Cube

Cube Core is a standalone semantic layer. Metrics, dimensions, joins and access rules are defined once in code and then exposed through SQL, REST and GraphQL to anything downstream, which ends the habit of re-implementing the same business logic inside every reporting tool that touches the warehouse.

It is headless and ships no interface of its own, leaving the analytics experience to you. It connects to SQL data sources including Snowflake, Databricks and BigQuery, query engines such as Presto and Amazon Athena, and application databases like Postgres. A built-in relational caching engine gives sub-second latency and high concurrency for API requests.

The case for a separate layer comes down to a few points.

• Reusable definitions: most semantic layers are proprietary and welded to one BI platform, so the model cannot travel.
• Standard APIs: internal BI, embedded analytics and AI agents all read the same model over the same interfaces.
• Governed foundation: access rules sit with the metrics instead of being rebuilt in each downstream tool.
• Caching: the relational cache absorbs concurrent requests that would otherwise land on the warehouse.
• Model portability: a model built here runs unchanged on the commercial Cube platform, and the reverse holds too.

Running it locally or self-hosting takes a Docker container with your project folder mounted and a port published for setup. It fits teams that want to own the stack, whether that means a custom BI experience, deeply integrated embedded analytics, or agents that need a governed foundation of metrics to reason over.