Heivol Codegraph
If you build regulated software, you already write down which decisions and controls govern your code. Codegraph connects those documents to the code itself, so "show me what backs this claim, and what changed since we filed it" is a query instead of a week.
What we're building
From the filing back to the code
Link a graph to the knowledge base holding your decision records and evidence documents. Then ask which code implements a given claim, and whether any of it has moved since the document was last updated. The staleness check is a comparison, not a judgement call, so it gives the same answer every time you run it.
Every answer says where it came from
Facts the type checker proved, patterns matched literally, and heuristic guesses are three separate things, and Codegraph keeps them separate. A blast radius returns what is certain as the answer and what is merely possible in its own bucket. Nothing quietly blends the two, because a graph you cannot trust to say "nothing else calls this" is not worth querying.
Your source can stay on your side
We are designing for a mode where your own build pipeline does the indexing and sends us only the resulting index. No clone, no copy of your repository, nothing to leak. Hosted locally in your own country either way, on Heivol's own hardware, with no hyperscaler in the path.
Built for estates, not single repos
Group many repositories into one workspace and query them as a whole. Which consumers are pinned to an old version of a shared package, how far two forked services have drifted apart, and what a change to one export actually reaches. Your agents and your people ask the same graph.