dbt seed is a dbt command that uploads static CSV files in your project’s data/ folder into your data warehouse as managed tables, making small reference datasets version-controlled and easily queryable.
dbt seed loads CSV files stored in a project’s data/ directory into your warehouse as tables, giving you version-controlled reference or lookup data with one command.
dbt seed automates table creation, handles schema changes, and ties data to Git history, whereas manual COPY commands live outside version control and require repetitive boilerplate.
During dbt seed, dbt reads each CSV, infers column types or applies column-level quoting, generates CREATE and INSERT statements, and stores a checksum in manifest.json for freshness checks.
Add settings in dbt_project.yml under seeds: to set database, schema, header, delimiter, quote_columns, and file-specific overrides.
seeds:
my_project:
+schema: staging
+quote_columns: false
users.csv:
+column_types:
id: integer
plan: varchar(10)
Use dbt seed --select my_seed or dbt seed --exclude large_seed to control which CSVs load, saving build time in CI pipelines.
Keep files under ~100k rows, store only static or slowly changing data, set explicit column types, and add tests for row_count and not_null keys.
Avoid dbt seed for large fact tables or frequently updated data; use proper ELT pipelines or warehouse-native staging instead.
In Galaxy’s SQL editor, seeded tables appear instantly in the sidebar metadata, letting you autocomplete against them and share validated seed queries inside Collections.
Version-controlled reference data eliminates hidden CSV uploads, keeps dev, CI, and prod in sync, and speeds testing by guaranteeing deterministic lookup tables.
No. dbt seed truncates and reloads the table each run. Use incremental models for updates.
Place them in the project’s data/ folder so dbt auto-detects them.
Use dbt seed --exclude filename in your prod job or configure env-specific select flags.
Yes. Galaxy indexes seeded tables, offering instant autocomplete, query sharing, and AI explanations just like any other table.
.avif){kind=logo}
