AI-powered document processing for Paperless-ngx. Drop a tag on a document — the LLM handles the rest.
Automatically assigns correspondents, document types, and tags using your LLM — no manual sorting.
Correspondent and document type are picked from your Paperless lists with a probability for every answer. Below your threshold the field stays as it is and the document gets a review tag instead.
Send document pages to a vision model for high-quality OCR on scanned PDFs, with optional text correction.
Define extraction prompts to pull structured data — invoice totals, dates, reference numbers — straight into Paperless custom fields.
Supports Ollama (local), OpenAI, OpenRouter, LM Studio and Grok. Mix providers — a local model for classification, a vision model for OCR, a small model for decisions.
Start, stop and check processing from cron, Home Assistant or your own scripts, and let Claude Desktop or any MCP client drive it through the built-in MCP server. One token, shown once.
The web interface ships with English and German translations. More languages can be added by editing a single JSON file.
docker run -d \
-p 8000:80 \
-v paperless-aissist-data:/app/data \
nyxtronlab/paperless-aissistThen open http://localhost:8000 and connect your Paperless-ngx instance.