Imports

Most teams already have some semantic or lineage assets. Keel can take them in so a model does not start from zero.

Available in v1.0

What	How	Result
OpenLineage run events	POST /api/v1/lineage/openlineage (API key with admin:*, or a user with ?workspaceId=)	Dataset and column lineage (inputs, outputs, the column-lineage facet) in scope openlineage:{namespace}/{job}. FAIL and ABORT events change nothing.
dbt manifest.json	POST …/lineage/dbt	Model dependencies as dataset-level lineage, mapped onto scanned tables (dbt:{project}); unmapped datasets are listed.
Glossary CSV / JSON	Glossary › ··· › Import or POST …/glossary:import	Terms (Draft), with counts and row errors.

External datasets are matched to scanned tables by the last parts of their name (db.schema.table), narrowed by the namespace's host, port and database; anything unmatched stays external. Imports are audited.

Model imports: dbt, LookML, Cube

Bootstrap a model from what your BI layer already defines. Open Import from the model bar's model menu, or Import from dbt / LookML / Cube on the Change requests page, pick the format and upload the files or a zip of the project. Preview shows what each object would do; Import opens a change request.

Format	Read	Reported, not guessed
dbt Semantic Layer (MetricFlow YAML + model SQL)	semantic models, entities → joins, dimensions, measures, simple / ratio / derived / cumulative metrics, config.meta.keel	other aggregations, conversion metrics, saved queries, Jinja SQL (imported verbatim, flagged)
LookML	views (sql_table_name, derived-table SQL), dimensions, dimension groups (grains), measures and filters, type: number measures as ratio / derived metrics, explore joins (roles from from:)	parameters, filter fields, Liquid, extends / refinements, access grants, datagroups, aggregate tables, cross-view field references
Cube (YAML)	cubes, dimensions, measures and filters, calculated measures as metrics, segments, hierarchies, joins	views, pre-aggregations, access policies, extends, JavaScript models, templated SQL

• Every imported object is a proposal: origin Imported, a confidence (lower when approximated) and the file:line it came from as evidence.
• Tables bind to the model's scanned sources by schema and table name; a table that is not in the latest scan is imported without a source and reported.
• The import lands on its own branch, keel/import-{format}-{id}, with a change request into main. Objects main already has identically change nothing. Objects main has differently are kept as conflicts: when merging, choose per object whether to keep main's version or take the imported one. Nothing on main changes without that decision.
• The report lists every construct that was skipped or approximated, and every file that could not be read, with its line. Uploads: up to 10 MB (20 MB unzipped), text model files only.
• API: POST …/models/{id}/imports:preview, POST …/models/{id}/imports (multipart format + files; EditModels), GET …/models/{id}/imports. Imports are audited.

Data-quality imports: dbt tests, Great Expectations

On Data quality › Import dbt tests / GE suite, choose the source the checks describe and upload dbt schema YAML files or Great Expectations suites (JSON), or a zip. Each check that maps onto a scanned table and column becomes a suggested rule (origin Imported, with its file and line) in Suggested rules — nothing runs until a Steward accepts it.

dbt	Great Expectations	Keel rule
not_null, dbt_utils.not_null_proportion	expect_column_values_to_not_be_null	Not null
unique, dbt_utils.unique_combination_of_columns	expect_column_values_to_be_unique, expect_compound_columns_to_be_unique	Unique
accepted_values	expect_column_values_to_be_in_set	Accepted values
relationships	—	Referential integrity
dbt_utils.accepted_range, dbt_expectations.expect_column_values_to_be_between	expect_column_values_to_be_between	Range
dbt_expectations.expect_column_values_to_match_regex	expect_column_values_to_match_regex	Pattern
dbt_utils.recency	—	Freshness
dbt_utils.expression_is_true, expect_table_row_count_to_be_between	expect_table_row_count_to_be_between / _to_equal	Custom SQL (read-only)

dbt severity: warn (or a GE .warning suite) becomes a Warning rule, otherwise Critical. Checks that do not map (an unknown table or column, an unsupported test) are listed with their line; importing again refreshes undecided suggestions, and a check Keel already suggested from the scan is not proposed twice. API: POST /api/v1/workspaces/{ws}/quality/rules:import (multipart format = dbt or great_expectations, dataSourceId, files; Govern).