OpenMetadata
OpenMetadata is an open-source metadata platform combining catalog, lineage, data quality tests, and collaboration on a single schema-first foundation with many connectors.
LicenseOpen source (permissive)· Apache-2.0
DeploymentSelf-hostedManagedHybrid
PricingFreeSubscription
Operational complexityMedium
WorkloadInteractiveBatch
LanguageTypeScript, Java, Python
Use it when
- You want catalog, column-level lineage, profiling, quality tests, and incident management in one open-source deployment, with none of it gated behind a paid tier.
- Connector breadth matters: 96 documented connectors across databases, dashboards, pipelines, and messaging.
- The metadata model should be a published standard: 700-plus JSON Schemas from which the platform, APIs, and client libraries are generated.
- Glossary changes deserve review: terms pass an approval workflow before they publish.
Think twice when
- Nobody can operate it: self-hosting means running the platform, its search index, and the ingestion framework; Collate is the managed escape hatch.
- You need access-request workflows for data; nothing here requests or grants permissions on source systems.
- The organisation buys governance platforms rather than installing tools; Collibra and Alation are that conversation.
How it runs
Self-hosted, or managed through Collate. An ingestion framework runs connectors on a schedule from the UI or an external orchestrator; RBAC and ABAC policies govern what users see inside the catalog; the profiler, no-code test suites, alerting, and incident management all ship in the open-source edition.
Details
CompareHow OpenMetadata answers the questions Data Catalogs turns on.
Data Catalogs
| How it works | |
| Metadata model | A published standard rather than an internal model, over 700 JSON Schemas covering entities, relationships, events and the REST API itself, from which the platform, its APIs and its client libraries are generated, and which adopt JSON Schema, RDF/OWL, SHACL, JSON-LD and PROV-O rather than inventing formats |
| Lineage | Column-level and editable: extracted automatically from warehouses and BI tools, built from query history where the source exposes it, written through the Python SDK, and corrected by hand in the UI when none of that is right; cross-database lineage resolves without cross-service access |
| Search and discovery | Search across every asset type with profile, quality status and ownership on the page, narrowed by domain, tag, glossary term and service, and entities are versioned, so how a description or a schema changed over time is itself discoverable |
| Business glossary | Terms in a hierarchy with an approval workflow behind them, so a term reaches the catalog after review rather than on save, with classification tags alongside for PII and retention |
| Quality checks | Native, and in the open-source edition, a profiler, no-code test cases grouped into suites, incident management with root-cause analysis, and alerting when tests fail. This is the clearest difference from DataHub, where assertions are a Cloud feature |
| Custom metadata | Custom properties on entities for the ordinary case, and beneath that the schemas themselves, the standard is community-governed and code-generated, so extending the model means extending JSON Schema rather than a vendor's object graph |
| Connections | |
| Connectors | 96 connectors, counted from the documentation (51 databases, 16 dashboards, 13 pipelines, 4 messaging, 3 metadata services, 2 each for ML models, storage, drives and search, and 1 API) run by an ingestion framework schedulable from the UI or an external orchestrator |
| Access | |
| Policy and compliance | RBAC and ABAC together, over metadata rather than data, rules name resources from tables and topics to glossary terms, domains, users and policies themselves, and operations from create and delete through ViewDataProfile, ViewSampleData and ViewQueries. The documentation is explicit that this controls what a user sees inside OpenMetadata, not authentication to the source system |
| Access requests | None for access to data. An approval workflow does exist, but it gates a glossary term's publication rather than a person's access to a table, so nothing here requests, approves or grants a permission |
| Cost | |
| Billing unit | Free to self-host. Collate, the managed offering, publishes its units and per-tier allowances but not its rates, Free at 5 users and 500 data assets, Premium at 25 and 5,000, Enterprise at 50 and 10,000 with unlimited options, buying more of either as needed; a free cluster is reclaimed after four weeks without a login |
Alternatives to OpenMetadata
