Hybrid graph + vector + keyword retrieval with LLM reasoning on top. Simple lookups stay fast; multi-hop questions traverse, and every answer returns citations plus the graph path behind them.
One pipeline from question to cited, path-backed answer.
Vector search finds text that sounds similar; it cannot tell you why an answer is true. GraphRAG retrieves through the relationships themselves (this contract, owned by that entity, amended on this date) so the LLM reasons over connected facts. The result is an answer with citations and a replayable path, which is what separates an assistant you demo from one you deploy.
Vector similarity finds what sounds related; BM25 finds exact terms; traversal finds what is actually connected. AtherGraph runs all three in one query surface.
A router matches each question to the right strategy: no traversal tax on simple lookups, no shallow answer to a deep question.
Answers arrive with their sources and the chain of connections that produced them. An artifact you can inspect, not an assurance you have to trust.
Retrieval respects the asking user’s rights. What you cannot read, the AI cannot retrieve on your behalf.
How it's built, and the reproducible numbers behind it.
SDKs, query examples, and local setup: what's live and what's coming.
Five phases, honestly labelled: shipped, live, in development.
8 questions, free gap analysis against EU AI Act and SR 11-7.
Book a session and we’ll walk this use case end to end on real data: architecture, queries, and what a 90-day path looks like for your team.