Cube

Overview

Cube is a semantic layer: you define metrics, dimensions and joins once, and every dashboard, embedded app or AI agent that asks a question gets the same answer. It is code-first, so models live in YAML or JavaScript under version control, and they move unchanged between the open-source and hosted versions.

Models are built from cubes, each mapped to a table, with measures and dimensions on top, then views that present query-ready datasets to consumers. Multi-fact views, dynamic models generated with Jinja or JavaScript, and a visual modeller in the commercial product round it out.

What is unusual is how many ways you can query it. There are SQL, REST, GraphQL and DAX interfaces, an MDX interface for native Excel pivot tables, and an MCP server so AI assistants can query the model directly.

Performance rests on two levels of caching. An in-memory cache is on by default; pre-aggregations are the real mechanism, materialising summaries into Cube Store, a columnar store built for the job, which also replaces the need for an external Redis for cache and queue.

Access control is declarative in the model, covering row-level and member-level rules, with four built-in roles and custom roles on the enterprise tier. Integrations include dbt, two-way Snowflake semantic views, Excel, Google Sheets and Slack, over 25 data sources, and orchestration through Airflow, Dagster or Prefect.

The split is clear: Cube Core is Apache 2.0 and self-hosted with Docker, while the hosted platform adds the visual tooling, AI features, dedicated deployments and enterprise controls. Recent direction is agentic: an analytics chat, agent skills for coding assistants, and MCP support.

Key features and capabilities

The same headings are used for every semantic layer & metrics entry, so two tools can be read side by side.

How metrics are defined

• Cubes defined in YAML or JavaScript, mapped to tables, with measures and dimensions
• Views present curated, query-ready datasets, including multi-fact views
• Models generated dynamically with Jinja or JavaScript
• Visual modeller and data model IDE in the hosted platform

Query interfaces

• SQL, REST, GraphQL and DAX interfaces
• MDX interface for native Excel pivot tables, in preview on the enterprise tier
• MCP server for AI assistants, on premium and enterprise tiers
• Analytics chat turns plain-language questions into queries against the model

Performance and caching

• Two-level caching: in-memory by default, then pre-aggregations
• Cube Store holds pre-aggregations in columnar form for sub-second queries
• Cube Store also serves as cache and queue, so no external Redis is needed
• Lambda pre-aggregations, and documented rules for when a query matches one

Governance and testing

• Declarative access policies in the data model
• Row-level security by user attribute, plus member-level rules
• Four built-in roles: admin, developer, explorer, viewer; custom roles on enterprise
• Audit log, and single sign-on with Google Workspace, Entra ID, Okta or SAML

BI and tool integrations

• dbt models pulled in and converted to cubes, on premium and above
• Two-way integration with Snowflake semantic views, on enterprise
• Excel, Google Sheets and Slack; agent skills for coding assistants
• Over 25 data sources: Snowflake, BigQuery, Databricks, Redshift, ClickHouse, DuckDB, Trino and others

How it runs

• Cube Core self-hosted with Docker, under Apache 2.0
• Hosted deployment types: shared for development, dedicated, and multi-cluster
• Shared deployments suspend after 30 minutes idle and do not refresh pre-aggregations on schedule
• Bring your own cloud on AWS, Azure or Google Cloud

Pricing

SubscriptionFree tier; Starter $40 per developer a month

Four tiers. Free is limited to 1,000 queries a day, two shared deployments and five workbooks. Starter is $40 per developer a month plus compute at $0.60 an hour for a dedicated deployment. Premium is $80 per developer, $40 per explorer and $20 per viewer, with unlimited queries, embedding and a 99.95% uptime commitment. Enterprise is quoted, and adds single-tenant installation, bring your own cloud and model, SSO and audit trails. AI tokens are billed separately, with a per-seat grant each cycle.

Vendor pricing page →

Demos and videos

• Demo: ClickHouse and Cube with dashboards, embedded analytics and AI agents Product demo · YouTube
• Cube agent skills Product demo · YouTube
• Cube channel Channel or playlist · YouTube

About Cube Dev

Cube is developed by Cube Dev, Inc., co-founded by Artyom Keydunov, chief executive, and Pavel Tiunov, chief technology officer. The company announced a $25m investment in June 2024 with Databricks Ventures among the investors. Its founding year and head office are not published on its own site. The open-source project is Apache 2.0 licensed, with the client libraries under MIT, and the hosted platform is built on the same core. Customers named on its site include Maersk, Wix, Intuit and Walmart.

cube.dev