Semantic models for autogenerated data models, through Snowplow Assistant and the MCP server
Autogenerated data models now come with a semantic model, so AI agents and BI tools have the context to read and query the table correctly. Both are available from Snowplow Assistant and any MCP client connected to the Snowplow MCP server.
Semantic model generation
An autogenerated data model gives you a flat, analysis-ready table for a tracking plan. The warehouse holds the data but not what it means: which rows are which business event, what a column's values stand for, or how to count. Your tracking plan already has that information in event specification names and descriptions, triggers, and instructions.
The semantic model projects that information onto the data model's table. It's an Apache Ossie document, formerly Open Semantic Interchange (OSI), with one dataset for the table, a field for every column, and an event count metric. It's composed from the tracking plan and the stored model configuration.
Convert it to the format your AI agent or BI tool reads with the converters in the Apache Ossie repository, for example Snowflake semantic views, Databricks metric views, Cube, or Omni.
Available through Snowplow Assistant and the MCP server
Until now, autogenerated data models were created in Console or through the API. You can now list, inspect, and create them from Snowplow Assistant or from any MCP client connected to the Snowplow MCP server. Semantic model generation is available only this way for now.
Ask the Assistant to create a data model for a tracking plan. It picks the destination, asks how your tracking is implemented, generates the SQL, and then produces the semantic model for the table you created.
For more information, see: