Paperless NGX 3.0 vs Paperless GPT: AI classification

In this German-language review, Paperless NGX 3.0 adds native AI suggestions for document metadata. Daniel Klozbücher compares it with Paperless GPT after two weeks of testing, using the same language model for both. His preference concerns workflow controls rather than the quality of the model's suggestions.

The native process starts with a button in the document detail view. It sends the OCR text to the configured backend, then returns suggestions that the user reviews and saves. The backend can use local AI through Ollama or public models from OpenAI. In the version reviewed, Klozbücher says workflows cannot trigger native classification automatically, users cannot edit its system prompts, and it lacks AI support for custom fields.

Paperless GPT can process documents automatically when they carry a chosen tag. Paperless workflows can assign that tag, connecting document ingestion to classification. The comparison also covers editable prompts for individual attributes and instructions for filling or creating custom fields.

Klozbücher recommends keeping the native RAG chat alongside Paperless GPT in a self-hosted setup. The chat searches documents and answers questions using a separate vector database, even without AI-generated attributes. It requires an additional embedding backend. He reports better chat results than with Paperless AI using the same language model; that is his test result, not a universal benchmark.