Pre-release. This repository is public from day one, but nothing in it is ready to install yet. The function catalogue has not been probed against a Fabric engine, and there is no plugin to add. Watch the repository or follow the v0.1 epic if you want to know when that changes.
The T-SQL reference for AI agents, scoped to Microsoft Fabric: the Warehouse and the SQL analytics endpoint of a Lakehouse. Every built-in function is probed against the real engine instead of typed from memory: whether it exists in Fabric, the error it raises if it does not, and the type it returns. System views and procedures come straight from the engine's own catalog.
It is the sibling of dax-for-agents, powerquery-m-for-agents and spark-for-agents, and it follows the same rule: an agent never reads a function name that did not come from the engine.
T-SQL does not describe itself. Built-in scalar functions such as LEN or DATEADD are not rows in
sys.objects, so there is no SHOW FUNCTIONS to export. The honest method has three steps:
- A list of candidates taken from the official T-SQL function documentation.
- Each candidate is probed on a Fabric engine with a minimal
SELECT, throughsp_describe_first_result_set, which compiles the statement and never runs it. The result records whether the name exists, the engine's error if it does not (not supportedis not the same asnot recognized), and the returned type. The first run used a SQL analytics endpoint; a Warehouse run is still to come. - System views and procedures come directly from
sys.all_objectsandsys.all_parameters.
| Piece | Source | Status |
|---|---|---|
Candidate list of built-in functions (catalog/candidates.json) |
official T-SQL function reference (MicrosoftDocs/sql-docs, CC-BY-4.0, pinned commit) |
first version |
Probe results (catalog/probe/functions.json) |
sp_describe_first_result_set on a minimal SELECT, compiled on a Fabric SQL analytics endpoint, never executed |
first run |
System views and procedures (catalog/probe/system_objects.json) |
sys.all_objects + sys.all_parameters on the engine |
first run |
| Fabric traps (unsupported vs SQL Server, types, collation) | drafts in lab/drafting/, each claim with its own probe |
drafts |
tsql-feedback skill |
report a wrong or missing card as a public issue | planned |
- Administering Warehouses, running loads or diagnosing queries. That is the
sqldw-cliskill of the Fabric skills plugin. - Spark SQL. That is spark-for-agents.
- Azure SQL, SQL Server or Fabric SQL database in general. Microsoft's own Agent-Skills summarises Azure SQL from Microsoft Learn; this repo links to it instead of duplicating it.
MIT.