OpenMetadata Integration

Bootstrap a Wren project from an OpenMetadata catalog instead of introspecting the live database.

Why import from OpenMetadata

When a team already catalogs its warehouse in OpenMetadata, the catalog holds context the raw database schema does not:

• Curated descriptions for tables and columns, written by data owners — usually better than database comments.
• Tags and owners (for example PII.Sensitive, Tier.Tier1).
• Foreign-key constraints, which become explicit Wren join paths in relationships.yml. Join inference is the step an agent is most likely to get wrong when it only sees table and column names.

What Wren imports

OpenMetadata input	Wren output
Table entities in the selected service / database / schema	models/<table>/metadata.yml
Column name, dataTypeDisplay, NOT_NULL / PRIMARY_KEY constraint	Column type (normalized for the datasource dialect) and not_null
PRIMARY_KEY table constraint	Model primary_key (a list for composite keys)
FOREIGN_KEY table constraints	relationships.yml entries (properties.source: openmetadata_foreign_key)
Table / column descriptions	properties.description
Table / column tags	properties.om_tags (list of tag FQNs)
Table owners	properties.om_owners (list of user or team names)
Table type	properties.om_table_type
Fully-qualified names	properties.om_fqn on every model and column
Database service type	data_source in wren_project.yml

properties.om_fqn makes the mapping between a Wren object and its OpenMetadata entity deterministic, so later tooling can look the entity up without guessing.

The generated wren_project.yml records where the project came from:

openmetadata:
  service: shop_pg
  database: shop
  schema: shop
  server_version: 2.0.5

What is not imported

• Lineage is not turned into relationships. OpenMetadata lineage records how data is derived (table B is computed from table A), not how tables join. Only foreign-key constraints produce relationships, so tables with no foreign keys get no relationships.
• Connection settings. Only the datasource type comes from OpenMetadata. Host, port, user and database are not imported (the address OpenMetadata ingests through is often not reachable from where Wren runs) and OpenMetadata masks passwords, so bind a Wren profile separately.
• Glossary terms, metrics and other knowledge assets, and incremental re-sync, are not covered by this command.

Foreign keys that point at a table outside the selection are skipped and listed in the command output. Tables with no columns in OpenMetadata, or whose name cannot be used as a model directory (for example one containing /), are skipped and listed too. Widen the selection (for example drop --schema) to include the referenced table.

Prerequisites

• An OpenMetadata server. Tested against 2.0.x (verified on 2.0.5). Other versions run, but the command prints a warning.
• A database service already ingested into OpenMetadata, with table constraints enabled in the metadata ingestion so foreign keys are present.
• A token with read access to the catalog: an OpenMetadata bot token (Settings → Bots) or a personal access token (Profile → Access Tokens).

Import

Pass the connection with flags or environment variables:

export OPENMETADATA_HOST=http://localhost:8585
export OPENMETADATA_TOKEN=<token>

wren context import openmetadata \
  --service shop_pg --database shop --schema shop \
  --path ./wren-shop --dry-run

wren context import openmetadata \
  --service shop_pg --database shop --schema shop \
  --path ./wren-shop

Flag	Meaning
--host / OPENMETADATA_HOST	OpenMetadata server URL. https:// is assumed when no scheme is given; write http:// explicitly for a local quickstart. /api/v1 is appended if missing
--token / OPENMETADATA_TOKEN	Bot or personal access token
--service	Database service name as shown in OpenMetadata (required)
--database	Only import tables from this database
--schema	Only import tables from this schema (requires --database)
--table	Only import this table; repeat for several
--data-source	Wren datasource to use when the service type has no automatic mapping
--dry-run	List the files that would be written
--force	Regenerate from OpenMetadata (see below)

Service types mapped automatically: Athena, BigQuery, ClickHouse, Databricks, Doris, DuckDB, MariaDB, MySQL, Mssql, Oracle, Postgres, Redshift, Snowflake, Trino.

Then bind a connection profile and finish the project:

wren profile add shop --datasource postgres --interactive
wren context validate --path ./wren-shop
wren context build --path ./wren-shop

Re-importing

OpenMetadata is the source of truth for the import, in the same way dbt is for wren context import dbt:

• Without --force the command only adds. Files that already exist are never modified, so descriptions or relationships you edited locally survive. Every collision is listed — models that already exist, relationships that were not added, and relationships whose local definition differs from OpenMetadata. Use this to pull newly cataloged tables into an existing project. If there is nothing new to write, the command exits with status 1.
• With --force the generated files (models/, views/, relationships.yml, instructions.md, wren_project.yml, AGENTS.md) are deleted and regenerated from OpenMetadata. Local edits to those files are discarded. Make durable description changes in OpenMetadata itself.