# Platform profile

> In Emolysis, a platform profile is a structured description of a data platform that the agent loads when you connect a workspace.

In Emolysis, a platform profile is a structured description of a data platform that the agent loads when you connect a workspace. It records the dialect, table format, merge idiom, incremental tools, deployment unit, compute options, sign-in method, known traps and how to sample data.

For example, the Databricks profile records that generated code uses Databricks SQL and PySpark, that tables are [Delta](https://emolysis.com/glossary/delta-lake) tables governed by [Unity Catalog](https://emolysis.com/glossary/unity-catalog), that changes are applied with `MERGE INTO`, that incremental loads can use Auto Loader and Lakeflow Declarative Pipelines, that a deploy produces a [Databricks Asset Bundle](https://emolysis.com/glossary/asset-bundle), and that work runs on jobs and SQL warehouses. Its known traps include differences between serverless and classic compute and missing Unity Catalog grants. The drafted Snowflake profile records a different value for nearly every field, from `MERGE` fed by streams to identifiers that fold to uppercase.

Profiles exist because the same pipeline shape needs different code on each platform. A merge that is correct on one platform can be slow or invalid on another, a deployment unit on one has no equal on another, and each platform has its own ways to fail, such as missing grants or case rules for names. Writing those differences down once, as data, is more reliable than hoping a general model remembers them each time.

Every file the agent generates is tagged with the platform and profile version it was written for, so it cannot be deployed to another platform without an explicit translation step. See also [coding agent](https://emolysis.com/glossary/coding-agent), [merge](https://emolysis.com/glossary/merge-upsert) and the [Databricks platform page](https://emolysis.com/platforms/databricks).

Profiles are versioned, because platforms change: a new version can add a trap or a tool, and existing files keep the version they were written with until they are updated. See also [template slot](https://emolysis.com/glossary/template-slot).

With it, the agent writes code the way that platform expects, only that platform's templates are offered, and lint rules check for the platform's known traps. See [Platform profiles](https://emolysis.com/docs/platform-profiles) for the fields and [Connecting Databricks](https://emolysis.com/docs/connect-databricks) for the first platform. Compare [template slot](https://emolysis.com/glossary/template-slot).

---

Page: https://emolysis.com/glossary/platform-profile. Emolysis is a product of Bandpey GmbH, Munich, Germany. It is in private development, and nothing is released yet.