Run AI-assisted data work
safely in production

Use AI coding assistants from your repo and IDE to write, inspect, and debug real data pipelines. Bauplan turns those changes into isolated, transactional runs with built-in rollback.

AI agents can iterate on code, but not on your data

Code is local and reversible. Data pipelines are not.
‍
Pipelines mutate shared state and failures leave production inconsistent. Traditional data platforms assume slow, manual change.

Bauplan is the execution layer built for fast, AI-generated iteration in production.

The missing execution layer that lets AI
work safely on production data

Work with data the same way you work with code

Everything in Bauplan is code, versioned in your repository and executed from your IDE. AI-generated changes run exactly as written, with no hidden state or manual steps.

Bring your AI coding assistant: we provide the safe execution layer.

QUICK START

Git-style safety for AI agents on production data.

Let AI agents work directly on production data without risk. Runs are isolated, publishes are atomic, and failed or bad changes can be rolled back immediately. Tests and expectations can gate publication before anything reaches your production tables.

LEARN MORE

Use cases

Automate data engineering workflows with simple agent skills,
from assessing the feasibility of a request to building and maintaining production pipelines at scale.

Data transformation pipelines

Build and iterate on data pipelines the same way you build software. Agents and engineers write transformations in code, run them against real data in isolation, inspect results, and iterate quickly before publishing a correct version to production.

Learn more

Safe data ingestion

Ingest new data without breaking downstream systems. Data is written to an isolated branch, validated with quality checks, and published atomically only if it passes, so bad files or schema issues never reach production tables.

Learn more

Debug & fix pipelines

Use your AI agent to diagnose import and pipeline issues, replay failed runs against the exact state that produced them, and propose a fix in a separate data branch. Review and merge when ready. Production stays untouched until you say so.

Learn more

Data exploration and discovery

Inspect schemas, sample rows, and run profiling queries on isolated branches, so engineers and AI agents can understand what is actually in the data before making or proposing changes.

Learn more

Integrations

Integrate with your storage, warehouses, and developer workflow.
Read more about Bauplan integrations in our docs.

Orchestra

A managed orchestration platform that lets data teams build, schedule, and monitor pipelines efficiently within a unified interface.

Metabase

An open-source and enterprise-ready BI tool that enables teams to explore, visualize, and share data insights in real time.

Snowflake

Connect Snowflake to Bauplan to access your data as an Iceberg REST catalog, integrating seamlessly with your object store.

Marimo

A reactive Python notebook and app framework that allows developers to build and share interactive tools entirely in Python.

Streamlit

A Python framework that turns scripts into interactive web apps for data exploration, dashboards, and internal tools.

Jupiter Notebooks

An interactive environment for Python that supports reproducible research, data analysis, and collaborative experimentation.

Temporal

A Python SDK that lets developers build resilient workflows and manage complex asynchronous tasks directly from their code.

Dagster

An orchestrator built for data applications, offering type-safe, observable, and testable pipelines for production workloads.

DBOS

A distributed operating system that provides high reliability, scalability, and consistency for workflow-driven applications.

Airflow

One of the most widely adopted orchestrators in data engineering, designed to programmatically author, schedule, and monitor workflows.

Prefect

A modern workflow orchestration platform built in Python that simplifies task automation, monitoring, and error handling across teams.

See Less

A whole data platform in your repo

Branch, inspect and merge data like code

Bauplan models the state of your data as branches and commits. Create branches, run changes, inspect history, and merge only when tests are passed.

import bauplan
import pandas as pd

client = bauplan.Client()

dev_branch = client.create_branch(
    branch=f"fritz.dev",
    from_ref="main",
)

tables = client.get_tables(ref=dev_branch)
print(tables)

preview = client.query("SELECT * FROM my_table LIMIT 5", ref=dev_branch)
print(preview.to_pandas())

client.merge_branch(source_ref=dev_branch, into_branch="main")

import bauplan
from bauplan.standard_expectations import expect_column_no_nulls

@bauplan.model()
@bauplan.python(pip={"pandas": "2.2.0"})
def clean_data(data=bauplan.Model("my_data")):
    df = data.to_pandas()
    return df.dropna()

@bauplan.expectation()
@bauplan.python("3.10")
def test_clean_data(data=bauplan.Model("clean_data", columns=["id"])):
    return expect_column_no_nulls(data, "id")

Native Python execution. No infrastructure to manage.

