Project

Second Brain

You never file a thing. It indexes your folders, email, calendar and browser history into one local file your AI assistants can search.

Source and install
The dashboard’s search page, run on sample notes made from this site’s pages with a word-count stand-in for the embedding model: the query “human in the loop” returns ranked passages, each with its file, chunk and score, led by the opening of the Ventura case study, then Clean-Data’s note on its locked send button
50+tools it offers AI assistants
27data connectors, from Gmail to browser history

What goes in, what comes out

Why I built it

I liked the idea of an Obsidian-style second brain and never kept one up: the filing is what falls away in a busy week. So I built one that files itself. My AI assistants search it, so they already have my context.

What I decided

  • It reads whatever lands in my folders. A background process pulls out the text, cuts it into chunks and embeds them for search.
  • One server, many assistants. It speaks MCP, the open protocol AI assistants use to reach outside tools, so any assistant that does, from Claude Desktop to Cursor, can use it. It also has a web dashboard of more than 30 pages.
  • Local first, with two calls out. The index is one SQLite file. The only calls out go to Anthropic, for chat, drafting and summaries, and to Voyage, for embeddings, each under a daily spending cap.
  • A name or half a sentence is enough. Keyword and vector results are merged and reranked, with a small boost for recent material.

Try a slice of it

Ask AI Ben, the chat on this site. It answers from notes I wrote by hand.