This page is for data governance leads, data platform owners and analytics teams who need people, and increasingly AI agents, to find the right data, understand what it means and use it within policy. It covers enterprise governance suites with stewardship workflows, metadata-driven catalogs aimed at the modern data stack, open source metadata platforms you host yourself, and the catalogs that Microsoft, Google, AWS, Databricks and Snowflake build into their platforms.
The main decision is scope. A catalog that governs one platform is often included in what you already pay for; a neutral catalog that spans many warehouses, databases and BI tools is a separate purchase and a long-term operating commitment for stewards and owners.
What this software does
A catalog starts by harvesting technical metadata: connectors or crawlers read schemas, tables, columns, views, jobs and reports from each source, either pulling on a schedule or receiving pushed events. It adds usage from query logs (who queries what, how often) and builds lineage, the graph of how data flows from source tables through transformations into dashboards and models. Column-level lineage, increasingly exchanged through the OpenLineage standard, is what makes impact analysis and root-cause work practical.
Governance features add business metadata on top: a business glossary of agreed terms, owners and stewards for each asset, certification of trusted datasets, data products with contracts, and workflows for proposing and approving changes. Classification tags sensitive columns such as personal data, manually or by automated scanning, so that policies, privacy assessments and masking can follow.
Catalogs differ most in whether they document policy or enforce it. Platform catalogs such as Databricks Unity Catalog, Snowflake Horizon Catalog and AWS Glue Data Catalog with Lake Formation sit in the query path, so their grants, row filters and column masks are applied at query time. Neutral catalogs mostly record policies and push them to platforms where integrations exist. Open table formats are also reshaping the market: Iceberg REST catalog endpoints let several engines share one set of tables and permissions.
Types of product in this category
- Enterprise catalog: Commercial, mostly SaaS platforms such as Collibra Platform, Alation, Atlan, Informatica CDGC, OvalEdge and Ataccama ONE that catalog many sources and add glossary, stewardship workflows, lineage and policy management across the estate.
- Open source catalog: Self-hosted metadata platforms such as DataHub, OpenMetadata and Apache Atlas, free under Apache 2.0, with managed commercial editions from DataHub and Collate; Amundsen is listed for reference only, as its repository was archived in September 2026.
- Cloud-native catalog: Catalogs built into a cloud or data platform, such as Microsoft Purview, Knowledge Catalog, AWS Glue Data Catalog, Unity Catalog and Snowflake Horizon Catalog, that index that platform's assets automatically and often enforce access policies, billed as platform usage.
Typical use cases
Self-service data discovery
Analysts search for "active customers" and find the certified table, its owner, its definition and how often it is used, instead of asking in chat or rebuilding a query from scratch.
Impact analysis before a change
Before dropping or renaming a column, an engineer checks column-level lineage to see which pipelines, dashboards and models depend on it and notifies their owners.
Privacy and regulatory programmes
Classifying personal data across databases and warehouses supports EU GDPR records of processing, data subject requests and privacy assessments, and gives masking and access policies something reliable to act on.
Central access policy on a data platform
On Databricks or Snowflake, grants, row filters, column masks and tag-based policies defined once in the platform catalog apply to every query and workload, with lineage and audit logs alongside.
Shared tables across engines
With an Iceberg REST catalog, Spark, Trino and other engines read the same tables under one set of permissions. AWS Glue Data Catalog, Unity Catalog and Snowflake Horizon Catalog all expose Iceberg REST endpoints.
Context for AI agents
Glossaries, certified datasets and semantic definitions give AI assistants and text-to-SQL agents the business context they lack. Several catalogs, including Atlan, DataGalaxy and Select Star, now expose metadata through MCP servers and APIs.
How to choose
- Scope: one platform or the whole estate. If most data lives in Databricks, Snowflake, Google Cloud, Azure or AWS, start with the platform catalog you already have. Choose a neutral catalog when you need one inventory across several warehouses, operational databases, ETL and BI tools; Atlan, for example, documents layering over Purview, Horizon and Unity Catalog.
- Connector depth for your sources. Count connectors you actually need, and check what each one extracts: schema only, or also lineage, usage, profiling and classification. Lineage is often source-dependent or sold as an add-on per connector, as at OvalEdge, and Microsoft documents that lineage, classification and policies are not available for every Purview source.
- Lineage granularity and standards. Check table versus column-level lineage, whether it crosses ingestion, transformation and BI tools, and whether you can push your own lineage through OpenLineage or APIs. Ask how lineage from SQL is parsed and how stored procedures and dynamic SQL are handled.
- Governance workflow fit. Formal operating models with domains, owners, stewards and approvals favour products with configurable workflows, such as Collibra. Lighter, engineering-led teams may prefer catalogs that automate documentation and certification. Platform catalogs are generally thin on glossary and stewardship workflows.
- Policy documentation or enforcement. Confirm whether access and masking policies are enforced at query time or only recorded. Unity Catalog, Horizon Catalog, Lake Formation and IBM Knowledge Catalog data protection rules enforce within their platforms; most neutral catalogs rely on integrations to push policy.
- Deployment and data residency. Options range from multi-tenant SaaS to single-tenant SaaS (Atlan), self-hosted editions (Collibra, Alation customer-managed, Secoda Enterprise, OvalEdge On-Prem) and fully self-hosted open source. Check where metadata extraction runs; Collibra Edge and Atlan Self-Deployed Runtime keep it inside your network.
- Adoption and seat model. A catalog is only useful if people use it. Check how readers are licensed: several vendors include unlimited or large numbers of view-only users (Ataccama, Secoda, DataGalaxy, OvalEdge) and charge for authors, connectors or governed assets.
- Operating an open source catalog. DataHub and OpenMetadata are capable and free, but self-hosting means running several services, including a search engine, and handling upgrades. Some features are reserved for the managed editions, so check parity before assuming the open source build covers your needs.
- Vendor and naming stability. This market changed fast: Informatica is owned by Salesforce, Secoda by Atlassian, data.world by ServiceNow, Select Star by Snowflake, Google's catalog is now Knowledge Catalog after Dataplex renames, IBM Knowledge Catalog is watsonx.data intelligence on IBM Cloud, and Amundsen is archived. Confirm standalone availability and roadmap in writing.