Skip to content

Analytics engineering that explains itself

OpenDataSuite (ODS) reads a dbt project's compiled artifacts, answers questions about them, and runs dbt on only what needs to run. Every answer comes with the evidence behind it.

Open source (Apache-2.0) · written in Rust · no account, no telemetry · pre-alpha

What it does

  • Build only what changed


    ods state build runs dbt on the models whose code or upstream data changed, reports each node's time and rows, and explains every decision.

    Getting started

  • A read-only dashboard


    ods serve shows the plan, the runs with each node's stats, lineage coloured by what the next run does, and the catalog.

    The dashboard

  • Column-level lineage


    Where every column comes from and what uses it, down to joins, filters and aggregations, with an offline explorer.

    Column-level lineage

  • Change impact


    Which models must run when a column changes, and which can be skipped because they don't read it.

    How impact is decided

  • Entity-relationship diagrams


    Keys and relationships from your tests, constraints and joins, each labelled with how it's known.

    ERDs

  • An MCP server for AI agents


    The same engines as read-only tools for agents, including help writing SQL for people new to a project.

    MCP

See it run

After an edit to one staging model, ods state build builds the 8 nodes that depend on it, reuses the other 6, and says why each one ran:

ods state build after an edit to stg_orders: 8 nodes build, 6 are reused, with each node's time, rows and reason

Try it in five minutes

pip install opendatasuite      # or: brew install buchochelliq-labs/tap/ods
cd my-dbt-project && dbt compile
ods lineage view --open        # explore the lineage, offline
ods state build                # build what needs it, and record the run
ods serve                      # the dashboard, on http://127.0.0.1:8765/

Getting started walks through each step on a demo project, and Install covers Homebrew, cargo-binstall and direct downloads.

A tour of ods serve: Home, the Runs page and a partial run's nodes

How it works

  • Conservative


    When ODS can't tell, it assumes the safer answer: it marks a model as affected, not skipped, and labels a relationship as a guess, not a fact.

  • Explainable


    Every decision carries a reason chain and evidence, as text and as JSON.

  • Provider-neutral


    dbt is the first project format and Databricks the first warehouse. Neither is built into the core. Principles

  • Open


    ODS works with dbt through its public artifact formats, and is built on open-source libraries. Legal and trademarks

Memo, the ODS mascot Tui, Memo's daughter

ODS is looked after by Memo and her daughter Tui. Meet the gang →

Status

ODS is pre-alpha. Commands, flags, output and file formats change without notice, and results can be incomplete or wrong. Use it to explore and to give feedback, and don't rely on it for production decisions yet. The roadmap says what comes next.

ODS is an independent open-source project. It is not affiliated with, endorsed by or sponsored by dbt Labs, Databricks or any other vendor named on this site.