An AI clerk for Paperless-ngx that may clock out for good

What it does

Paperless-AI hooks into Paperless-ngx and passes new documents through an LLM—local Ollama models, OpenAI, Gemini, DeepSeek, or any OpenAI-compatible endpoint—to automatically assign titles, tags, types, and correspondents. It also layers a RAG-based chat interface on top of your archive, so you can query documents in plain English instead of digging through folders.

The interesting bit

The project builds a semantic search index over your personal paperwork, turning “What was my last electricity bill?” into a database query rather than a memory exercise. The author admits the repo is currently unmaintained and may not survive a rewrite, partly because Paperless-ngx itself is shipping official AI integration soon.

Key highlights

• Backend agnostic: supports Ollama (Mistral, Llama, Phi, Gemma), OpenAI, DeepSeek, OpenRouter, Perplexity, Together, LiteLLM, VLLM, FastChat, Gemini, and others.
• RAG chat interface for natural-language questions across your entire document history.
• Rule engine to auto-filter which documents get processed and apply tags without human review.
• Optional /manual web mode for reviewing sensitive documents before tagging.
• Dockerized deployment with health monitoring and persistent volumes.

Caveats

• The project is currently unmaintained; the author is unsure whether a promised rewrite will ever ship, especially with upstream Paperless-ngx adding native AI features.

Verdict

A pragmatic stopgap if you want LLM-powered document triage in Paperless-ngx today. Avoid if you need guaranteed long-term support or prefer to wait for first-party integration.