otzar
I kept forgetting what I'd already written, so I built something to search my own notes.
I have a habit of writing things down and then forgetting I wrote them. Years of markdown, org files, PDFs, papers, source code. Keeping all of it was never the problem — finding it again was. grep needs the exact word I used, but my memory works in terms of concepts.
So I built otzar, which is Hebrew for treasure, and also for thesaurus. It searches my files by word and by meaning at the same time, and tells me where it found things: this line, that page, that paragraph.
Everything runs on my laptop. A Go program you type at, a Clojure server doing the actual work with Lucene, Ollama for the embeddings. No cloud, no database to maintain, no API key — my journal shouldn’t need a privacy policy.
The same program is also an MCP server, so Claude searches the exact index I do. Read-only, deliberately: I want it finding things in my notes, not editing them.
None of that is where the fun was, though. Building a personal tool is its own pleasure, closer to knitting yourself a glove: it only has to fit one hand, and you’re the one wearing it.
I’ve been using it every day for a month. The surprise wasn’t how well it retrieves. It’s how often I go looking for something and find I’d already tackled a similar issue and what my insights were.