Most data teams that use dbt have already written down what revenue, active customers and margin mean, in semantic_models and metrics YAML. A BI tool that cannot read those files makes them write it again, and two definitions of the same figure start to drift the day they are created. ATX AnalyticsTable reads a dbt MetricFlow project or an Apache Ossie (Open Semantic Interchange, OSI) model into its own governed semantic layer, previews what will be created and what will not, and can write the catalog back out as dbt YAML or OSI. The same definition then serves dashboards, pipelines, Ask Greg, the REST API and AI agents through MCP.
Settings → Semantic objects → Import dbt / OSI… takes the files and a target connection. Nothing is written until you confirm the preview.
Each part of the source becomes a native semantic object with its own URN, so it can be certified, versioned, restored and queried like one written by hand.
Define measures, dimensions and joins once; dbt MetricFlow and OSI definitions import and export.
An object with no native form is never half-imported. The preview lists it with the reason, and so does the report after the import. That covers conversion metrics (a funnel needs each entity's events in order, which the planner does not model), offset_to_grain, a filter that names another model's dimension or uses other Jinja, and an OSI metric whose expression has no usable dialect.
Settings → Semantic objects → Export dbt / OSI writes the catalog, or a selection plus everything it depends on, as one dbt MetricFlow semantic_models.yml or as an OSI 0.2 or 0.1 model. The same export is available over the API as POST /api/semantic/export.
OSI has no word for some of what a governed metric holds, such as dimension levels, the exact aggregation, join cardinality, calculated-member formulas, time windows, filters and parameters. Those travel in a custom_extensions entry for the vendor ATX_ANALYTICSTABLE, which the importer restores, so ATX → OSI → ATX loses nothing while another tool still sees a plain OSI model. What dbt cannot hold, such as a SQL-view dataset, a custom aggregation or a formula with parameters, is left out and listed with the reason.
Imported objects are tagged imported:dbt or imported:osi and start uncertified. With the publish gate on, dashboards, reports and the API serve them only after a steward has certified them, in a four-eyes workflow recorded against each definition.
Once a definition is in the catalog it is the source for charts, pivots and KPI widgets, the METRIC() formula in a cell, the Governed metrics step of a scheduled pipeline, Ask Greg's answers, and the headless metrics endpoints of the REST API. ATX AnalyticsTable is also an MCP server, so Claude, Copilot Studio, Cursor or your own agent can search, describe and query the same metrics, read-only.
Yes. It reads a dbt project's semantic_models and metrics YAML, or target/manifest.json, into its own semantic layer on a connection you choose. A preview shows what will be created and what is left out, with the reason, before anything is written.
No. It reads the definitions from the files and runs its own queries on your warehouse through its own connection. It does not run dbt models or need dbt Cloud.
The Open Semantic Interchange, developed as Apache Ossie: a vendor-neutral format for datasets, relationships and metrics, so a semantic model can move between tools. ATX AnalyticsTable reads and writes versions 0.1 and 0.2.
Yes. Export writes dbt MetricFlow YAML or an OSI model. What OSI cannot express travels in an ATX extension that the importer reads back, so a model exported to OSI and imported again is unchanged.
This site needs JavaScript for the full experience. Every page above is reachable without it.