Cube is an open-source semantic layer that lets a data team define metrics, dimensions, and access rules once in code, then serve them to BI dashboards, applications, and AI agents through one consistent API. It suits teams whose warehouse data must feed many consumers without each one redefining what a metric means.[1]
Quick verdict
Data platform teams serving BI, embedded analytics, and AI agents from one warehouse benefit most from Cube. Its code-defined data model with pre-aggregations and caching delivers consistent metrics over SQL, REST, GraphQL, and MCP interfaces. The material constraint is operational: production self-hosting is a multi-service deployment, and the managed cloud bills per developer plus compute. If you want governed metrics without running infrastructure, choose Cube Cloud on at least the Starter plan; if you need a no-code dashboard with no data team, choose a simpler BI tool instead.[1][2][3][4]
Key features
Code-defined semantic model Cubes, views, joins, and access rules live in JavaScript or YAML under version control, so metric definitions are reviewable, branchable, and reusable across every consumer.[1][2]
Pre-aggregations with Cube Store Declared pre-aggregated rollups are built and served from the built-in caching engine, which keeps queries fast over large warehouse tables.[1][2]
One model, four query APIs The same definitions serve BI tools over SQL, applications over REST and GraphQL, and AI agents through an MCP server, so numbers stay identical everywhere.[1][2]
JWT-based access control Row-level and member-level permissions derive from the signed security context on each request, keeping multi-tenant and per-user filtering inside the model rather than in each app.[1]
Broad warehouse and BI connectivity Sources include Snowflake, BigQuery, Databricks, Postgres, MySQL, ClickHouse, Trino, and dbt-built models, while outputs feed Superset, Tableau, Power BI, Metabase, and Looker Studio.[1]
dbt and Looker model reuse Teams can build on dbt models already in the warehouse and migrate Looker LookML with the lkml2cube path instead of rewriting metric logic.[1]
Docker Compose and Kubernetes deployment The core runs self-hosted from an all-in-one Docker Compose image up to Kubernetes via community Helm charts and Terraform, with single-store versus multi-store Cube Store topology.[1]
Managed Cube Cloud option The same model deploys to Cube Cloud with a free tier, Playground exploration, scheduled reporting, and Team and Enterprise governance for teams that prefer not to operate the stack.[1]
Who will love it
Platform teams unifying metrics across BI and apps: one versioned model behind SQL, REST, and GraphQL ends the era where every dashboard quietly computes revenue differently.[1][2]
Teams feeding governed data to AI agents: the MCP server gives agents a curated metric surface with access control instead of raw warehouse credentials.[1]
Embedded analytics builders: customer-facing products get per-tenant filtered numbers from the same model, with JWT scoping doing the hard multi-tenant work.[1]
Who should avoid it
Non-technical teams wanting point-and-click BI will stall: modelling happens in code against a live warehouse and needs SQL-fluent data staff.[1][2]
Tiny datasets with one dashboard do not justify the pre-aggregation and multi-service machinery; a single Postgres table behind Metabase needs none of it.[1]
Teams with no infrastructure budget or appetite face operating API, refresh workers, and Cube Store, or per-developer plus compute billing on the managed cloud from the first paid tier.[1][2]
What the reports suggest
Three named firsthand G2 reports: Alejo B., Chief Data Officer at a data product company (February 2026); a Verified User in Commercial Real Estate running production data apps (January 2025); and Steven L., Director of Engineering after 2.5 years on Cube (June 2023). One labelled excerpt of 20 words is quoted exactly; the other two reports are referenced by link.
3 cited community sources. Purposive sample of three inspected G2 firsthand reports, not a prevalence estimate. All reporters are technical data staff; managed-cloud-only and non-technical experiences are not represented. The 2023 setup-complexity report predates later deployment and Playground work in the official docs.
Field reports converge on the same strength: sophisticated metric modelling served as governed data products through pre-aggregation caching.[1][2]
The recurring costs are operational and documentary: confusing docs, pre-aggregation limits, weak language tooling, and cloud support and scaling visibility criticisms.[1][2][3]
Usage deepens over time: the longest-tenured reporter describes growing caching and pre-aggregation reliance across 2.5 years, which favours teams adopting Cube as infrastructure rather than a trial.[1]
Alternatives and compatibility
Conditional alternatives Teams needing visual no-code BI without a warehouse model may fit Apache Superset or Metabase better; teams rebuilding metrics inside dbt may prefer the dbt Semantic Layer. Neither replaces Cube's cross-API serving role.[1]
Existing catalogue neighbours The catalogue relates Cube to Appsmith and Retool as alternativeTo entries; Cube complements rather than replaces those app builders when dashboards need governed metrics underneath.[1]
Migration paths Looker teams can carry LookML forward with lkml2cube, and dbt-built warehouse models are reusable inputs, so prior modelling investment is not lost on adoption.[1]
Decisive tradeoff If the warehouse feeds several tools and the next consumer is an AI agent, Cube solves the consistency problem; without data engineers or a warehouse, a no-code BI tool fits better. Governance power against operational weight is the decision.[1][2][3]
How people experience it
Exact excerpt
Ability to create sophisticated logic to define metrics, dimensions, cubes and make it available as data products through preaggregation caching.
G2 review published February 3, 2026, reporting a data product company running Cube with preaggregation caching.
Questions worth asking
What is the best Cube alternative for non-technical users?
Is Cube really no-code?
No. Exploring data in the Playground and viewing dashboards can be click-driven, but defining cubes, joins, pre-aggregations, and access rules is JavaScript or YAML written by data staff, and production deployment needs infrastructure work.[1]
How much does Cube cost per month?
Cube Core is free Apache-2.0 software you host yourself, with infrastructure as your only bill. Cube Cloud lists a free tier, then Starter at $40 and Premium at $80 per developer per month plus compute usage, with concurrency and support caps rising by tier; SAML SSO, audit logging, and dedicated infrastructure are Enterprise-only. Pricing checked October 2026 on the official pricing page.[1]
Can you self-host Cube?
Yes, under the Apache-2.0 licence, with an all-in-one Docker Compose image for evaluation and Kubernetes via community Helm charts and Terraform for production. Expect a multi-service deployment with JWT authentication and separate Cube Store workers at production scale rather than a single binary.[1]
Sources and editorial review
Source research. Reviewed on October 9, 2026. Written by Sanjib Sah. AI assisted; reviewed by a human editor.
We have not run or tested this product.
Three named firsthand G2 reports: Alejo B., Chief Data Officer at a data product company (February 2026); a Verified User in Commercial Real Estate running production data apps (January 2025); and Steven L., Director of Engineering after 2.5 years on Cube (June 2023). One labelled excerpt of 20 words is quoted exactly; the other two reports are referenced by link. Purposive sample of three inspected G2 firsthand reports, not a prevalence estimate. All reporters are technical data staff; managed-cloud-only and non-technical experiences are not represented. The 2023 setup-complexity report predates later deployment and Playground work in the official docs.
- Cube documentation (official)
official · Observed October 9, 2026
Official documentation hub: data-modeling concepts, connectivity and APIs, caching, access control, deployment topology, and AI integration.
- Cube pricing (official)
official · Observed October 9, 2026
Official pricing: free tier allowances, Starter $40 and Premium $80 per developer per month plus compute, Enterprise-only SAML SSO, audit logging, and dedicated infrastructure.
- Cube deployment guides (official docs)
official · Observed October 9, 2026
Official self-host deployment topology: Docker Compose all-in-one, Kubernetes via Helm and Terraform, single-store versus production multi-store setup, JWT auth, Cube Store workers.
- G2 review: Chief Data Officer, data product company (Cube)
community · Observed October 9, 2026 · Published February 3, 2026
Firsthand G2 report by a Chief Data Officer at a data product company: sophisticated metrics, dimensions, and cubes served with preaggregation caching; cloud support and scaling visibility criticized.
- G2 review: Verified User in Commercial Real Estate (Cube)
community · Observed October 9, 2026 · Published January 21, 2025
Firsthand G2 report by a commercial real estate data engineer: CubeJS for production data apps, simple REST API, preAggs as ML and AI feature store, Git branches; docs and preAgg limits criticized.
- G2 review: Director of Engineering, 2.5 years on Cube
community · Observed October 9, 2026 · Published June 25, 2023
Firsthand G2 report by a director of engineering after 2.5 years: slicing and dicing, growing caching and pre-aggregation use; advanced self-host setup at scale described as complex.