🆕 Mesh Datasets
Learn how Mesh Datasets enable you to join, query, and activate data across multiple sources in Activations Workbench — all within a single dataset.
Mesh Datasets have been deprecated and are no longer available in new Fivetran accounts:
- July 1, 2026: Existing datasets become read-only; no new datasets can be created.
- August 1, 2026: Datasets will be deleted.
You can work with support to migrate the sources of their mesh datasets to Fivetran connections, and potentially dbt models.
Activations makes it it easy to connect or create Datasets from many different sources. Any SQL source can easily join data within that source. But what about joining across sources? Now you can use Mesh Datasets to write SQL that spans datasets built on any SQL source together.
Writing Mesh SQL
To support querying across SQL sources, Activations provides a virtual query engine that has access to a magic datasets.* schema. Every SQL dataset you've created in Activations Workbench is automatically available as a virtual table in the schema. For example, if you have a dataset called VIP Users, you'll find that datasets.vip_users is now queryable when writing the SQL for your Mesh Dataset.
Mesh SQL is regular Postgres-compatible SQL syntax so any standard Postgres functions should work as expected. This is true even if your SQL sources are Snowflake, BigQuery, or some other service. Querying a Mesh Dataset primarily runs in Activations' virtual query environment. The query engine automatically determines when a source-specific dataset needs to be executed and automatically pushed down to the appropriate source using the SQL defined in that referenced dataset.
Using Mesh SQL AI Assistant
Mesh Datasets also support using AI to generate SQL. Give your Mesh Dataset a description and then press the Generate SQL button. The LLM will use the metadata about all of your datasets to generate what it thinks is the most appropriate SQL.
LLMs are magic but they're not perfect! Activations recommends using the generated SQL as a starting point for your Mesh Dataset rather than a perfect query every time. Make sure to review the logic and the results before you sync it to any destinations.
For more information on how we work with LLM providers, see our privacy docs on AI Integration & Data Privacy.
Supported Sources
All SQL Sources including Snowflake, BigQuery, Databricks, Redshift, as well as Census Store are supported and can be joined in Mesh Datasets.
Streaming Sources such as Kafka and HTTP are not supported. Google Sheets, S3, and file Sources are also not supported.
Warehouse writeback is not a supported feature with Mesh Datasets.