Repository navigation
Home
github-actions[bot] edited this page Jun 21, 2026
·
3 revisions
Paperless IQ adds LLM intelligence to your Paperless-NGX archive — self-hosted, privacy-first, and built for non-English collections.
It connects to your existing Paperless-NGX instance and layers on:
- AI metadata analysis — an LLM reads every document and suggests structured metadata (tags, correspondent, document type, storage path, custom fields)
- Approval workflow — every suggestion is reviewed in a queue before anything is written back to Paperless
- Conversational search — ask natural-language questions about your archive and get grounded, cited answers
- Long-term memory — key facts from past conversations are extracted and injected into future ones
- Library grooming — detect and merge duplicate tags/correspondents/types; scan for mis-tagged documents
No cloud dependency required. Runs fully air-gapped with Ollama and the local ChromaDB vector store.
| Page | What it covers |
|---|---|
| Getting-Started | Prerequisites, Docker installation, first-run checklist |
| Document-Analysis | How the analysis pipeline works, OCR vs. vision mode, the approval queue |
| Discovery | Conversational search, RAG pipeline, long-term memory |
| Automation | Inbox poller, scheduled batch runs, cron expressions |
| Grooming | Entity descriptions, duplicate detection, mismatch scanning |
| LLM-Providers | Setup guide for Ollama, Anthropic, OpenAI, and Amazon Bedrock |
| Vector-Stores | ChromaDB vs. Qdrant vs. Bedrock KB; migration between backends |
| Settings | Every configurable setting with caveats |
| Troubleshooting | Common problems and their fixes |
| Architecture | System overview, module responsibilities, data flows (developer reference) |
| Paperless-NGX v3 | Paperless IQ | |
|---|---|---|
| LLM metadata suggestions | Title, tags, correspondent, doc type | All of those + custom fields |
| LLM providers | OpenAI-compatible + Ollama | Ollama · Anthropic · OpenAI · Amazon Bedrock |
| Approval workflow | Inline suggestions | Full queue — editable, stackable, batch actions |
| Image-only PDFs | Azure cloud OCR | On-premise vision analysis |
| Document search | FAISS | ChromaDB · Qdrant hybrid search · Bedrock KB |
| Long-term memory | — | Facts extracted per session, deduplicated, reinjected |
| Re-ranking | — | LLM · local cross-encoder · Amazon Bedrock Rerank |
| Audit log | — | Field-level change history |
| Runs air-gapped | Yes | Yes (Ollama + local ChromaDB) |