Unity Catalog governs the Databricks platform. Google’s Knowledge Catalog, formerly Dataplex Universal Catalog, catalogs a Google Cloud estate. Atlan’s Context Layer for AI crawls both through its Databricks and Knowledge Catalog connectors, so an agent whose question spans Databricks and BigQuery can draw on both in one place. GA and Preview labels follow each vendor’s own release notes.
Google’s docs home reads “formerly Dataplex Universal Catalog,” so both names appear here. On April 10, 2026 Google renamed the product, and it kept the API, IAM, and CLI names. Each catalog describes its own platform well, and neither is better than the other in the abstract. The useful comparison starts from what each one covers, then moves to lineage, access control, open-format reach, and what an agent can call.
| Dimension | Unity Catalog (Databricks) | Knowledge Catalog, formerly Dataplex (Google Cloud) |
|---|---|---|
| What it is | Unified governance layer for data and AI built into Databricks | AI-powered data catalog, formerly Dataplex Universal Catalog |
| What it governs | Securable objects: tables, views, volumes, functions, models, MCP services; foreign catalogs through Lakehouse Federation | Entries and aspects across Google Cloud sources, plus third-party via connectors |
| Access to catalog objects | Privileges, attribute-based policies (GA), row and column filters, workspace bindings | IAM roles on projects, entry groups, entry types, aspect types, glossaries |
| Lineage | Automatic, column level, for queries run on Databricks | BigQuery, Vertex AI, cataloged entries; column level for BigQuery and Apache Spark jobs |
| Open-format reach | Iceberg REST endpoint; also available as open source | Iceberg REST cataloging GA; federation of Databricks Unity in Preview |
| Agent interface | MCP services governed as securable objects | Catalog MCP servers, local and remote; lineage remote server Preview |
What is the difference between Unity Catalog and Dataplex?
Unity Catalog governs assets inside the Databricks platform, while Knowledge Catalog ingests metadata from Google Cloud sources and takes third-party metadata through connectors or a managed connectivity pipeline. Google’s metadata overview lists BigQuery, Dataform, Vertex AI, Looker, Bigtable, Spanner, AlloyDB, Cloud SQL, Pub/Sub, and Cloud Storage as sources (Google Cloud docs, last updated 2026-10-27).
Names cause the first confusion. According to Google’s Knowledge Catalog release notes, Dataplex Universal Catalog was renamed on April 10, 2026. The Google Knowledge Catalog explainer covers the rename lineage, and Knowledge Catalog vs a neutral catalog for BigQuery covers where Google’s view ends.
For a data architect, the platform that holds your assets decides which catalog answers first, and the layer above both decides what an agent knows.
What is Unity Catalog?
Unity Catalog is the unified governance layer for data and AI built into Databricks. Its governed objects are securable objects, meaning tables, views, volumes, functions, models, and services, arranged in a three-level namespace of catalog.schema.object (Databricks docs, Unity Catalog overview, last updated Sep 11, 2026). An open-source implementation is also available.
Permissions start closed. Per the Databricks permissions model (last updated Oct 15, 2026), users initially have no access to data in a metastore, every securable has an owner with all privileges and the right to grant, and the MANAGE privilege lets others manage grants.
Attribute-based access control (ABAC) became generally available on April 28, 2026 (Databricks release notes). It uses governed tags to target data, with row filters and column masks on tables (Databricks docs, ABAC, last updated Oct 22, 2026). Databricks labels three policy types Beta: metastore-level policies, policies on views, and DENY policies. For related features, see Unity Catalog business glossary and domains, and which features are GA or preview, and Unity Catalog Metrics.
On the agent side, Unity Catalog governs MCP services as securable objects, and Genie One MCP shows how agents reach Unity Catalog data. The Unity AI Gateway sits alongside it.
What is Dataplex, now called Knowledge Catalog?
Dataplex Universal Catalog is now called Knowledge Catalog, a rename Google announced on April 10, 2026. Google’s docs home opens by describing it as an AI-powered data catalog, and its core objects are entries, aspects, and business glossaries. An entry represents a single data asset, per Google’s metadata overview.
