An agent can produce fluent SQL and still give the wrong business answer when the same metric means different things across systems. The harder problem is not query generation; it is preserving trusted meaning from a question through to a result. This project addresses that gap by making enterprise definitions, relationships, and lineage available as shared context for agents.
What Sets It Apart
- It builds an ontology from enterprise metadata and can connect existing semantic models, so teams can improve agent grounding without discarding definitions they already maintain.
- Query validation against certified definitions and lineage shifts evaluation beyond whether SQL executes: answers can be checked against the business meaning they are supposed to represent.
- Provenance is returned with answers, making it easier to inspect where a result came from rather than treating the model's response as an unexplained conclusion.
- MCP, CLI, and API interfaces let multiple agent harnesses use the same context layer instead of each rebuilding its own interpretation of the data.
Who It’s For
Great fit for teams building agents over relational business data that need consistent metric definitions, traceable answers, and a shared semantic layer. Look elsewhere if your work is primarily unstructured-document retrieval or if you need a provider-neutral setup without configuring model endpoints: the default inference path uses NVIDIA NIM, and the project is oriented around connected databases. Also account for deployment security—UI-added database credentials are stored unencrypted in Postgres unless Vault is configured.
