Bauplan vs Datastrato Iceberg Lakehouse Platform

Bauplan is a data lakehouse for running Python and SQL pipelines directly on Apache Iceberg tables in a user's object storage, without managing servers or clusters. It uses serverless execution for Python pipelines and supports Python, SQL, and mixed workloads on the same tables. Data branches, isolated runs, atomic publishes, and rollback are intended to safeguard changes to production data. Bauplan also describes workflows where AI agents build pipelines, explore data, and manage a lakehouse through its MCP Server and Skills. Integrations listed include Airflow, Prefect, Dagster, Temporal, DBOS, Jupyter, Marimo, Streamlit, Snowflake, BigQuery, Metabase, Superset, Fivetran, and Airbyte. Bauplan says customer data stays in the customer's S3 bucket and offers PrivateLink or bring-your-own-cloud options. Its free Sandbox uses a shared environment where all data is public, which the company recommends for toy projects and demos rather than company data. Dedicated costs $700 a month after the first 14 days free and includes a dedicated private instance, data in the customer's S3, 10GB memory, and unlimited queries, users, and agents. Enterprise pricing is custom. Bauplan states that it is SOC 2 Type II compliant.

Datastrato Iceberg Lakehouse Platform is a federated metadata catalog for connecting data sources and governing access across clients. Built on Apache Gravitino and Iceberg REST Catalog, it provides a standards-compliant REST catalog with metadata caching and table maintenance for production lakehouses. It can register Iceberg, Hive, Glue, relational databases, Kafka, filesets, and models in place without copying metadata. Spark, Flink, Trino, and Dremio are named Iceberg catalog clients, and the same namespace resolves across BigQuery. Governance includes catalog-level role-based access, row filtering, column masking, tag-based policies, and auditing across engines and agents. S3, GCS, and ADLS are supported with short-lived scoped credentials per principal. Agents can use an MCP metadata interface and GVFS under the same access controls. Per-table compaction, snapshot expiry, and orphan cleanup jobs can be sent to Airflow, Kubernetes, EMR, or Databricks. Enterprise is installed with a Helm chart in Kubernetes and activated by license key. Pricing is on request; the commercial distribution includes support with a defined response commitment and maintained release line.