A privacy-first tool that lets immigrants photograph any official document — tax notice, school letter, lease, utility bill, benefits letter — and get a plain-language explanation in their native language, plus urgency detection and concrete next steps.
Everything runs locally. Nothing leaves your machine.
- Upload or photograph any official document
- Plain-language explanation in 11 languages
- Document type classification (Tax Notice, Eviction Warning, etc.)
- Urgency countdown — "You have 23 days to respond"
- Numbered next steps
- Key terms glossary with simple definitions
- Encrypted local history — past scans stored on-device only, encrypted with a key that never leaves your machine
Download and run the installer from https://ollama.ai
After installation, Ollama runs automatically in the background.
Open a terminal and run:
ollama pull gemma4:latest
This downloads ~3 GB. Only needed once.
# Clone / download this project, then:
pip install -r requirements.txtTerminal 1 — start Ollama (skip if it's already running):
ollama serveTerminal 2 — start the app:
uvicorn main:app --reloadOpen your browser at http://localhost:8000
- Click the upload area or tap "Tap to photograph or upload" on mobile
- Select your document image (or take a photo directly on a phone)
- Choose the language you want the explanation in
- Click Analyze Document
- Wait ~30–60 seconds for the local model to process
| Variable | Default | Description |
|---|---|---|
OLLAMA_MODEL |
gemma4:4b |
Ollama model name |
OLLAMA_URL |
http://localhost:11434/api/generate |
Ollama API endpoint |
Spanish · Chinese (Simplified) · Tagalog · Vietnamese · Arabic · Hindi · Korean · French · Portuguese · Russian · English
Past scans are saved automatically to .history/records/ as Fernet-encrypted files. The encryption key lives in .history/key.bin and never leaves your machine. Both paths are in .gitignore so they won't be accidentally committed.
To clear all history, delete the .history/ directory. A new key is generated automatically on the next run.
This tool uses Gemma 4 E4B running fully on-device via Ollama. Your document images and their contents are never uploaded to any server or third-party API. All processing — including history storage — happens locally on your machine.