Murakumo
1.1 Quality rules on the canvas, data contracts and a query console

Design the model. Get the whole data platform.

Draw the entities, how they relate and where each column comes from. Choose your warehouse, orchestrator and the rest of the stack. Murakumo writes the DDL, the dbt project, Spark and Polars jobs, tests, data contracts, orchestration, CI and Terraform, as ordinary code in your repository.

$ pip install "murakumo[canvas,verify,duckdb]"
$ murakumo init my_platform
$ murakumo canvas my_platform   # the visual editor

One model, every layer

The model is the single source of truth. Change a column once and every layer, test, contract and pipeline that uses it changes with it, and murakumo verify builds it all on DuckDB to prove it works before it reaches your warehouse.

The model

trip
trip_id key
pickup_zone → zone
fare_amount ≥ 0
zone
zone_id key
borough one of 6

What Murakumo writes

RawLanding tables, loads, audit columns
StagingTyped, renamed, tested, quarantine
IntermediateJoins, history, Data Vault
MartsFacts, dimensions, SCD types
SemanticMetrics and exposures
DDLdbt projectPySparkPolars testsdata contractsDAGsCI Terraformreplication guide

Choose your stack

Every part is a choice in murakumo.yaml or on the canvas. Each warehouse gets SQL in its own dialect, and switching one is a setting, not a rewrite.

Warehouse

DuckDBSnowflakeDatabricksBigQueryRedshiftPostgreSQLFabricSynapse

Extract

PythonLambdaGluedltAirbyteFivetran

Load

BuildSnowpipeAuto LoaderS3 or local landing

Transform

dbtPySparkPolarsIcebergDelta

Orchestrate

AirflowDagsterPrefectDatabricks WorkflowsSnowflake TasksCloud ComposerStep FunctionsAWS GlueGoogle WorkflowsData Factory

Stream

Kafka consumerFlinkSpark StreamingSnowpipe Streaming

Quality

dbt testsGreat ExpectationsSodaODCS contracts

Ship

GitHub ActionsGitLab CITerraformSlim CI

Everything a platform needs, on the canvas

Each feature works from the canvas and from the command line, and the generated code runs without Murakumo.

Data quality you configure

Pick a rule per column or table: not empty, unique, regex, ranges, freshness and more. Each one fails the run, warns, or sets the bad rows aside in a quarantine table.

Quality rules →

Data contracts

A contract per table in the Open Data Contract Standard, checked on every load. Import a contract to start a model from one.

Data contracts →

A query console for every stage

Read-only SQL on the source, what landed, raw, staging, quarantine and marts, and on Parquet, CSV, Iceberg or Delta files in S3, GCS or Azure.

The query console →

Import what you have

Start from a database, a data dictionary, files, an API, protobuf, an existing dbt project or a contract, and keep it in sync as it changes.

Getting metadata in →

Modelling patterns built in

Kimball, Data Vault, SCD types 1, 2, 3, 4 and 6, CDC, masking and row-access policies, and a semantic layer, each from a setting rather than hand-written SQL.

Modelling patterns →

Made for teams

Environments, a schema of your own, reviews of each model change and what it reaches, and CI that tests only what changed.

Working as a team →

No lock-in, by design

Murakumo writes a replication guide with every platform: how to build all of it by hand. Anyone can rebuild the platform, change it outside Murakumo, or stop using Murakumo, and lose nothing.

  • Every account, role and key, and every source connection
  • Every table, model and test, as dbt and as plain SQL
  • Every schedule, pipeline and CI step, and why it is built that way

The replication guide →

$ murakumo guide --check Rebuilding from the guide alone, on DuckDB, then with Murakumo, twice over... The guide rebuilds the project: all 36 tables match a Murakumo build, in rows and values, after each of 2 runs of new files; every test passes, and running it again changes nothing.

Who it is for

Open source under Apache-2.0. AI features stay off until you turn them on.

Consultancies

Stand up a client's platform on the stack they already run, and hand it over with a guide they can follow without you.

In-house data teams

Keep the model, the code and the documentation in step as the business changes, with reviews and environments.

Solo engineers

Go from a data dictionary to a tested, scheduled platform on your laptop with DuckDB, then deploy it to a warehouse.