What is a context layer for AI?

Summary

• A context layer gives AI agents the business meaning of data, not just access to it: definitions, relationships, calculations, authoritative sources, expertise and permissions. Databricks puts it this way: "A schema describes how data is structured. An ontology describes what that data means in the context of the business."
• Databricks' context layer is Genie Ontology, which its docs call "the unified context layer that gives the Genie family of AI experiences a business-aware map of your organization." The docs label it Public Preview.
• Genie Ontology combines Unity Catalog semantics that you define, govern and certify (metric views, domains and Pages) with inferred context: snippets that Genie extracts from your existing assets and usage.
• Permissions follow the user. Snippets are gated by Unity Catalog permissions, so Genie "only uses snippets extracted from assets you have permission to see."
• You don't have to model everything first. Databricks recommends modeling the critical "head" and letting Genie Ontology infer the "tail," and it cautions that an ontology "by itself doesn't magically make an agent accurate."

What is a context layer for AI?

A context layer is the part of an AI architecture that tells agents what data means and which sources to trust. Databricks' ontology post says an ontology "connects technical assets such as tables, metrics and queries to business concepts: definitions, relationships, calculations, sources of expertise and rules about how those concepts should be interpreted." Databricks' product for this is Genie Ontology, which is in Public Preview.

Why do agents need more than a schema?

A human analyst fills the gaps. The post says an analyst "knows which of five revenue tables Finance trusts, what 'active customer' means this quarter, or why one calculation should be used instead of another." An agent has no knowledgeable person between the data and the decision, so "the agent has to discover the meaning itself." Giving it access to more data "doesn't solve this." It needs "definitions, relationships, calculations, authoritative sources, expertise and permissions." When the model doesn't know which internal source holds the ground truth, the post says, it "fills the gap with inference."

What does a context layer hold?

What it holds	Example from the Databricks docs
Metric definitions	"An 'active user' is a distinct user, deduplicated across all platforms."
Authoritative sources	"Revenue questions should be answered using the curated Finance Genie Agent."
Business rules	"A 'qualified lead' only counts once a demo is booked."
Permissions	Snippets are gated by Unity Catalog permissions, so Genie only uses snippets extracted from assets you have permission to see.

The first three rows are the docs' own examples of inferred snippets.

How does Genie Ontology work?

The docs say Genie One and Genie Code search the same ontology, "so context you curate once applies to both surfaces." The launch post says Genie One and Genie Agents are "powered by Genie Ontology." The ontology brings together two categories of context:

• Unity Catalog semantics: "Modeled context that you define, govern, and certify through features like metric views, domains, and Pages."
• Inferred context: "A map of snippets that Genie automatically extracts and maintains from your existing assets and usage, including metric views, dashboards, SQL queries, and Genie Agents."

Each snippet "gets an authority score based on where it was generated from, how often it is used, and how fresh it is." When you ask a question, Genie "ranks the most relevant snippets, resolves conflicts, and answers using only the sources you are permitted to see." To see which knowledge sources Genie used, click the citation icons in a response.

How do you build one without modeling everything?

Databricks' advice is to "model the head and learn the tail." People should "explicitly define and govern the small set of concepts that cannot be wrong, things like revenue, compliance rules and core KPIs," while the wider ontology learns the rest from how the organization works. The operationalizing post suggests a layered path:

• Add table descriptions and column comments in Unity Catalog, and apply tags to classify and organize.
• Define measures in metric views, and capture shared business terms in Pages.
• Certify the assets you trust.
• Start with one high-value domain, then expand.
• Test with a representative set of business questions for each priority domain, and trace any failure back to the source asset.

FAQs

Does a context layer guarantee accurate answers?

No. Databricks says "an ontology by itself doesn't magically make an agent accurate." What improves accuracy is "supplying the agent with the right, authoritative context at the point where it's reasoning."

How is a context layer different from a semantic layer or knowledge graph?

Databricks says semantic layers and knowledge graphs "have delivered real value, particularly for business-critical concepts." The difficulty comes when organizations "try to manually model the entire enterprise," because business knowledge "changes too quickly and is spread across too many places." Databricks' answer is to model the head and learn the tail.

Can agents outside Databricks use it?

Yes, through the Genie One MCP server. Per the docs, "an MCP client or agent sends a natural-language question, and Genie searches your enterprise data, writes SQL, and returns an answer grounded in Genie Ontology with deep links back to your cited Databricks sources." Unity Catalog permissions are enforced on every request. The docs mark the Genie One MCP server generally available, while Genie Ontology itself is Public Preview.

What lifecycle stage is Genie Ontology in?

Public Preview, according to the Databricks docs ("This feature is in Public Preview"). The June 16, 2026 launch post says Genie One, Genie Agents and Genie Code are generally available but gives no separate availability statement for Genie Ontology. Confirm current terms before you depend on it.