lakeFS

Overview

lakeFS gives object storage the things Git gives code: branches, commits, merges, tags and reverts. It sits in front of S3, Azure Blob or Google Cloud Storage and adds a versioning layer, so a data lake can be changed on a branch and promoted atomically.

The mechanism that makes this practical is that branches are metadata operations, not copies. Creating an isolated environment over production-scale data costs nothing in storage, which is what teams use it for: test a pipeline against real data, validate the result, then merge.

It does not run pipelines. Orchestration stays in Airflow, Dagster or whatever you already use, and lakeFS is driven through the S3-compatible API, a CLI, REST API or SDKs.

Quality gates come from hooks bound to versioning events. A pre-merge hook can check schemas, formats or PII and abort the merge if it fails, which is how write-audit-publish is implemented in practice. Hooks run as embedded Lua, a webhook, or an Airflow DAG.

Licensing needs care. lakeFS Community is Business Source Licence 1.1, source-available rather than open source, and production use is limited to internal purposes; older pages that say Apache 2.0 are out of date. Enterprise adds role-based access, single sign-on, auditing, mounts and an Iceberg REST catalogue.

Treeverse acquired the DVC project in late 2025, and the 2026 releases add a datasets layer, multi-tenancy, auditing into Iceberg tables and object-level access control.

Key features and capabilities

The same headings are used for every dataops & ci/cd entry, so two tools can be read side by side.

What it does for data teams

• Git-like branching, committing and merging over object storage
• Isolated environments over production data without copying it
• Atomic promotion, and rollback to recover from bad data
• Does not run pipelines; it gives them a versioned substrate

Pipelines and automation

• No CI runner of its own; orchestration stays in Airflow, Airbyte, Kubeflow and similar
• Driven through the S3-compatible API, CLI, REST API and SDKs for Python, Spark and R
• Automation is expressed as hooks on versioning events rather than YAML jobs

Testing and change checks

• Hooks on commit, merge, branch, tag, revert and cherry-pick events
• Pre-merge hooks abort the operation on failure, gating bad data
• Hook types are embedded Lua, webhooks, or triggering an Airflow DAG
• No value-level table diffing; that is a different category of tool

Environments and isolation

• Zero-copy branches give isolated development and test environments
• Promotion between stages is an atomic merge; tags name specific commits
• Enterprise adds branch and object lifecycle management for cleaning up
• Multi-tenancy creates isolated environments within one deployment

Approvals, secrets and audit

• Role-based access, single sign-on and auditing are Enterprise-only
• Auditing streams operations into Iceberg tables for querying
• Object-level attribute-based access control below repository level
• Pre-merge hooks act as the protected-branch equivalent

Integrations

• Object stores: S3, Azure Blob, Google Cloud Storage, MinIO and Ceph
• Engines: Spark, Trino, DuckDB, Databricks, Athena, Dremio and Delta Lake
• Catalogues: AWS Glue and Unity Catalog; AI tools including MLflow and SageMaker
• No dbt or Snowflake integration page in the current docs

How it runs

• lakeFS Cloud, fully managed on AWS, Azure or Google Cloud with private link
• Self-managed as binaries, Docker images or a Kubernetes Helm chart
• Enterprise covers on-premises, government clouds and air-gapped environments
• Community is BUSL 1.1 and self-hosted only

Pricing

Price on request

Community is free to use under the Business Source Licence, self-hosted, with no role-based access, single sign-on or auditing. Enterprise, whether on lakeFS Cloud or self-hosted, is quote-only: no tiers, prices or trial length are published. A trial is offered through a contact form.

Vendor pricing page →

Demos and videos

• lakeFS Iceberg integration: environment-wide data version control Product demo · YouTube
• Tutorial: getting started with lakeFS Tutorial · YouTube

About Treeverse

lakeFS is built by Treeverse, founded in 2020 by Oz Katz and Einat Orr; the company does not publish a head office and describes a distributed team. It has raised $43m in total, including a $20m growth round in July 2025 led by Maor Investments with Dell Technologies Capital, Norwest and Zeev Ventures. The Community edition is BUSL 1.1 with an Apache 2.0 change licence after four years; Enterprise is commercially licensed. Named users include Volvo, Bosch and Lockheed Martin.

Founded 2020 · lakefs.io