Loctran — private AI PDF translator

Translate PDFs locally. No cloud. No API key. Just Ollama.

Features

Screenshots

1. Home	1.1 PDF Upload

2. Translation Configured	2.1 Translation In Progress

3. Result	3.1 Translation Complete

30-second install

The default install includes the Web UI and all dependencies (OCR, OpenCV, Ollama client). A plain pip install loctran is enough to start the app.

pip install loctran
ollama pull glm-ocr
ollama pull translategemma:4b
loctran
# opens Web UI at http://127.0.0.1:8000

# CLI translation example
loctran translate document.pdf --lang French

How it works

PDF
 └─► rasterise pages (pypdfium2)
      └─► dual-pass OCR (Tesseract normal + inverted)
           └─► deduplicate & group words into segments
                └─► batch translate (Ollama LLM)
                     └─► HTML overlay output

Each page becomes an image with absolutely-positioned translation boxes sized to match the original text bounding boxes. For PDFs with a digital text layer, pdfplumber extracts text directly — no OCR needed.

Requirements

• OS: macOS, Linux, or Windows
• Python ≥ 3.9
• Ollama running locally — download
• Tesseract — brew install tesseract tesseract-lang (macOS) or apt install tesseract-ocr tesseract-ocr-all (Linux)

On startup, Loctran will try to start Ollama if it is installed and will pull the configured OCR and translation models when they are missing. The first launch still depends on the user having Ollama available and network access for any model downloads.

Run loctran doctor to check everything at once:

loctran-doctor v0.1.5
─────────────────────────────────────
✓  Python         3.11.9
✓  Tesseract      5.3.4  (langs: eng fra deu jpn +47)
✓  Ollama         0.3.1  (running)
✓  glm-ocr        pulled (2.2 GB)
✓  translategemma:4b pulled (3.3 GB)
─────────────────────────────────────
All required dependencies satisfied.

Web UI

Start the server and open your browser:

loctran serve
# → http://localhost:8000

Upload a PDF (up to 50 MB), choose source and target languages, pick an OCR and translation model, then watch the real-time progress bar. Cancel any running job with the cancel button. The translated HTML opens automatically when done.

CLI reference

Usage: loctran [OPTIONS] COMMAND [ARGS]...

Commands:
  serve      Run the local web UI server.
  translate  Translate a file or folder using local OCR + Ollama.
  doctor     Run environment diagnostics for dependencies and models.

# Translate to Spanish using a higher-quality translation model
loctran translate report.pdf --lang Spanish --model translategemma:12b

# Extract text only, save to custom folder
loctran translate scan.pdf --extract-only --output ~/Desktop/extracted

# Use smaller batches to avoid context overflow on long documents
loctran translate book.pdf --lang German --batch-size 3

# Run dependency diagnostics
loctran doctor

FAQ

Does this send my documents anywhere? No. Everything runs locally on your machine. Loctran talks only to Ollama at localhost:11434. No telemetry, no analytics, no cloud.

Which Ollama models work? Any locally installed Ollama model appears in the Loctran model picker automatically. Run ollama list to see what is available. For this project, use glm-ocr for OCR and translategemma:4b for translation. On 16 GB+ machines, translategemma:12b is the higher-quality option.

What about scanned PDFs? Loctran automatically detects whether a PDF has a digital text layer. If it does, pdfplumber extracts text directly (fast, accurate). If not — or if you pass --force-ocr — Tesseract runs a dual-pass OCR (normal + inverted image) to catch light-on-dark text. Pass --use-ai-ocr to route OCR through an Ollama vision model for the highest accuracy on complex layouts.

Docker

Loctran needs a running Ollama instance. Inside a container, localhost doesn't reach the host, so pass OLLAMA_HOST:

docker build -t loctran .

docker run -p 8000:8000 \
  -e OLLAMA_HOST=http://host.docker.internal:11434 \
  -v ~/Documents:/docs \
  loctran

The container runs with --no-desktop --no-browser automatically. Mount a volume at /docs to access your files from the Web UI.

Contributing

See CONTRIBUTING.md for development setup, running tests, and submitting PRs.

License

This project is dual-licensed:

• Open source — GNU Affero General Public License v3.0 (AGPL-3.0). You may use, modify, and distribute Loctran under the AGPL, which requires that any modified version or networked service built on Loctran also be released under the AGPL with full source code.
• Commercial — A proprietary license is available for organisations that cannot comply with the AGPL's copyleft obligations. Contact anzal.ks@gmail.com for terms.

© 2026 Anzal K Shahul. All rights reserved.

The recommended AI models (TranslateGemma, GLM-OCR) carry their own licenses. See THIRD_PARTY_LICENSES.md for details.