Paperless-GPT
A self-hosted AI add-on for paperless-ngx that extracts text and organizes documents using local Ollama models or cloud APIs. Open source under MIT.
Paperless-GPT adds AI text extraction and document organization to an existing paperless-ngx library. It runs in Docker on your own server and suits people who want less manual sorting of scanned paperwork. The project is open source under the MIT license.
It can generate document titles, tags, creation dates and correspondents, plus extract information into custom fields. Its web interface lets you review and edit suggestions before applying them, or let automatic processing handle documents. You can also adjust prompts and ask for summaries or specific information across a selection of documents.
For local LLM processing, it connects to Ollama, with support for reasoning models such as qwen3:8b and vision models such as minicpm-v. Using local backends keeps that processing on your hardware. Cloud alternatives include OpenAI, Anthropic Claude and Mistral, which receive the document text or images needed for the selected task and require API credentials.
OCR can use vision models or dedicated services, including self-hosted Docling Server, Google Document AI, Azure Document Intelligence and Mistral OCR. With Google Document AI, it can produce PDFs whose text is searchable and selectable while retaining the scanned page's appearance. It can save those files locally or upload them to paperless-ngx.
The web interface has no built-in authentication; access protection needs to come from your hosting setup.
