Bauplan

Bauplan is a serverless execution layer for Python and SQL pipelines working directly on Apache Iceberg tables in your object storage. It is aimed at developers and AI agents who need to change production data with controlled release paths. The central compromise is clear: experimentation is free but public, while private deployment starts at $700 a month.

Overview

Bauplan runs Python, SQL, or mixed workloads on Iceberg tables without asking your team to operate servers or clusters. Its branches, isolated runs, atomic publishes, and rollback are designed for safer pipeline changes: developers can work away from production, validate results, then publish or reverse changes in a controlled manner.

Execution uses serverless function-as-a-service infrastructure, with Bauplan handling execution, isolation, and scaling for Python pipelines. The platform is therefore a focused pipeline execution layer rather than a general analytics interface. It connects to orchestrators including Airflow, Prefect, Dagster, Temporal, and DBOS, and to working environments and destinations such as Jupyter, Marimo, Streamlit, Snowflake, BigQuery, Metabase, Superset, Fivetran, and Airbyte.

Key features

Production data safeguards

Data branches and isolated runs separate development from live tables. Atomic publishes help prevent partial updates, while rollback gives teams a defined way to undo a release. These controls are particularly relevant when a pipeline mistake could propagate into reports, models, or downstream services.

Iceberg and mixed workloads

Bauplan keeps data in customer-managed object storage using Apache Iceberg, an open table format. Python and SQL can operate on the same tables, supporting teams that combine programmatic transformations with relational analysis. External storage also avoids moving customer data into a proprietary repository.

Serverless and AI execution

Function-as-a-service execution removes cluster provisioning from day-to-day pipeline work and lets the service manage isolation and scaling. Bauplan says AI agents can build pipelines, explore data, and manage a lakehouse through its MCP Server and Skills, making the product relevant to teams incorporating agents into engineering workflows.

Security and deployment controls

Bauplan states that it is SOC 2 Type II compliant. Dedicated and Enterprise deployments add options such as single tenancy, PrivateLink, and bring-your-own-cloud. Customer data remains in the customer’s S3 bucket and can be processed through PrivateLink or inside the customer’s VPC with BYOC. Encryption in transit, restricted production access, monitoring, incident response, and retention and disposal policies round out the stated controls.

Pricing

Sandbox — free. This shared Bauplan environment includes public datasets or your own uploads, the full CLI, SDK, and API, and community support on Slack. Every Sandbox data item is public, so it is appropriate for toy projects, demonstrations, and learning—not company data or confidential experiments.

Dedicated — $700 a month. The first 14 days are free. You get a dedicated private instance, data kept in your S3, 10GB memory, unlimited queries, users, and agents, private Slack support, and SOC 2 Type II and GDPR compliance. This is the practical entry point for private production use, but the fixed monthly cost will be excessive for occasional or very small workloads.

Enterprise — custom pricing. Enterprise provides 50GB or more memory, unlimited queries and users, dedicated customer support, and a forward deployed engineer, with SOC 2 Type II and GDPR compliance plus PrivateLink or BYOC.

Platforms

Bauplan offers web and API access. Its external-storage model, CLI, SDK, and API support integration into existing engineering and orchestration environments rather than restricting work to a single browser workspace.

Who it's for

Bauplan fits data-platform developers and AI agents building, validating, and shipping pipelines against production data. It is strongest for teams already using Iceberg and object storage that want serverless execution plus release safeguards. The public Sandbox suits demos and evaluation; private production teams need to justify the Dedicated subscription or negotiate Enterprise terms. It is a poor fit when a public free environment cannot be used even for evaluation, or when you need a broad analytics workspace rather than an execution layer.

Pros and cons

• Pro: Branches, isolated runs, atomic publishes, and rollback provide a coherent safety model for production data changes.
• Pro: Python and SQL share Apache Iceberg tables in your own object storage, reducing dependence on proprietary storage.
• Pro: Broad orchestration, notebook, BI, warehouse, and ingestion integrations support existing data stacks.
• Pro: Dedicated and Enterprise offer private deployment paths, including PrivateLink and BYOC.
• Con: Sandbox data is public, ruling it out for confidential company work.
• Con: Private use starts at $700 a month, a substantial commitment for small teams.
• Con: Enterprise pricing is custom, so larger deployments require a sales process before costs are clear.

Alternatives

For a broader comparison of this category, see Data Lakehouse Platforms and, for tools centered on tracking data changes, Data Version Control Tools.

• Oracle Autonomous AI Lakehouse is a better choice when you want Oracle’s autonomous lakehouse offering with an always-free option and API, Linux, self-hosted, and web access.
• Starburst Data Platform suits teams seeking a free tier with up to three clusters for ad hoc queries, with paid usage priced per contact.
• Starburst Galaxy is the hosted alternative with a free three-cluster tier and a Pro plan starting on request.
• AetherLake is an option for internal, evaluation, or non-commercial use under its free license.
• Amazon SageMaker Autopilot fits pay-as-you-go machine-learning automation rather than Iceberg pipeline execution.
• Apache Hudi is preferable when you want a free, open-source lakehouse platform for Linux or self-hosted deployments.
• Databricks Notebooks offers a free web-based edition with one serverless workspace and limited compute.
• Google Cloud Lakehouse for Apache Iceberg is worth considering for Google Cloud’s Iceberg metadata services and free metadata-storage tier.

Verdict

Choose Bauplan if your developers or AI agents need to run Python and SQL against production Iceberg data with branching, isolated execution, and controlled publishing, while keeping storage in your own cloud. Look elsewhere if public-only evaluation is unacceptable, if $700 a month is beyond your budget, or if you need a general analytics environment instead of a specialized pipeline execution layer.