Cube

Cube is an agentic analytics platform built around a universal semantic layer. You model your metrics, dimensions, and joins once in Cube, then expose them to dashboards, embedded analytics, and increasingly to LLM agents through MCP connectors, Claude, ChatGPT, and Slack. The pitch, borrowed from a Brex quote on the homepage, is that 'the LLM is the engine, but the semantic layer is the map' — without that map, natural-language analytics tools tend to invent metric definitions and return numbers that don't reconcile with the finance team's.

It's aimed at two audiences: data teams who want governed Analytics Chat and workbooks for internal use, and SaaS companies (Brex, Webflow, Drata, Wix and 100+ others) embedding multi-tenant analytics into their own products. Cube Core is open source with ~18k GitHub stars; Cube Cloud is the managed, paid tier with embedded analytics, Creator Mode, and SSO — pricing is gated behind a sales conversation. If you're building a RAG or agent stack that needs to answer quantitative questions over a warehouse, this is the layer that stops the model from guessing.

Integrations span ClickHouse, Snowflake, BigQuery, Databricks, Postgres and most warehouses on the read side, with SQL, REST, GraphQL, and MCP on the serving side. The caveat: Cube is infrastructure, not a finished BI app — you still model the semantic graph yourself, and the AI features shine brightest once that modeling work is done.