paperless-gpt adds AI to Paperless-ngx in two ways: better OCR using vision-capable language models, and automatic titles, tags, and correspondents. Its distinguishing feature is a review interface, so you approve suggestions before they are written to your documents.

Features

• LLM-enhanced OCR: vision models read poor-quality scans, handwriting, and odd layouts better than traditional OCR, and can produce searchable PDFs with an accurate text layer
• Generated titles, tags, and correspondents
• A manual review screen for suggestions
• Custom prompt templates, editable in the web interface
• Ad-hoc analysis of selected documents
• Alternative OCR providers, including Google Document AI and Azure Document Intelligence

LLM providers

Ollama for local models, plus OpenAI, Anthropic, Mistral, and Google Gemini.

Deployment

Run it as a Docker container alongside Paperless-ngx in the same Compose file, with a Paperless API token and your provider settings. Tag documents in Paperless with a trigger tag to have paperless-gpt process them.

paperless-gpt has no built-in authentication, so do not expose it directly; put it behind a reverse proxy with authentication, such as Tinyauth, or keep it on your LAN.

Privacy

Cloud providers receive the full content of every document you process. For tax, medical, and financial papers, a local model via Ollama keeps everything at home; vision models need a capable GPU to run at a useful speed. See our Ollama guide.

paperless-gpt or Paperless-AI?

Paperless-AI adds document chat but is currently unmaintained. paperless-gpt focuses on OCR quality and reviewed metadata, and is actively developed.

License

MIT.

Links

• Repository