Access runs through IAM. The general Dataplex Administrator, Editor, and Viewer roles do not cover metadata resources such as entry groups and entries; separate Catalog Admin, Editor, and Viewer roles do, and viewing a data aspect also needs read access to the source data, per the Knowledge Catalog IAM roles page (last updated 2026-11-03).
Governance workflows route access requests to approvers, and approved requests receive IAM roles and group memberships automatically. Metadata-change requests for glossaries and entries are in private preview, and Google’s governance workflows page carries a Pre-GA warning. Google’s agentic data cloud comparison puts it next to a context layer, and the agentic data catalog and AI agents for data catalog pages cover how catalogs serve agents.
Unity Catalog vs Dataplex: head-to-head comparison
The two products diverge on scope and access model and converge on column-level lineage, open-format reach, and MCP access for agents.
| Dimension | Unity Catalog (Databricks) | Knowledge Catalog (Google Cloud) |
|---|---|---|
| Primary scope | Databricks platform assets, plus foreign catalogs through Lakehouse Federation | Google Cloud estate plus federated sources |
| Lineage capture | Automatic, column level, queries run on Databricks | BigQuery, Vertex AI, cataloged entries; column level for BigQuery and Apache Spark jobs |
| Lineage timing and retention | Catalog Explorer indefinite; system tables 395-day rolling window; data starts Sep 1, 2024 | Graph takes three hours for most jobs, up to 24 for some |
| Access control | Privileges, ABAC (GA; Beta on three policy types), row and column filters | IAM roles; separate Catalog roles for metadata resources |
| External engine access | Iceberg REST endpoint; managed Iceberg read and write; foreign Iceberg and Delta with Iceberg reads read only | Automated cataloging of Iceberg REST catalog for Lakehouse runtime catalog: GA (2026-04-16) |
| Federation | Lakehouse Federation: governed, read-only access through foreign catalogs; BigQuery a listed source | Federated Databricks Unity catalog: Preview |
| Boundary to plan for | Foreign Iceberg tables read through the endpoint need REFRESH FOREIGN TABLE to pick up the latest snapshot | Metadata permissions and data permissions are separate roles |
Take an estate with Delta tables in Databricks and sales data in BigQuery. Unity Catalog describes the first, and Knowledge Catalog describes the second. An agent asked for revenue by region needs both descriptions, plus the rule for which one wins when the definitions differ.
Lineage shows the scope split clearly. Databricks states that Unity Catalog captures lineage automatically for queries run on Databricks, down to the column level (Databricks docs, lineage in Unity Catalog, last updated Nov 4, 2026). Google’s lineage page (last updated 2026-10-26) covers BigQuery, Vertex AI, and cataloged entries. An agent question that crosses both platforms is a data lineage for AI question.
Access follows the same split. Per the Databricks Iceberg clients page (last updated Oct 27, 2026), external engines configure access through the endpoint /api/2.1/unity-catalog/iceberg-rest, and the cross-engine page (last updated Oct 20, 2026) says reads only are supported when fine-grained controls apply. Databricks release notes record cross-engine ABAC as generally available on October 1, 2026. Role-based access control in an AI context platform describes how role-based access applies in a context layer, and why governance rules should be readable by AI agents explains why the rule has to be machine-readable, not a spreadsheet.
Do Unity Catalog and Dataplex work together through Iceberg REST?
