Agentic Data Engineering

Your data engineering agent

It builds and maintains data pipelines on the platform you already run.

Try the sandbox Deploy from Your Cloud Marketplace Read how it works

You stay the accountable driver; your data engineering agent works across build and operate, with context carried across Intents and skills and tools, your own added as plugins. Every change passes the gates before it reaches your data platform. — You stay accountable; your data engineering agent works inside a boundary and ships through a gate

Today every pipeline change is a ticket in a data engineer's backlog

53% of a data engineer’s time goes to maintaining existing pipelines (Fivetran, 2026). An agent can take that load if it has somewhere to be wrong on real data that production never sees, and guardrails it cannot reason around.

Demand from AI agents grows around the data team, alongside AI / ML and business intelligence workloads.

Watch the demand grow

What it does

Before building

Onboard your estate

It reads the repository and platform you already run and writes the context every stage reads: vocabulary, decisions, conventions. Agent Readiness findings follow, each with its evidence.

Vibedata Studio Code View showing the NYC taxi onboarding review report, with findings R-1 to R-9, evidence and status. — Review findings link to evidence · Example: NYC taxi data on DuckDB

Vibedata Studio Domain Memory tab showing four NYC taxi records for billable trips, quality verdicts, week boundaries and rate calculations. — Review findings link to evidence · Example: NYC taxi data on DuckDB

Before building

Isolate the change

Every change gets its own branch and workspace, built natively on your platform. Production never sees it, so a wrong attempt is disposable and agents run side by side.

Studio provisions the driver weekly aggregate change on Microsoft Fabric, showing the workspace capacity check and isolated workspace creation. — Studio provisions an isolated workspace · Example: NYC taxi data

Studio shows the isolated branch, conversation and ephemeral DuckDB database for the driver weekly aggregate change in a details drawer over the dimmed chat. — Studio provisions an isolated workspace · Example: NYC taxi data

While building

Spec and build it

Vibedata takes the request to a spec you approve, then designs, plans and builds on subagents, unattended. It asks only what your code and marts leave open.

Studio explores a weekly driver view for the Queens shift-bidding cycle using documentation, model inventory and lineage tools. — A plain request starts exploration of the estate · Example: NYC taxi data on DuckDB

Studio shows the exploration history, an answered weeks-and-rates question and an open Sunday-midnight shift question. — A plain request starts exploration of the estate · Example: NYC taxi data on DuckDB

While testing

Review and verify

A gate runs the change against real data on your compute and records evidence per requirement. You read the evidence and the exceptions, then merge the pull request.

Build summary with Validation evidence for the driver weekly aggregate in Vibedata Studio. — The build summary shows validation evidence · Example: NYC taxi data on DuckDB

Change view showing direct changes, downstream impact and consumers reached by the driver weekly aggregate. — The build summary shows validation evidence · Example: NYC taxi data on DuckDB

After release

Keep it running

Vibedata watches the platform, read-only, files what breaks as a GitHub Issue, and ships the fix as a pull request for a person to merge.

Vibedata Studio Domain automations with a daily health check scheduled at 10:00 GST. — A daily health check runs on a schedule · Example: NYC taxi data on DuckDB

Vibedata Studio Health view showing core.fct_trip with a failing enrichment rejoin test. — A daily health check runs on a schedule · Example: NYC taxi data on DuckDB

After release

Give your agents context

The context Vibedata keeps on your estate answers questions in Vibedata, and grounds the agents your users already work in, with your grain, definitions and decisions.

Claude retrieves the billable metric, its definition and the trip-grain table from Vibedata, gets SQL, then runs it and reports revenue by borough. — Claude uses domain context to report revenue by borough · Example: NYC taxi data · Domain MCP in private preview

ChatGPT retrieves the billable metric, its definition and the trip-grain table from Vibedata, gets SQL, then runs it and reports revenue by borough. — Claude uses domain context to report revenue by borough · Example: NYC taxi data · Domain MCP in private preview

What a coding agent lacks for data work

Isolation

Being wrong is survivable, because a wrong attempt is disposable.

Guardrails

Scoped to the data platform: schema changes, warehouse writes, and the deploy path.

Context

Data engineering context: grain, freshness, and which mart the business reads.

Cross-platform

The same loop, skills and gates on DuckDB, Microsoft Fabric and MotherDuck, built natively on each.

The same team ships more

The same team ships five times more with Vibedata, once it changes what it focuses on and how it builds. We measure it as time to a pull request, and as the human time behind each accepted change.

Where it runs

In production today

DuckDB
Microsoft Fabric
MotherDuck

Design partner open

Databricks
Redshift

Coming next

BigQuery
Oracle
Postgres
Snowflake

Yours, and standard

Runs in your own cloud subscription, under your keys.
Writes dlt and dbt code you own, as your platform's own objects.
Model-tiered per lifecycle stage, on your own model keys and endpoints. The gate decides what ships.

Vibedata, and native per platform

An isolated workspace for every change, built natively on each platform, with the agent working inside it.
Real-data gates on every change, and an agent-proof deploy lock.

See it run

Sign in with GitHub for the hosted sandbox, seeded on DuckDB. Or deploy from your cloud marketplace into your own subscription.

Try the sandbox Deploy from Your Cloud Marketplace Read how it works

Questions

What is Vibedata?

Vibedata is your data engineering agent. It builds and maintains data pipelines on the platform you already run, with the four things a general coding agent lacks for data work: isolation, guardrails, context and cross-platform. It covers the whole lifecycle: onboarding your estate, isolating and building each change, review and verify, keeping it running, and giving your agents context.

Where does it run?

In your own cloud subscription, deployed from your cloud marketplace. There is no multi-tenant SaaS.

Which platforms does it support?

In production today: DuckDB, Microsoft Fabric and MotherDuck. Design partner open: Databricks and Redshift. Coming next: BigQuery, Oracle, Postgres and Snowflake.

How do I try it?

There are two equal ways in: a hosted sandbox seeded on DuckDB that needs no install and no tenant setup, or a deployment from your cloud marketplace into your own subscription. Sandbox sign-in is through GitHub.

Does it read my data?

It reads metadata, code and execution traces, never your rows. Profiling and verification run as queries on your own compute. Agents act under scoped, short-lived credentials, with tamper-evident, immutable audit logging.