Point an LLM at raw tables and it re-derives joins and metric logic on every prompt — so the same question returns different numbers. Cube's semantic layer gives the agent certified metrics, dimensions, joins, and access rules to select from, so answers are consistent, governed, and explainable.
Just like these companies:
Grounded, not guessing
The agent selects from metrics, dimensions, and joins that are already defined and tested — it doesn't reinvent the logic on each prompt.
Because the definitions are fixed, revenue means revenue every time — no drift between two phrasings of the same question.
Every answer traces back to a named definition in the model, so you can see exactly how a number was produced.
Model Context Protocol
Claude, ChatGPT, Cursor, or one you build — each queries the same governed model as a tool over MCP.
Define metrics and joins once; every connected agent inherits the same context without a separate integration.
If it speaks MCP, it can query Cube — your custom agents get the governed model for free.
Governed end to end
Row-level and column-level rules are part of the governed definitions the agent selects from, not bolted on afterward.
Permissions apply before the query runs, so an agent can't return a metric or a row a user isn't allowed to see.
The same rules that govern internal answers scope what each end user sees in an embedded experience.
Extensible at query time
The agent composes new calculations on top of certified metrics and dimensions, without redefining what they mean.
The base stays trustworthy while the analysis stays open-ended — you don't trade one for the other.
The same governed context serves your team's analysis and the analytics you ship to customers.
Why it works
Brex grounds its embedded agentic analytics on Cube's governed model, so answers stay consistent and scoped to each customer. Brex chose Cube over the dbt Semantic Layer and LookML.
The semantic layer is what makes the AI useful.
The future of reporting isn't a chart, it's an insight. Large language models are becoming a commodity — the LLM is the engine, but the semantic layer is the map. A well-modeled ontology is the difference between 'I don't understand that question' and a correct, contextualized answer with a chart and a clear explanation. Cube gives us the foundation to make that real for every customer.Dan MeshkovStaff Software Engineer, BrexRead the Story
Cube becomes our single source of truth for metric definitions and powers everything from customer-facing dashboards to AI-driven quarterly business reviews. CSMs gain back dozens of hours each quarter, enabled by Cube’s semantic layer and agentic analytics.
Anthony Cronander
Senior Analytics Engineer, Drata
We integrated Cube Cloud smoothly with ClickHouse, leveraging both for fast query execution while maintaining the abstraction needed for different teams to access data without diving into database-specific complexities.
Rocio Delgado
Senior Engineering Manager at Webflow
Without Cube, our data analysts might have to write 20 different queries for a single core business metric. With Cube, that metric is defined once in the data model, and every downstream tool uses that definition along with the associated calculation logic.
Dr. Jun Huang
Global Head of Data Science at Alcon
See how Cube can help you get the most of your data