Structure-aware code retrieval for Onyx (open to collaborating)

[rs545837]

Hi Onyx team,

I maintain sem, an open-source tool that builds a deterministic, entity-level graph of a codebase (functions, classes, methods, and the call and import edges between them) and answers structural questions from it. Things like what calls this, what breaks if I change it, semantic diffs between commits, and who last touched a function.

I have been following Onyx as a way to search and answer over connected sources, and code tends to be one of the harder ones because flat text chunks lose the structure. sem is built for exactly that layer. Instead of returning code snippets, it can return a function plus its callers and callees, or tell you that changing something affects twelve other things, deterministically and without hallucinated edges.

Before assuming anything, I wanted to ask whether structure-aware code retrieval is something Onyx would find useful, and if so, what the right integration shape would be on your end. sem runs as both a CLI and an MCP server, so it could plug in as a connector, a retriever, or a tool, whatever fits your architecture best. Happy to prototype whichever direction you prefer.

Thanks for building Onyx in the open.