Kivgraph

I kept watching my AI coding agent search through the same repository to answer structural questions like “where is this used?” and “what breaks if I change it?”

So I built Kivgraph: a local semantic code graph for AI coding agents. It maps symbols, dependencies, and cross-repository relationships instead of making the agent reconstruct them from repeated file reads.

The benchmark that convinced me to keep working on it covered 37 repositories and 29 questions. Kivgraph and grep + reading both reached 28/29 exact answers, but the graph used 36k tokens versus 268k—about 7.4× less context. Grep was still cheaper for simple searches, so this is a complement, not a replacement.

It is Apache-2.0, runs locally, needs no API key, and supports Go, TypeScript, Rust, Python, and Dart.

GitHub: https://github.com/Luqueee/kivgraph

Website: https://kivgraph.dev

What codebase questions make your agent burn the most context?