Transformed tables
Robot Data Engineer cleans, dedupes and joins raw tables into modeled tables, built as Metabase transforms, so every question starts from clean data.
Robot Data Engineer · Beta
Generate your data pipeline with an AI data expert
Robot Data Engineer takes you from raw data to Data Apps and dashboards. It profiles your database and builds what your analytics runs on: transformed tables, a semantic layer of metrics, measures and segments, and builds reports on top of a verified foundation. It works from your input, and when it isn't sure how your business works, it asks you.
Run this in your terminal
$ npm install -g @metabase/rde
$ rde init
Needs macOS or Linux, Node.js, and Docker or Java 21.
robot-data-engineer — session
> Build a semantic layer and core business dashboards from our warehouse.
✓ Profiling schema — 1,875 tables across 6 domains
✓ Writing 34 transforms + 226 tests
✓ Defining 18 metrics, 9 segments in the semantic layer
✓ Building 6 dashboards on the curated tables
✓ Pull request #142 opened
+2,410 −0
· awaiting review
Trusted by 100,000+ companies
Data engineering doesn't have to be the bottleneck
AI has made analytics creation easier than ever. That puts more demand on tidying up the raw data, modeling it, and defining metrics to make sure it's accurate and trustworthy. Generic agents can write decent code, but they don't know your data.
Semantic layer
It defines your metrics, measures and segments, with descriptions, in Metabase's semantic layer in Data Studio. Every dashboard and AI answer uses the same definition of revenue or churn, and Robot Data Engineer builds on those definitions when you ask for more.
Analytics, Data Apps, and dashboards
It builds Data Apps and dashboards on the semantic layer, so the numbers match on every dashboard, for whichever area of the business you point it at.
How Robot Data Engineer works
Describe what you need and let Robot Data Engineer build it. You can review and approve before anything goes live.
- Set up. Run
rde init. It installs the Metabase CLI and the Robot Data Engineer skills for your coding agent. - Describe. Give it a plain language prompt, like "build a sales pipeline dashboard from our Stripe data." More context gives better results.
- Explore. Robot Data Engineer reads what's already in your Metabase, like existing tables, metrics and naming conventions, and profiles the raw tables the request touches. Then it proposes what it will build. If something is ambiguous, like which of four billing intervals is right, it asks you.
- Build. It writes the transforms and their tests. Then it defines the metrics and builds the dashboard on them, so the same number is never calculated two ways.
- Review. You can decide to add an approval gate before anything hits production. If you use remote sync, it arrives as a branch and pull request.
Isolated, verified, and reviewed
Generic coding agents write code and leave the checking to you. Robot Data Engineer builds away from production and verifies its work before anything ships.
Isolated from production
Robot Data Engineer builds in a sandboxed environment, separate from the tables and dashboards your team relies on.
Verified as it builds
Every transform gets tests, and a table isn't built until they pass. When a definition changes, it reruns the tests and checks every dashboard that depends on it.
Checked against numbers you trust
Give it a reference, like last quarter's revenue from finance, and it reconciles its numbers and explains any gap.
You decide what ships
Add an approval step before anything reaches production, or have changes arrive as a pull request. Approved transforms and definitions move to production through Git. Read about agent-driven development
Speed up, no matter where you're starting from
Data engineering for teams that don't have one, or more specialist muscle for teams that do.
New setup
A new Metabase instance with raw data and nothing built yet. Robot Data Engineer builds the tables, metrics and first dashboards for each area of the business.
Adding a database
Metabase is already running and a new database needs modeling. Robot Data Engineer follows your existing naming and structure instead of inventing its own.
Migrating from legacy tools
Your business logic and transformations already live in other tools. Robot Data Engineer reads that logic, flags the gaps, and rebuilds it as transforms and definitions in Metabase.
Changing what exists
A definition needs to change. Because each definition lives once in the semantic layer, Robot Data Engineer finds it, shows the before and after, reruns the tests, and checks every dashboard that depends on it.
Frequently asked questions
What is Metabase Robot Data Engineer? 
Metabase Robot Data Engineer is an AI data engineer that turns raw warehouse data into tested transforms, a semantic layer of metrics and measures, Data Apps and official dashboards in Metabase. It runs in your coding agent, like Claude Code, Codex or Cursor, and makes its changes through the Metabase CLI. You can require approval before anything it builds reaches production.
What is an AI data engineer?
An AI data engineer is an AI agent that does data engineering work: profiling raw tables, writing and testing transformations, and defining metrics. Metabase Robot Data Engineer is an AI data engineer built for Metabase. It follows data engineering playbooks, and you can review its work before it reaches production.
What is a semantic layer, and why does Robot Data Engineer build one?
A semantic layer is a shared set of definitions for the numbers a business tracks, like revenue, active customers or churn. In Metabase, those definitions are metrics, measures and segments, managed in Data Studio. Robot Data Engineer builds the semantic layer so every dashboard and AI answer calculates a number the same way, and so it can build on its own work later. See metrics in Metabase.
How is Robot Data Engineer different from using Claude Code or Cursor on my own?
A generic coding agent can write SQL, but it doesn't know your Metabase data model, your existing metrics, or how to test and ship its work. Robot Data Engineer gives your agent the Metabase CLI and a set of data engineering playbooks, it's a specialized agent.
How does Robot Data Engineer make sure AI-built data is accurate?
Robot Data Engineer writes a test for every rule in a transform, and a table isn't built until its tests pass. Each metric is defined once, so dashboards never calculate the same number two ways. When a definition is ambiguous, it stops and asks.
Can AI build a semantic layer?
Yes. Metabase Robot Data Engineer profiles your database and builds a semantic layer from it: metrics, measures and segments with descriptions, defined once in Metabase. Dashboards and Metabase's AI features then use those same definitions.
Do I need a data engineer to use Robot Data Engineer?
Robot Data Engineer is built for teams without a data engineer and for data teams that need more capacity. It works out what it can from your data and asks you about the rest, like how you define revenue or which records count.
Can Robot Data Engineer help migrate from another data or BI tool?
Yes. Robot Data Engineer reads the business logic in your existing tools, flags the gaps, and rebuilds it as transforms and definitions in Metabase.
What access does Robot Data Engineer need to my data warehouse?
Robot Data Engineer has no access of its own. It works through the Metabase CLI with an API key, so it can only do what that key's permissions allow.
Try Robot Data Engineer
Two commands set up a local Metabase to build in, the Metabase CLI, and the Robot Data Engineer skills for your coding agent.
Run this in your terminal
$ npm install -g @metabase/rde
$ rde init
Needs macOS or Linux, Node.js, and Docker or Java 21.