The two products meet at the Iceberg REST boundary: Unity Catalog exposes an Iceberg REST endpoint, and Google lists Databricks Unity as a federated Iceberg REST source in Preview. According to Google’s release notes (April 16, 2026), federated support for Databricks Unity and two other vendors’ catalogs is available in preview, and the metadata overview repeats the Preview label. As of November 5, 2026, no later release note moves federation to GA.
| Feature | Product | Status as recorded | Source and date |
|---|---|---|---|
| Federated Databricks Unity and two other vendors’ catalogs via Iceberg REST | Knowledge Catalog | Preview | Release notes, 2026-04-16 |
| Automated cataloging of the Iceberg REST catalog, Lakehouse runtime catalog | Knowledge Catalog | GA | Release notes, 2026-04-16 |
| Governance workflows (access requests) | Knowledge Catalog | Pre-GA | Docs page, last updated 2026-11-03 |
| Data lineage remote MCP server | Knowledge Catalog | Preview | Release notes, 2026-05-27 |
| ABAC | Unity Catalog | GA | Databricks release notes, 2026-04-28 |
| ABAC: metastore-level, on views, DENY policies | Unity Catalog | Beta | Databricks ABAC page, 2026-10-22 |
| Cross-engine ABAC | Unity Catalog | GA | Databricks release notes, 2026-10-01 |
Check the stage before you build on a Preview path. A federated Iceberg route suits prototyping, and a production dependency is easier to plan once the stage is settled and a fallback is in place.
Federation is one way to cross the boundary, and a layer above both catalogs is another. Can a data catalog sit on top of existing catalogs walks through that option. Single-stack lock-in vs a neutral context layer covers how a neutral context layer relates to a single platform’s catalog, and context layer vs data catalog vs semantic layer separates the three jobs.
When should you use Unity Catalog, Dataplex, or both?
The right choice follows the shape of your estate.
- Start with Unity Catalog when most governed data and models live in Databricks and workspaces share a metastore.
- Start with Knowledge Catalog when the estate is BigQuery, Vertex AI, Looker, Cloud Storage, and other Google Cloud sources.
- Use both when data lives on both platforms, and plan for the Iceberg REST boundary and the Preview status of federation.
An agent that crosses both estates draws on context from each catalog. The full-stack AI platform vs best-of-breed context layer trade-off applies directly, and is your data catalog keeping up with what AI agents need gives a way to test each catalog. For the Databricks side, Databricks Genie context requirements covers what Genie needs in order to answer well.
What to settle before comparing Unity Catalog and Dataplex
Three points keep the comparison fair to both.
- Confirm that the “formerly Dataplex” names refer to one product. The rename took effect on April 10, 2026, and the API, client library, CLI, and IAM names did not change.
- Confirm the federation stage. Google’s April 16, 2026 note labels it Preview.
- Name the scope first. A platform-wide governance layer and a catalog of one cloud’s sources each cover different ground.
The third matters most. Each team tends to start from the catalog nearest its own data, and enterprise AI context silos can follow when the scopes differ. The question of data catalog vs context layer is really a question about which scope each tool was built for.
How does Atlan work across Unity Catalog and Knowledge Catalog?
Atlan’s Context Layer for AI crawls both catalogs into one Enterprise Data Graph and gives agents one place to read from. It sits above each catalog and replaces neither. The context layer is the category Atlan defines, and the enterprise context layer is its architecture.
An agent whose question spans Databricks and BigQuery tables meets two descriptions of the estate, one from each catalog.
Atlan’s connectors cover both sides. According to the Atlan Databricks connector documentation, the connector crawls Unity Catalog schemas, tables, views, volumes, and AI models and builds column-level lineage automatically. The Knowledge Catalog connector crawls aspect metadata, data quality results, and data profiling results and attaches them to the matching BigQuery assets. Atlan’s docs describe connecting 80+ data sources, and those two feed the Enterprise Data Graph. How to implement an enterprise context layer for AI gives the sequence, and context engineering for AI agents explains how the context gets shaped for each question.
Agents reach that context through the Atlan MCP server, a hosted server where agents can search enterprise context, traverse lineage, and read governed definitions and glossaries, with every action respecting the permissions already set in Atlan. The MCP architecture deep dive covers the design, MCP for data lineage covers the lineage tools, and an MCP connected data catalog shows the pattern end to end. A semantic layer for AI agents then supplies the metric definitions. For the Databricks side specifically, Genie Ontology and the Atlan Context Layer shows how the two split the work.
