Unity Catalog enforces access and captures column-level lineage inside Databricks. Collibra records policies, stewardship workflows, and AI lifecycle stages across the enterprise. Atlan’s context layer imports Unity Catalog tags, pushes tag updates back to Databricks through reverse sync, and serves context to agents through a hosted MCP server.
The two products overlap more than a clean split suggests. Both hold lineage, both register AI assets, and Collibra documents an integration that reads Unity Catalog metadata. They differ on the point of record: the system where access is enforced, or the system where policy and stewardship decisions are written down.
Teams adding AI agents to either estate meet a third question, which is where an agent reads business definitions at runtime. A context layer answers it, and the enterprise context layer carries that answer across systems.
Unity Catalog and Collibra at a glance
| Dimension | Unity Catalog | Collibra |
|---|---|---|
| Where it runs | Built into Databricks, with cross-engine ABAC (GA October 1, 2026) | Across systems, with Databricks connected through Edge |
| Governance record | Privileges, ABAC policies (Beta labels on parts), audit log system table | Policies, workflows, stewardship structure |
| Access enforcement | Privileges, row filters, column masks | Policy adoption and monitoring in Policy Manager |
| Lineage | Automatic, column level, for queries run on Databricks | System-to-system lineage as a listed product |
| AI assets | MCP servers in system.ai, governed with grants |
Registry and lifecycle for agents, models, use cases |
What does Unity Catalog govern?
Unity Catalog governs data and AI assets inside Databricks through privileges, policies, lineage, and audit records. According to the Databricks Unity Catalog overview (September 11, 2026), it is “the unified governance layer for data and AI built into Databricks.”
Access control combines privileges, attribute-based policies, row and column filters, and workspace bindings. Securable objects follow a catalog.schema.object namespace, activity lands in the audit log system table, and Catalog Explorer handles discovery. The Context Layer for Databricks guide maps how those native pieces fit together.
Lineage is automatic. According to the Databricks Unity Catalog data lineage page (updated November 4, 2026), Unity Catalog captures lineage “down to the column level” for queries run on Databricks, for tables registered in a Unity Catalog metastore. Column lineage applies to registered tables, and a source or target referenced by file path is outside it.
Attribute-based access control (ABAC) uses governed tags as attributes, with policies attached at the metastore, catalog, schema, or table level (Databricks ABAC documentation, October 22, 2026). Metastore-level policies, views support, and DENY policies carry Beta labels on that page. A Databricks release note dated October 1, 2026 announces cross-engine attribute-based access controls as generally available. The same status care applies to business glossary and domains, where GA and Preview labels differ by feature.
Core components of Unity Catalog
- Access privileges and ABAC: privileges, row filters, column masks, and tag-driven policies, with Beta labels on parts of ABAC.
- Lineage: automatic capture down to the column level for queries run on Databricks.
- Audit log system table: the record of activity across governed objects.
- Catalog Explorer: discovery and browsing of securable objects, including governed metrics and the semantic layer built on them.
These pieces cover assets registered in Unity Catalog. What an agent should read when its answer spans more than one system is a separate job.
What does Collibra govern?
Collibra governs policies, stewardship, workflows, and the lifecycle of data and AI assets across the enterprise. According to Collibra’s Platform products and features documentation (September 18, 2026), Data Catalog is a metadata catalog for business and data stewards to discover, describe, assemble, and govern data sets.
The same page lists Stewardship, Policy Manager, AI Governance, Data Lineage, and Workflow Designer as products. Stewardship establishes an organization’s governance structure, and Policy Manager adopts, implements, and monitors digital policies. Workflow Designer is a process and form modeling web application.
AI assets get their own lifecycle. According to Collibra’s Lifecycle management and AI Governance documentation (October 7, 2026), oversight runs from creation to archive for AI Agent, AI Agent Version, AI Base Model, AI Model Version, AI Use Case, and Data Product asset types. Stages can carry assessments and sign-offs.
Those sign-offs are the stage gates an AI governance framework usually formalizes, and they sit beside the agent memory governance questions teams hit once agents persist state. AI agent governance splits into agent controls and governed data, and these records speak to the first half.
Core components of Collibra
- Data Catalog: a metadata catalog for business and data stewards to discover and describe assets.
- Stewardship: the structure that assigns who is responsible for which data.
- Policy Manager: adoption, implementation, and monitoring of digital policies.
- Workflow Designer: process and form modeling for approvals and reviews.
A catalog record serves stewards. A context catalog is how Atlan delivers definitions, ownership, and lineage to agents.
