One project. One set of operations.

Weaver reads declarations in Python, SQL and TMDL, then uses them to build and operate Fabric items.

Source files describe the project; installed state records what has been deployed.

Native declarations

T-SQL

A comment above an ordinary query. Identity, keys and schema sit in native SQL metadata; the SELECT stays ordinary T-SQL.

Warehouse/Curated/Parcel.Event.sql
/*
Table ID: Parcel.Event
Description: Parcel events.
Lineage: Example input.
Primary key: parcel_id
Schema:
  parcel_id: varchar(20)
*/
select 'P-1001' as parcel_id;

Python

A class with a read() method. Declare the Table in a module docstring, then return a Spark DataFrame using ordinary Python.

Tables/Parcel__Event.py
"""Table ID: Parcel.Event
Description: Parcel events.
Lineage: Example input.
Schema:
  parcel_id: string
"""
from weaver import Table

class Parcel__Event(Table):
    def read(self):
        return self.spark.createDataFrame(
            [("P-1001",)], "parcel_id string"
        )

TMDL

A native annotation. Weaver applies modelling conventions through overlays, leaving your authored model files unchanged.

Parcel.tmdl
model Model
    annotation 'Weaver.AutoHideForeignKeys' = true

Declarations identify each object and specify its keys, schema or modelling rules.

File paths and naming conventions supply the rest of the structure.

Native authoring →

The Catalogue brings everything together

The Catalogue records definitions, dependencies and run outcomes for registered managed items.

Native definitionsPython · SQL · TMDL
Build
Installed Fabric estateLakehouses · Warehouses
Semantic Models · Reports
Records definitions
and installed bindings
Operate on the
installed estate

Catalogue

Fabric Warehouse

Installed definitions

  • Definitions
  • Signatures
  • Dependencies
  • Metadata & bindings

Operational evidence

  • Bookmarks
  • Logs
  • Statistics
  • Test results
Read stateLoad / Test
results
Load · Test · HealthUse the recorded estate
Build installs Fabric objects and records their definitions and bindings. Load and Test use that state, operate on the installed estate and record outcomes. Health reads the evidence.

The Catalogue is an ordinary Fabric Warehouse. Its tables hold the shared state used by Build, Load, Test and Health.

Inside the Catalogue →

The diagram shows a Catalogue-managed estate. Standalone Power BI deployment can also work without a Catalogue.

From declarations to execution

Dependencies
Coordinate related work across Fabric items.
Incremental processing
Bookmarks, row changes and deletes.
Concurrency
Run independent work together.
Fault tolerance
Record failures without losing completed work.
Tests & Assumptions
Check the data, not just the execution.
Modelling conventions
Apply standards across Semantic Models.

Weaver generates loading machinery, resolves supported dependencies and applies declared validation and modelling rules.

Python imports and supported T-SQL references establish dependencies; Spark SQL needs explicit declarations. Incremental policies use bookmarks and authored change/delete logic, not automatic detection of every source change.

One operating model

  1. BUILDInstall the definitions.
  2. LOADProcess data and refresh models.
  3. TESTVerify the results.
  4. HEALTHSee what needs attention.

MIRROR — Develop against existing data without copying everything.

WIPE — Destructively reset selected physical destinations.

Reports are Build-only, with no independent Load or Test. Mirror borrows Lakehouse/Warehouse data. Wipe includes unmanaged contents within selected Lakehouses/Warehouses and normally the resolved Catalogue; named Semantic Model resets retain it. Reports cannot be wiped.

Operations and their boundaries →

One project. Different environments.

Development

Warehouse/CuratedCurated_Dev

Production

Warehouse/CuratedCurated

Same source. Different physical destinations.

Workspace configuration selects destinations; Build records the installed bindings. Changing configuration alone doesn't move installed objects or rewrite arbitrary external connections.

Run Weaver from desktop Python, inside a Fabric Notebook, or both.

Work across environments →

Source files and installed work

This separation distinguishes what the project declares from what Fabric has installed.

Recorded outcomes show what has actually happened since.

Get started →Built for scale →