# Sources

> SQL Server and PostgreSQL as sources: Emolysis reads from them and never deploys to them. What that means, the shapes they feed and what is not decided.

Status: PostgreSQL is in development as a source, tested against local PostgreSQL servers; managed services are not verified. SQL Server is planned as a source. A source is a system Emolysis reads from, not one it deploys to. Landing a source's data on your production platform, Databricks first, then Snowflake and Microsoft Fabric, is planned. The platform pages cover each one: [SQL Server](https://emolysis.com/platforms/sql-server) and [PostgreSQL](https://emolysis.com/platforms/postgresql).

## Source or platform

| Role     | What Emolysis does                                                         | Examples                                |
| -------- | -------------------------------------------------------------------------- | --------------------------------------- |
| Platform | Writes code for it, tests it on DuckDB first, and deploys to it            | Databricks, Snowflake, Microsoft Fabric |
| Source   | Reads schemas and samples from it so pipelines can land the data elsewhere | SQL Server, PostgreSQL                  |
| Local    | Runs your development loop and the golden tests                            | DuckDB                                  |

## What a source profile says

Sources use the same profile format as platforms, so the agent knows how to read them correctly. See [Platform profiles](https://emolysis.com/docs/platform-profiles).

- **SQL Server:** T-SQL, including bracket-quoted identifiers. Types that need care include `datetime` versus `datetime2`, `money`, `nvarchar`, `uniqueidentifier` and `rowversion`. Change options include Change Data Capture, Change Tracking, and `rowversion` or updated-at watermarks.
- **PostgreSQL:** PostgreSQL SQL, including quoted, case-sensitive identifiers. Types that need care include `jsonb`, arrays, `numeric` without a declared scale, and `timestamp` versus `timestamptz`. Change options include logical replication, and updated-at or sequence watermarks.

Which change options Emolysis supports first is not decided.

## Which shapes read from a source

Three [shapes](https://emolysis.com/docs/shapes) start at a source: [Ingest (batch)](https://emolysis.com/shapes/ingest-batch) for a full or periodic extract, [Incremental append](https://emolysis.com/shapes/incremental-append) for append-only tables with a watermark column, and [CDC merge](https://emolysis.com/shapes/cdc-merge) for tables whose rows change or are deleted. Which one fits depends on whether the source can tell you what changed. See [change data capture](https://emolysis.com/glossary/change-data-capture) and [watermark](https://emolysis.com/glossary/watermark).

## What you do

You add a source in a project and enter its connection details. Credentials are kept in your machine's credential store, never in a project file. For development, Emolysis samples a slice of the source into DuckDB on your machine, so the pipeline can be tested without touching the source again. That sample read is a query on the source system.

## Not decided yet

- How production loads from a source run, and where.
- Which change options come first for each source.

## Where to start

The first platform to read about is [Databricks](https://emolysis.com/docs/connect-databricks).

---

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