Unity Catalog vs Collibra: head-to-head comparison
The two products differ most in where the governance record lives and who writes it, and they overlap on lineage and AI asset registration. Unity Catalog is the enforcement point inside Databricks. Collibra is the enterprise record of policy, stewardship, and lifecycle decisions.
| Dimension | Unity Catalog | Collibra |
|---|---|---|
| Primary focus | Enforcement and lineage inside Databricks | Policy, stewardship, and lifecycle records across the enterprise |
| Policy form | Privileges, row filters, column masks, ABAC (Beta labels on parts) | Digital policies adopted and monitored in Policy Manager |
| Lineage scope | Column level, for queries run on Databricks | System-to-system lineage as a listed product |
| Audit and sign-off | Audit log system table | Lifecycle stages with assessments and sign-offs |
| AI assets | MCP servers in system.ai, invoked under grants |
Registry and lifecycle for agents, models, and use cases |
| Integration path | Source of metadata that Collibra’s Edge integration synchronizes inbound | Edge integration for Unity Catalog metadata and Databricks AI model links |
| Status labels in docs | Cross-engine ABAC GA; metastore-level, views, and DENY policies Beta; Unity Gateway MCP servers Beta; Genie One MCP GA | AI model and agent integrations documented through Edge |
Picture a Databricks-heavy estate with a governance office. A Unity Catalog column mask protects a customer table on read, and column lineage shows which Databricks queries touched it. Collibra holds the policy text, the steward’s sign-off, and the AI use case record for the agent that reads the table.
The overlap is real, so “Unity Catalog enforces, Collibra defines” is a simplification, and neither product is better than the other: each is the right record for a different job. Both hold lineage, and both register AI assets. For agents, the useful form of that lineage is AI-ready data lineage, traced into and out of the model.
Unity Catalog’s own modeling layer lives in Databricks, a boundary covered in Unity Catalog vs a cross-platform semantic layer. The data catalog vs context layer distinction applies the same way to Collibra’s catalog, and the question of how the data catalog is changing is really a question about who reads the record, humans or agents.
When should you use Unity Catalog, Collibra, or both?
Choose by where your enforcement and stewardship records need to live, then check the overlap. Unity Catalog fits access control inside Databricks, Collibra fits enterprise policy and stewardship, and both fit an estate that needs each record.
| If your team… | Start with |
|---|---|
| Needs access control and lineage for data queried in Databricks | Unity Catalog |
| Needs policy approval, stewardship roles, and lifecycle sign-offs across several systems | Collibra |
| Runs Databricks plus other systems and needs both enforcement and a policy record | Both, connected through Edge |
That last check matters here. If Unity Catalog holds the lineage and Collibra holds the approvals, the need may be a connection between the two, not another tool. An agentic data catalog approaches the same problem from the catalog side.
How do Unity Catalog and Collibra connect?
Collibra documents an Edge-based integration that brings Unity Catalog metadata into its catalog. According to Collibra’s integration steps for Databricks Unity Catalog via Edge (October 20, 2026), one step is titled “Synchronize inbound metadata from Databricks Unity Catalog.”
For AI assets, Collibra’s About integrating AI models and agents page (October 9, 2026) documents integrations through Edge. For Databricks AI, model versions link to the Delta tables used as dataset inputs for their source MLflow run, and model deployments link to the Delta table that serves as the endpoint’s inference table.
That flow runs inbound to Collibra and gives its stewards a view of what exists in Databricks. Serving an agent one consistent definition of “active customer” is the job of a unified context layer, and the enterprise data graph is the structure that holds them.
What do AI agents need across Unity Catalog and Collibra?
An agent working across Databricks and other systems reads context from more than one place. Each system supplies part of it, and the agent-facing surfaces differ.
| Surface | What it provides | Source and date |
|---|---|---|
| Databricks SQL MCP | Runs with the caller’s Unity Catalog and warehouse permissions | Databricks docs, October 23, 2026 |
| Genie One MCP | GA managed MCP server in the Databricks AI product family | Databricks docs, October 23, 2026 |
| Collibra AI asset registry | Registers and monitors AI agents, models, and use cases with lifecycle stages | Collibra docs, October 7, 2026 |
According to Databricks’ managed MCP servers documentation (October 23, 2026), MCP servers live in the system.ai schema, and admins govern who invokes them with grants. The dbsql, sandbox, and web_search servers are in Unity Gateway Beta, Genie One MCP is GA, and other workspace endpoints are Public Preview. These are capabilities of the Databricks AI product family, and Genie One MCP and Databricks’ runtime agent governance each get their own walkthrough. The wider MCP server landscape for Databricks shows how the server types relate.
Collibra’s AI Governance registers and monitors AI agents, models, and use cases across the enterprise (Platform products and features, September 18, 2026).
What an agent reads at runtime is context: business definitions, ownership, lineage, and quality signals, held across systems. Policy decisions stay where they are enforced. An MCP-connected data catalog moves that record to the agent. MCP delivers business context to agents in the same way, and context engineering is the discipline of deciding which context reaches which agent.
How does Atlan’s context layer work alongside Unity Catalog and Collibra?
Atlan is the context layer for AI. It holds the definitions, ownership, lineage, and quality signals agents read across the systems it connects, and leaves policy decisions where they are enforced. Unity Catalog stays the enforcement point inside Databricks, and Collibra stays the enterprise record of policy and stewardship.
