Lesson 2 of 10 · Foundations
Catalog, semantic layer or AI context layer: which does what
Short answer
Catalogs inventory data for people, semantic layers define metrics once, and AI context layers assemble lineage, definitions and rules for agents at query time. Many products now span more than one.
By The Lineage Desk · Best Data Catalogs · Published
Desk research from public vendor pages. Editorial assessment.
3 min read
Three product categories now claim to get data ready for AI agents, and their marketing overlaps. It helps to separate what each was built to do before comparing products.
What was a data catalog built for?
A data catalog is an inventory. It crawls warehouses and tools, lists tables, dashboards and owners, and lets people search and document them. Alation says it pioneered this market. Collibra built its business on governance around the catalog. Catalogs were designed for humans who browse, and much of their value came from people writing descriptions and certifying assets.
Source: Alation homepage
Source: Collibra homepage
What does a semantic layer add?
A semantic layer defines metrics, dimensions and joins once, so every tool computes revenue the same way. Cube describes itself as an agentic analytics platform built on a semantic layer; AtScale calls itself a universal semantic layer for AI and analytics; Honeydew is a semantic layer for AI and BI. For agents, the attraction is clear: ask for a certified metric by name and get a consistent answer. AtScale's MCP server does exactly that. The cost is that someone must build and maintain the model, and logic that lives outside it stays invisible.
Source: Cube homepage
Source: AtScale homepage
Source: Honeydew homepage
What is an AI context layer?
An AI context layer assembles what an agent needs for a specific task: lineage, definitions, trust signals and access rules, delivered at query time, usually over MCP. Atlan now calls itself the context layer for AI; DataHub calls DataHub Cloud an enterprise context platform; Euno calls itself an AI context platform and computes its context graph ahead of time from APIs and query logs.
Source: Atlan homepage
Source: DataHub Cloud context platform
Why do the categories blur?
Each category is moving toward the others. Catalogs add MCP servers and semantic features: Alation now ingests and governs Snowflake and Databricks semantic models. Context platforms generate metrics and semantic views: Atlan lists a Metrics Generator and Semantic Views. Semantic layers add chat and agent interfaces. Snowflake and Databricks fold catalogs and semantic layers into their platforms.
Source: Alation press release: AIOS expansion with six products
Source: Atlan homepage
How should you choose between them?
Start from the gap that hurts most. If agents pick the wrong tables, you need discovery and lineage. If they find the right table and compute the metric differently from the dashboard, you need definitions. If they find sensitive or uncertified data, you need governance that travels with the context. Most enterprises have all three gaps, which is why the ranking on this site weighs them together and why the weight-your-own tool lets you change the balance.
Whichever category you start from, ask the same question of every product: what exactly will my agent receive when it asks about this table, and who keeps that answer current?
The next lesson looks at MCP, the delivery mechanism most of these products now share.
Key points
- Catalogs inventory for people; semantic layers define metrics; context layers assemble context for agents.
- Every category is borrowing features from the others.
- Choose from the gap that hurts most: discovery, definitions or governance.
Next lesson
3. How MCP delivers context to agents