To test either catalog against agent runtime needs, the catalog AI readiness check walks four dimensions: a verified agent interface, where enforcement stops, what the pricing meter counts, and whether a definition survives renames.
Start from scope, then check the status label
Pick the catalog that matches where your assets live, and plan a fallback for anything labeled Preview. Each catalog answers for its own platform, and an agent question that touches both draws on two sets of definitions. Context that sits above both is one way to reconcile them. Score your starting point with an AI readiness assessment before you build.
FAQs about Unity Catalog vs Dataplex
1. What is Dataplex called now?
Dataplex Universal Catalog is now called Knowledge Catalog. Google announced the rename in its release notes on April 10, 2026. The API, client library, CLI, and IAM names did not change, so existing code and permissions keep working under the Dataplex names. Google’s docs home describes it as an AI-powered data catalog.
2. What does Dataplex do?
Knowledge Catalog ingests metadata from Google Cloud sources such as BigQuery, Vertex AI, Looker, and Cloud Storage into entries and aspects. It adds business glossaries and lineage. Governance workflows route access requests to approvers, and Google labels that feature Pre-GA on its documentation page.
3. What is the difference between Unity Catalog and Dataplex?
Scope. Unity Catalog governs assets inside the Databricks platform, including tables, volumes, functions, models, and MCP services. Knowledge Catalog catalogs metadata across a Google Cloud estate and takes in third-party metadata through connectors. Access to catalog objects differs too: privileges and attribute-based policies in Unity Catalog, IAM roles on catalog resources in Knowledge Catalog.
4. Can Knowledge Catalog catalog Databricks Unity Catalog assets?
Through federation, in Preview. Google’s April 16, 2026 release note lists Databricks Unity as a federated Iceberg REST source, labeled Preview, and as of November 5, 2026 no later release note moves it to GA. Check the current stage before you build a production path on it.
5. Does Unity Catalog capture column-level lineage?
Yes, for queries run on Databricks. Databricks documents automatic capture down to the column level, aggregated across workspaces on one metastore. System tables keep a 395-day rolling window, and lineage data starts on September 1, 2024. Catalog Explorer retention is indefinite.
6. How can an AI agent use both catalogs together?
Each catalog offers its own agent access. Unity Catalog governs MCP services as securable objects, and Knowledge Catalog ships local and remote MCP servers, with the lineage remote server in Preview. Atlan’s Context Layer for AI crawls both catalogs into the Enterprise Data Graph, and agents read it through the Atlan MCP server.
Sources
- Knowledge Catalog release notes, Google Cloud
- About metadata in Knowledge Catalog, Google Cloud
- Knowledge Catalog IAM roles, Google Cloud
- Data lineage in Knowledge Catalog, Google Cloud
- Manage access requests with governance workflows, Google Cloud
- About MCP servers in Knowledge Catalog, Google Cloud
- Knowledge Catalog documentation home, Google Cloud
- Unity Catalog overview, Databricks documentation, last updated Sep 11, 2026
- Unity Catalog permissions model, Databricks documentation, last updated Oct 15, 2026
- Attribute-based access control, Databricks documentation, last updated Oct 22, 2026
- Lineage in Unity Catalog, Databricks documentation, last updated Nov 4, 2026
- Access Databricks tables from Apache Iceberg clients, Databricks documentation, last updated Oct 27, 2026
- Cross-engine ABAC, Databricks documentation, last updated Oct 20, 2026
- Lakehouse Federation, Databricks documentation, last updated Sep 29, 2026
- Databricks release notes, April 2026 (ABAC GA, April 28) and October 2026 (cross-engine ABAC GA, October 1), Databricks on AWS documentation
- Databricks, Atlan Documentation
- What does Atlan crawl from Google Cloud Knowledge Catalog, Atlan Documentation
- Atlan MCP, Atlan Documentation
- Metadata: The Critical Foundation for AI Agents & Natural Language to SQL Use Cases, Atlan