Atlan connects to Unity Catalog in two documented ways. According to Atlan’s docs on managing Databricks tags, Atlan imports Unity Catalog tags into the catalog so classifications stay consistent across sources, and reverse sync pushes tag updates for Databricks assets back to Databricks. Reverse sync updates existing tags in Databricks and neither creates nor deletes any.
Atlan’s Unity Catalog governance FAQ lists what is not carried into Atlan and stays in Unity Catalog: privilege grants and access control lists, row filters and column masks, and audit log events. Atlan also reads the same Databricks system tables that Unity Catalog lineage uses, including system.access.column_lineage, per Atlan’s docs on Atlan and Databricks lineage.
Atlan’s hosted MCP server is available on every Atlan tenant with nothing to install, and its actions respect the permissions set in Atlan (Atlan MCP overview). This is the coexistence pattern: Unity Catalog for enforcement, Collibra for enterprise policy and stewardship, and Atlan serving agents context from the systems it connects. Genie Ontology and the Atlan context layer shows the Databricks side of that pairing, and neutral context layers compared with runtime agent control shows where each sits.
An agent query that crosses two systems crosses a boundary, which is the subject of platform-native context layers. The core components of a context layer and the business context layer describe what has to exist on the far side of it. The human side of that handoff, why teams hesitate to give agents control, is covered in fear, trust, and control in enterprise agents.
Enforcement, policy records, and agent context are three jobs
Unity Catalog is where Databricks access is enforced. Collibra is where enterprise policy and stewardship records live. The overlap on lineage and AI assets is real, so neither one is a simple substitute for the other.
An agent needs a third thing: context it can read at query time. A context layer supplies it without taking over enforcement or the policy record. Teams ready to wire that into a working agent can follow how to build an AI agent harness, where the context question comes before the loop.
FAQs about Unity Catalog vs Collibra
1. Is Unity Catalog a data governance tool?
Yes, for data and AI assets inside Databricks. Databricks describes Unity Catalog as the unified governance layer for data and AI built into Databricks, covering privileges, attribute-based policies, row and column filters, lineage, and an audit log system table. Enterprise-wide policy and stewardship workflows are the work Collibra documents.
2. What is the difference between Unity Catalog and Collibra?
Unity Catalog enforces access and records lineage and audit activity inside Databricks. Collibra holds policies, stewardship structure, workflows, and lifecycle sign-offs across the enterprise. Both hold lineage and both register AI assets, so the practical difference is where the governance record lives and who writes it.
3. Can you use Unity Catalog and Collibra together?
Yes. Collibra documents an Edge-based integration with a step titled “Synchronize inbound metadata from Databricks Unity Catalog” (Collibra docs, October 20, 2026). That pairing puts enforcement in Unity Catalog and the policy record in Collibra.
4. Does Unity Catalog capture column-level lineage?
Yes, for queries run on Databricks. Per Databricks docs (updated November 4, 2026), Unity Catalog captures lineage automatically down to the column level for tables registered in a Unity Catalog metastore.
5. Does Collibra govern AI agents and models?
Yes. Collibra’s Lifecycle management and AI Governance documentation (October 7, 2026) covers oversight from creation to archive for AI Agent, AI Agent Version, AI Base Model, AI Model Version, AI Use Case, and Data Product asset types. Stages can carry assessments and sign-offs.
6. Are Unity Catalog attribute-based access control policies generally available?
Parts are generally available and parts are Beta. A Databricks release note dated October 1, 2026 announces cross-engine attribute-based access controls as generally available. The ABAC documentation (October 22, 2026) labels metastore-level policies, views support, and DENY policies Beta. Confirm the status of the specific feature in current Databricks docs before you depend on it.
7. How does Atlan help AI agents use Unity Catalog and Collibra context?
Atlan imports Unity Catalog tags and can push tag updates back to Databricks through reverse sync, and it reads lineage from the same Databricks system tables Unity Catalog uses. Its hosted MCP server connects AI clients to that context. Enforcement stays in Unity Catalog and the policy record stays in Collibra.
Sources
- Unity Catalog overview, Databricks documentation (September 11, 2026)
- Unity Catalog data lineage, Databricks documentation (updated November 4, 2026)
- Attribute-based access control, Databricks documentation (October 22, 2026), and ABAC policies, Databricks documentation (October 15, 2026); both are documentation-area names
- Databricks release notes, October 2026, entry dated October 1, 2026 (cross-engine attribute-based access controls, generally available)
- Managed MCP servers, Databricks documentation (October 23, 2026)
- Platform products and features, Collibra documentation (September 18, 2026)
- Lifecycle management and AI Governance, Collibra documentation (October 7, 2026)
- Steps: Integrate Databricks Unity Catalog via Edge, Collibra documentation (October 20, 2026)
- About integrating AI models and agents, Collibra documentation (October 9, 2026)
- Manage Databricks tags, Atlan documentation
- Unity Catalog governance and connections, Atlan documentation
- Atlan vs Databricks lineage, Atlan documentation
- Atlan MCP overview, Atlan documentation
