AetherLake

AetherLake is a self-hosted lakehouse stack managed through a web control panel. It suits teams that can run Kubernetes and want storage, streaming, SQL, orchestration, and BI in one environment. Its breadth is useful, but production deployment calls for careful setup and a separate license for commercial use.

Overview

AetherLake combines object storage, a governance catalog, federated SQL, stream processing, orchestration, BI, and identity. Its components include MinIO, Trino, Apache Polaris, Kafka, Flink, dbt, Airflow, Superset, Spark, Milvus, Prometheus, Grafana, PostgreSQL, and Keycloak. That range gives data teams a shared foundation for several workflows, while leaving them to deploy and operate a substantial set of services.

The web control panel brings together platform status, pod health, resource use, SQL, catalog, Kafka, Flink, dbt, and observability views. It can restart services and provides workspaces for dbt, Kafka, Flink SQL, Trino queries, and Iceberg catalogs. English and Turkish are supported.

Key features

Streaming and lakehouse workflows

Kafka and Flink are enabled by default. Project examples show Kafka data flowing into Iceberg tables and Trino querying Kafka topics. The dbt examples use Bronze, Silver, and Gold stages with Kafka, Flink, Parquet, Iceberg, Superset, and Trino. These workflows make AetherLake a stronger fit for teams building streaming and analytical pipelines together than for those seeking only a query service.

Identity, security, and telemetry

Keycloak provides OIDC single sign-on across services, with role names for data administrators, scientists, and engineers. Credentials are generated randomly for each installation. Before exposing a cluster, the project advises replacing placeholder values, setting production secrets and MCP server Airflow authentication, and configuring real ingress TLS certificates. The external Kafka listener uses TLS and SCRAM-SHA-512 authentication.

An anonymous heartbeat tracks installation counts, detected cloud environments, and component health. It excludes personal information, schemas, and dataset contents, and can be disabled. Teams should still decide whether this telemetry fits their operating policy.

AI assistant access

The MCP server can give Claude Desktop and Cursor access to platform status, service logs and restarts, Trino queries, Polaris catalogs, and Airflow DAGs. That offers a way to work with platform operations and data tools through supported assistants; it does not remove the need to manage the underlying cluster and credentials.

Pricing

AetherLake is free for internal, evaluation, development, testing, and non-commercial use. The free plan is a practical route for teams assessing the stack or running it for non-commercial work, but commercial use requires a separate license.

Plan	Price	Terms
Internal, evaluation, and non-commercial use	free	Free for internal, evaluation, development, testing, and non-commercial use; commercial use requires a separate license.
Commercial license	Custom pricing	Case-by-case licensing; may include revenue-sharing terms.

The license permits internal use, modification, and self-hosting under BUSL-1.1. Commercial hosted or managed offerings require a license, and releases convert to Apache-2.0 after four years. Organizations planning a commercial deployment should resolve licensing terms before building around the platform.

Platforms

AetherLake is self-hosted and web-based. The quick start requires Kubernetes, Helm 3.12 or later, kubectl, Docker, and an NGINX ingress controller, so it assumes an existing capacity to run and maintain a Kubernetes environment. Deployment materials cover AWS, Azure, and Google Cloud Platform.

Who it's for

AetherLake makes sense for data engineering teams that want to combine a catalog, SQL analytics, streaming ingestion, transformations, orchestration, and BI in a self-managed environment. Its default Kafka and Flink setup, Iceberg examples, and service workspaces support teams building connected data pipelines. It is a poor fit for readers who need a managed service, lack Kubernetes expertise, or want to avoid production security and licensing work.

Pros and cons

• Broad stack in one control panel: storage, catalog, SQL, streaming, orchestration, BI, identity, and observability sit within one project.
• Useful streaming examples: Kafka and Flink are enabled by default, with documented paths into Iceberg and Trino.
• Operational controls and assistant access: the panel supports health and resource views and service restarts; the MCP server exposes selected platform and data operations to Claude Desktop and Cursor.
• Substantial deployment burden: Kubernetes and several supporting tools are prerequisites, and production exposure requires replacing placeholders and configuring secrets and TLS.
• Commercial terms need planning: commercial use requires a separate, case-by-case license that may include revenue sharing.
• Telemetry is enabled as an option to review: the anonymous heartbeat can be disabled, but teams should account for it in their deployment decisions.

Alternatives

For a broader comparison, see Data Lakehouse Platforms.

• Oracle Autonomous AI Lakehouse is worth considering for readers who want a freemium lakehouse with an always-free plan subject to capacity limits and a free trial.
• Starburst Galaxy offers a free plan capped at three clusters for ad hoc queries, plus a free trial; choose it when that query-focused hosted option better fits than operating AetherLake's Kubernetes stack.
• Databricks Notebooks has a free edition with one serverless workspace and limited compute and usage, making it an option for readers who want a constrained serverless notebook workspace.
• Starburst Data Platform has a free tier for up to three clusters in standard ad hoc execution mode, plus a free trial, for readers comparing another platform with a defined free cluster allowance.
• Amazon SageMaker Autopilot is a paid option with a free trial and pay-as-you-go pricing, for readers looking for that model instead of AetherLake's self-hosted stack.
• Bauplan provides a free public sandbox with CLI, SDK, and API access, suitable for readers who want to explore in a shared environment.
• IOMETE offers a free self-hosted plan capped at 100 vCPUs and an enterprise plan, for readers wanting a stated compute limit and self-hosted alternative.
• LavaLake has a free Core plan for one machine and up to 16 vCPU, with a no-time-limit term, for readers whose workload fits those single-machine limits.

Verdict

Choose AetherLake if your team can operate Kubernetes and wants one self-hosted environment spanning streaming, lakehouse tables, SQL, orchestration, and BI. Its integrated components and control panel are the main draw. Look elsewhere if you need managed operations or cannot take on production hardening and case-by-case commercial licensing.