Semlith launches tomorrow. Where do your coding agents struggle most when they search?

Hi everyone, I'm Aakash. Semlith goes live here tomorrow, Saturday 31 Oct, at 12:01am Pacific.

It's a free, open-source local index of your code and documents that coding agents like Claude Code, Cursor and Codex search over MCP instead of grepping. I started it because my agents kept quoting PDFs and Word files they couldn't actually open. In my test, 41 of 127 quotes weren't in the file the agent named. With Semlith it was none.

Before the launch I'd really like to hear how this goes for you. Where do your agents get it wrong most, in code, in docs, or when the answer is in another repo?

I'll be around all day tomorrow to answer questions.