Code intelligence over MCP
One graph query
instead of dozens of reads.
An agent exploring a codebase file by file spends most of its context on navigation. Indexing it once into a knowledge graph turns that into a structural query.
Published numbers, not mine. Every figure on this page is the project's own reported measurement, with the arXiv preprint linked so a reader can check it.
Five structural queries
Project-reported- Indexing
- Linux kernel in 3 minutesReported at 28M lines across 75K files.
- Evaluation
- 31 real-world repositoriesDescribed in the linked preprint.
Figures above are quoted from the project README and its preprint, not measured here.
The project also states plainly that it reads your codebase and writes to your agent configuration files.