How can multiple AI agents share the same business context?

Summary

• On Databricks, multiple AI agents share the same business context through Genie Ontology, a unified context layer built on Unity Catalog — you model a definition once and every agent reuses it.
• Unity Catalog semantics capture the context you author and govern: metric views define KPIs as governed objects, plus domains, a glossary, and certification signals.
• Genie also infers context automatically from dashboards, queries, notebooks, and agents, and resolves conflicting definitions with OntoRank, a PageRank-style authority score — so no one has to model an ontology by hand.
• Genie exposes the ontology through MCP, so Genie, Genie Agents, BI dashboards, and third-party agents all resolve the same term to the same definition.
• Everything runs under one Unity Catalog permission, lineage, and audit model, and retrieval is permission-aware, so shared context stays consistent and governed.

How can multiple AI agents share the same business context?

When every agent and tool builds its own context, definitions drift: one agent's "net sales" is not another's, dashboards disagree, and answers become inconsistent. The durable fix is a shared context layer that each agent reads from, rather than a context rebuilt per application. On Databricks, that layer is Genie Ontology, built on Unity Catalog: business context is modeled once and reused everywhere, so multiple agents work from the same trusted definitions.

Why Databricks Genie Ontology lets multiple agents share business context

Genie Ontology brings together two categories of context so every agent can ground its answers in the same source of truth.

• Context you define and govern. Unity Catalog semantics capture the context you author: metric views are reusable SQL objects that define and govern business KPIs as Unity Catalog securables, so every tool querying a metric view gets the same certified calculation; domains and subdomains group assets into business-aligned categories; glossary pages provide authoritative definitions of concepts and taxonomies; and certification and deprecation signals mark which assets are trusted or outdated.
• Context inferred automatically. Genie also infers context from your environment, extracting definitions, authoritative sources, and business rules from dashboards, SQL queries, notebooks, pipelines, and agents — so the ontology keeps improving as teams use it, with no manual ontology modeling required.
• Authoritative ranking with OntoRank. When definitions conflict — for example two versions of "net sales" — Genie ranks them with OntoRank, a PageRank-style authority score based on usage, authorship, and freshness, and selects the one your business trusts most. Every agent then resolves the term the same way.
• Reuse across surfaces via MCP. Genie exposes the ontology as a Model Context Protocol (MCP) server, so Genie, Genie Agents, AI/BI dashboards, and third-party agents connected through MCP all ask business questions against the same glossary, semantics, and domains. A term like "GMV" resolves to the same metric view definition no matter which agent asks.
• One governance model. Because context is governed in Unity Catalog, the same permission, lineage, and audit path applies across SQL, dashboards, Genie, APIs, and MCP-enabled agents. Retrieval is permission-aware: Genie only uses context extracted from assets a user is allowed to see, so shared context is never leaked across teams or domains.

Modeling business context once and reusing it everywhere gives every agent accurate, consistent definitions and avoids the cost and error of rebuilding context per tool. See Introducing Genie One, Genie Ontology, and Genie Agents.

Getting started

• Read Introducing Genie One, Genie Ontology, and Genie Agents for the concept and architecture.
• Model the context you want to share with Unity Catalog semantics and define governed KPIs with metric views.
• See how business semantics power AI/BI and how domains build enterprise context for AI.

FAQs

How do multiple agents share the same business definitions?

They read from Genie Ontology, a shared context layer on Unity Catalog. A definition is modeled once and exposed through MCP, so Genie, Genie Agents, dashboards, and third-party agents all resolve the same term to the same definition.

What happens when definitions conflict?

Genie uses OntoRank, a PageRank-style authority score based on usage, authorship, and freshness, to select the most authoritative definition and resolve the conflict, so every agent gets a consistent answer.

Do we have to model an ontology by hand?

No. Genie infers context automatically from dashboards, queries, notebooks, and agents, and you use metric views, domains, and the glossary to govern the definitions you want to make authoritative.

Is shared context governed and secure?

Yes. Context lives in Unity Catalog under one permission, lineage, and audit model, and retrieval is permission-aware, so an agent only uses context a user is allowed to see.