Vat
My personal second brain, searches across pretty much everything I've written and read on my Google Drive, Apple Notes, ChatGPT exports, Gmail etc and returns high-quality ranked results.
What It Does
I indexed and imported all of my mail, ChatGPT history, all notes, documents and PDFs across my entire ecosystem and built a database that allows fast AI-style semantic queries using a local LLM or Claude.
How it works
It reads all my notes, docs, ChatGPT exports and mail into one SQLite file, splits everything into passages, and builds two indexes over them, a keyword one and an embeddings one.
When I ask something it runs both, merges the results, reranks the top handful, and hands only those passages to a model to write the answer with citations. Everything sits on my laptop; I open it in the browser and the model is either Ollama running locally or Claude, whichever I point it at.
Built with
- Python 3.12, Typer, SQLite (FTS5), sentence-transformers (bge-base + cross-encoder rerank), NumPy, FastAPI, and Ollama or the Claude API for answers.