AI · Web App
Paperless-AI
Ask your documents anything — AI tagging and semantic search for Paperless-ngx.
Paperless-AI makes a document archive answer questions. It is an AI layer for Paperless-ngx that classifies, tags and — crucially — lets you ask your documents in plain language, turning a passive filing cabinet into something you can actually query.
The challenge
Self-hosted document management solves storage, but not findability. Once an archive grows to thousands of scanned invoices, contracts and letters, manual tagging stops scaling and keyword search misses anything not phrased exactly right. The real questions people have are semantic — “When did I sign my rental agreement?”, “What was the last electricity bill?” — and a plain full-text index simply cannot answer them.
The approach
I built an AI extension around Paperless-ngx that automates classification and tagging and adds retrieval-augmented generation over the whole archive. Documents are embedded and indexed so a natural-language question retrieves the right passages, which a language model then answers with grounded, cited context. It is deliberately deployment-flexible:
- Works with OpenAI-compatible APIs or fully local models via Ollama, so privacy-sensitive users keep everything on-premise.
- Automatic classification and smart tagging drive hands-off document workflows.
- An intuitive web interface for search, chat and configuration.
The outcome
The archive becomes conversational: precise, natural-language answers across the full corpus, and tagging that maintains itself. It is one of my most-used open-source projects, precisely because it removes the friction between having documents and getting answers from them.