Definition
Low-latency knowledge graph querying is retrieval of meaning — entities, relationships and governed facts — fast enough for an interactive user or an agent loop, with source and permission attached.
The problem
A correct answer that arrives after the decision is an archive. Agents in particular cannot wait on a warehouse job. The graph has to be the query surface, not an export into another index.
What exists today
- Warehouse SQL, which is precise on tables and slow or brittle on meaning.
- Vector search, which is fast on similarity and weak on permission and identity.
- Heavyweight graph databases queried as analytics, not as a live semantic layer.
Deca10 research question
What query plan keeps entity lookup, path constraint and provenance payload inside an interactive bound for organizational questions?
Evidence
Latency targets will be published against a defined query set: entity lookup, one-hop governed path, and sourced metric questions. Benchmarks in progress.