I've been using Claude Code a lot, but finding past sessions is a nightmare. The built-in --resume flag just gives you a flat list. If I want to find a specific database refactoring chat from last week, I have to scroll manually and guess based on truncated titles.
I got tired of this, so I built a searchable TUI for it. You type what you're looking for, hit Enter, and it instantly drops you back into the terminal chat via claude --resume <id>.
I wanted the search to actually be good, so it doesn't just use grep. It's written in Rust and does local hybrid search -> BM25 via SQLite FTS5 for exact keyword matches, plus semantic search using an all-MiniLM-L6-v2 ONNX model to find conceptual matches. It merges them with Reciprocal Rank Fusion.
Everything runs 100% locally. No search data goes to the cloud, and the embedding model is small enough (~80MB) to run fast on any hardware.
It's completely open source. I'd love to hear what you think, especially from claude code power users.
SilverElfin•1h ago
What about being able to see everything it does on your behalf? Like to verify it’s not doing something unintended? Like not the chat sessions but the behind the scenes stuff. Is there a solution for that?
madzarm•1h ago
the tool is open source and completely local - you can go take a look at the actual code and see that no unintended things are happening behind the scenes :). It just parses your local claude code chat files, indexes them into a local SQLite DB and runs 'claude --resume <id>' with the id it parsed from the history.
madzarm•1h ago
I got tired of this, so I built a searchable TUI for it. You type what you're looking for, hit Enter, and it instantly drops you back into the terminal chat via claude --resume <id>.
I wanted the search to actually be good, so it doesn't just use grep. It's written in Rust and does local hybrid search -> BM25 via SQLite FTS5 for exact keyword matches, plus semantic search using an all-MiniLM-L6-v2 ONNX model to find conceptual matches. It merges them with Reciprocal Rank Fusion.
Everything runs 100% locally. No search data goes to the cloud, and the embedding model is small enough (~80MB) to run fast on any hardware.
It's completely open source. I'd love to hear what you think, especially from claude code power users.