What it is
Privacy-first personal AI knowledge assistant that runs entirely locally via Ollama. Ingest anything (YouTube videos, GitHub repos, Obsidian notes) and query it through a local LLM with RAG. No data leaves your machine.
Stack
Ollama (llama3.1:8b) · pgvector (PostgreSQL) · nomic-embed-text · Obsidian PARA vault · CLI + voice interface · Groq (optional fallback)
Status
Phase 1 complete — core RAG (ingest, chunk, embed, query). Phase 2 in progress — GitHub/Obsidian connectors + voice. Phase 3 planned: research agent, SearXNG, ntfy notifications. Phase 4: hybrid search, reranking, web UI.
Looking for
Beta testers who want a local-only second brain. Contributors interested in the voice assistant or research agent phases.
Notes
This project is also the backbone for the Vault mind map — the LLM will write markdown files directly into the Obsidian vault, auto-updating the public graph on push.