Customer warehouses

Connect BigQuery, ClickHouse, Snowflake, Elasticsearch, or OpenSearch with least-privilege credentials.

Last verified 2026-08-02

Customer-owned mode stores and queries data in infrastructure you operate. Create the credential first or during the warehouse workflow, then select the backend-specific target.

BigQuery

Required target fields:

  • Dataset, for example analytics;
  • Location, for example EU.

The credential is a Google service-account JSON document. Default behavior uses the model key as the table name, ensures the dataset, and can add compatible columns.

Use a dedicated Google Cloud service account with only the dataset permissions required for table management, writing, and querying.

ClickHouse

Required target field:

  • Database, for example default.

An optional cluster name can be supplied. The credential includes host, username, password, database, and whether the connection is secure. Default behavior uses the model key as the table name and can create tables or add columns.

Snowflake

Required target fields:

  • Database, for example REPORTPLANE;
  • Schema, commonly PUBLIC or a dedicated schema.

The credential contains the account, username, authentication secret, compute warehouse, and role. Use a dedicated least-privilege role rather than ACCOUNTADMIN in production.

Elasticsearch

Optional target fields include an index prefix and ingest pipeline. Default behavior derives an index from the model key and can create it when allowed.

Supported credential shapes include compatible username/password and API-key configurations. The service URL must be reachable from ReportPlane.

OpenSearch

OpenSearch uses the same general index strategy and optional prefix/pipeline concepts. Choose OpenSearch explicitly so the adapter applies the correct compatibility behavior.

Connection checklist

Before saving:

  1. confirm the endpoint is reachable;
  2. verify TLS and certificate policy;
  3. use a non-personal service identity;
  4. grant only required database/index permissions;
  5. confirm the database, dataset, schema, or index naming policy;
  6. test with a non-production model first.
Was this page helpful?Send feedback