I'm trying to build a simple, local, and semantic code + context search for coding agents like Claude Code.
I noticed that when trying to work with Claude Code I burned my token allocation really quickly and with parallel agents, no better. But when I looked at the other repos things felt really complex, for me at least.
I call it "contextd" and imagine it kind of like a context daemon you can always query in your repository for implementation plans, specs or code, etc. I always liked that git lived in your repo in the .git file, so I wanted this to work like that too but without 3rd party hosted databases or multi-repo pollution.
It's written in python at the moment, uses some local models and lanceDB but I might have it re-written in rust for speed :)
I noticed that when trying to work with Claude Code I burned my token allocation really quickly and with parallel agents, no better. But when I looked at the other repos things felt really complex, for me at least.
I call it "contextd" and imagine it kind of like a context daemon you can always query in your repository for implementation plans, specs or code, etc. I always liked that git lived in your repo in the .git file, so I wanted this to work like that too but without 3rd party hosted databases or multi-repo pollution.
It's written in python at the moment, uses some local models and lanceDB but I might have it re-written in rust for speed :)
https://github.com/aquaflamingo/ctxd