Pipelines are ordinary Python and SQL functions. Declare environments and quality checks in code. Execution is managed by the platform.

One control loop for humans and agents

A few predictable primitives for developer and AI agents. Every workflow follows the same loop: branch → run → inspect → merge.

import bauplan

client = bauplan.Client()
dev_branch = f"fritz.dev_branch"

client.create_branch(branch=dev_branch, from_ref="main")

state = client.run(project_dir="./my_project", ref=dev_branch)

if state.job_status.lower() != "success":
    raise Exception(f"{state.job_id} failed: {state.job_status}")

client.merge_branch(source_ref=dev_branch, into_branch="main")

Latest from our blog

AI-first data engineering, Git-for-data semantics, and serverless execution over Iceberg

READ ALL POSTS

Bauplan’s MCP Server

The First Step Towards the Agentic Lakehouse

Ciro Greco

Data engineer agents

From Prompt to Pipeline: Cloud-Native Agents for Data Transformation and ETL

Ciro Greco, Jacopo Tagliabue and Federico Bianchi

Everything-as-Python

From notebook to prod with Bauplan and marimo

Ciro Greco

Git for Data: Formal Semantics of Branching, Merging, and Rollbacks (Part 1)

How formal methods help ensure safe, reproducible workflows in data lakehouses

Ciro Greco and Jacopo Tagliabue

Python Over Data Lakes

Functions, Declarative Environments and Data Management

Ciro Greco

Hello Bauplan

Bauplan is a serverless data platform that treats pipelines, models, and tables like software.

Ciro Greco

FAQs

What if I already have Databricks or Snowflake?

Great! Bauplan is built to be fully interoperable. All the tables produced with Bauplan are persisted as Iceberg tables in your S3, making them accessible to any engine and catalog that supports Iceberg. Our clients use Bauplan together with Databricks, Snowflake, Trino, AWS Athena and AWS Glue, Kafka, Sagemaker, etc.

What does Bauplan replace in my AWS data stack?

Bauplan consolidates pipeline execution and data versioning into one workflow: branch, run, validate, merge. You can keep S3 and your orchestrator; you remove a lot of cluster complexity and glue. For example, an Airflow DAG that spins up an EMR cluster, submits Spark steps, then runs an AWS Glue crawler to refresh the Glue Data Catalog before triggering downstream jobs becomes: Airflow triggers a Bauplan run on an isolated branch that writes Iceberg tables directly to S3.

How do you keep my data secure?

Your data stays in your own S3 bucket at all times. Bauplan processes it securely using either Private Link (connecting your S3 to your dedicated single-tenant environment) or entirely within your own VPC using Bring Your Own Cloud (BYOC).

Do I need to learn a new data framework or DSL?

No. Bauplan is just Python (and SQL for queries). That why your AI assistant can immediately write Bauplan code with no problem.

What does Git-for-Data mean?

Bauplan allows you to use git abstractions like branches, commits and merges to work with your data. You can create data branches for your data lake to isolate data changes safely and enable experimentation without affecting production, and use commits to time-travel to previous versions of your data, code and environments in one line of code. All this, while ensuring transactional consistency and integrity across branches, updates, merges, and queries. Learn more.

Run AI-assisted data worksafely in production

AI agents can iterate on code, but not on your data

The missing execution layer that lets AIwork safely on production data

Work with data the same way you work with code

Git-style safety for AI agents on production data.

Use cases

Data transformation pipelines

Safe data ingestion

Debug & fix pipelines

Data exploration and discovery

Integrations

Orchestra

Metabase

Snowflake

Marimo

Streamlit

Jupiter Notebooks

Temporal

Dagster

DBOS

Airflow

Prefect

A whole data platform in your repo

Branch, inspect and merge data like code

Native Python execution. No infrastructure to manage.

One control loop for humans and agents

Latest from our blog

Bauplan’s MCP Server

Data engineer agents

Everything-as-Python

Git for Data: Formal Semantics of Branching, Merging, and Rollbacks (Part 1)

Python Over Data Lakes

Hello Bauplan

FAQs

Run AI-assisted data work
safely in production

The missing execution layer that lets AI
work safely on production data