Search your own files. Inspect the sources.
PersonalRAGVault is a local-first document retrieval project for contracts, manuals, notes, and code stored on hardware the operator controls. It can ingest supported files, create local embeddings, retrieve relevant passages, and return source references. Optional local-model answers are available through Ollama. No NextEleven-hosted cloud account is required for the standard local workflow.
Why it exists
Most “AI over your docs” products want an upload. Operators with real contracts and internal notes need the opposite.
Control the storage path
A hosted workflow sends files or extracted content to another system. A local workflow can avoid that upload step when external services are not enabled.
Inspect the sources
Source references make an answer easier to verify against the underlying document. Citations do not guarantee that retrieval or generated text is correct.
Docker + GPU just to try it
PersonalRAGVault is a small Python CLI: venv, ingest, query. Optional UI and OCR extras when you want them.
What it does
Ingest → embed → retrieve → ask. Built for a personal or team vault on hardware you control.
Ingest
- PDF, Markdown, text, JSON, DOCX, code, calendars, bookmarks
- WhatsApp chats, Notion exports, optional OCR for images
- Dedup on re-ingest · skip noise dirs (.git, .obsidian, node_modules)
Query
- Hybrid vector + keyword (FTS5) with RRF
- Presets: fast / balanced / quality
- Streaming answers + automatic citation display
Operate
- doctor and config edit for local health
- Folder watch, purge, reindex, compact
- Optional Streamlit UI for multi-turn chat
Integrate
- Local FastAPI: /query, /ingest, streaming SSE
- Optional bearer token for agent/tool access
- Graph neighbors for linked notes
What's ready, and what isn't
Open source under MIT and designed for personal vaults and private operator-managed installations. It is not a NextEleven-hosted multi-tenant service.
What you get today
- Documented local ingestion and retrieval paths for supported macOS and Linux environments
- CPU embeddings + Chroma storage
- Retrieval-only mode or local LLM answers
- CLI, optional UI, optional local API
- Public repo with CI and install scripts
What we won't pretend
- A NextEleven-hosted cloud vault (you run it)
- Enterprise SSO/SCIM out of the box
- Guaranteed accuracy without good source docs
- A replacement for your DMS or dealership desk products
The repository demonstrates retrieval, API, evaluation, testing, and documentation work. Hardening for a specific organization would require separate requirements and verification.
Get started
Clone, set up, inspect, ingest, and query using the documented local workflow.