment method is Docker Compose. The app stores its data in a SQLite database inside a bind-mounted data directory so updates and backups are simple.
How do I access Lector remotely?
The docs recommend using a WireGuard-style setup with Tailscale so you can access Lector from your phone or work device without exposing public ports. If you want HTTPS inside your network, the docs suggest putting it behind a reverse proxy such as Caddy.
Which language packs are available?
The website lists Afrikaans to English as available, with about 2900 sentence pairs, a dictionary, and frequency data. Spanish to English and German to English are marked as coming soon, and the reader itself can work with any language for reading.