Auto-tagging documents in Paperless-ngx with Ollama

What they built

Paperless-ngx 3.0 added optional LLM features: suggested title, sender, document type and keywords for new documents, plus a document chat.

Invoices, contracts and official notices contain personal data, so the article recommends running the model locally via Ollama so documents never leave the premises. The AI complements the rule-based classification; it doesn't replace review.

Setup

• Paperless-ngx 3.0 or newer
• Ollama on your own hardware (or a model service in Germany)

Gotchas

• Reliable for clear documents, less so for unusual correspondence, so keep a review inbox

“The AI complements the classic rule-based and learned classification, but does not replace review.”— bettersorted.de

Do it yourself

1. Run Ollama and pull a model
2. Enable the AI features via PAPERLESS_AI_* settings with the ollama backend
3. Keep your list of senders, document types and tags manageable
4. Review suggestions for the first weeks and always for tax documents
5. Use rules for clear cases and AI for unclear ones; measure how often suggestions are right

PAPERLESS_AI_ENABLED=true
PAPERLESS_AI_LLM_BACKEND=ollama
PAPERLESS_AI_LLM_MODEL=<your-ollama-model>
PAPERLESS_AI_LLM_ENDPOINT=http://ollama:11434