Cube

TL;DR

• What it is: An open-source semantic layer that unifies metric definitions across BI tools, embedded analytics, spreadsheets, and AI agents
• Who it’s for: Engineering teams building embedded analytics, data engineers maintaining a metric store, and teams deploying AI agents that query internal data
• Cost savings: Cube Cloud (the managed version) is enterprise-priced; self-hosting Cube Core on a $20/month VPS replaces the need for proprietary BI middleware that can run $500–$2,000/month per team at scale
• Key strength: Define metrics once in a single data model; every downstream tool — from Looker to a GPT-based agent — queries the same definition
• Key weakness: The open-source core (Cube Core) is headless — there is no built-in dashboard UI, so you still need a visualization layer on top

---

What is Cube

Cube (formerly Cube.js) is an open-source semantic layer for analytics. At its core, it solves a persistent problem in data-heavy organizations: the same metric — revenue, active users, churn — gets defined differently in Tableau, Looker, a Python notebook, and an ad-hoc SQL query. Cube Core lets you define that metric once in a schema file and expose it through REST, GraphQL, and SQL APIs so every consumer queries the same underlying logic.

The project has 19,650 GitHub stars and is built by the Cube company, which also offers Cube Cloud — a managed platform that extends the open-source core with a user interface, AI-powered analytics, and enterprise access controls. The open-source Cube Core is the headless engine; Cube Cloud wraps it into a complete product.

Cube Core supports all major SQL data sources: cloud warehouses (Snowflake, Databricks, BigQuery), query engines (Presto, Amazon Athena), and application databases (Postgres, MySQL). It has a built-in relational caching engine designed for sub-second query latency at high concurrency.

The architecture has four layers: the data source connector, the semantic model (where you define dimensions, measures, and relationships), the API layer (REST, GraphQL, SQL), and the caching/pre-aggregation engine. You write your data model in YAML or JavaScript, and Cube handles the rest.

---

Why people choose it over top alternatives

vs. dbt Semantic Layer

dbt is the dominant transformation tool and added a semantic layer of its own. dbt’s semantic layer is tightly coupled to the dbt transformation workflow — if you are already running dbt models, its built-in metrics are a natural extension. Cube is database-agnostic and not tied to any transformation tool, which matters if your stack includes Databricks, Spark, or direct database queries alongside dbt. For teams not standardized on dbt, Cube offers more flexibility as a standalone semantic layer.

vs. LookML (Looker)

Looker pioneered the semantic layer concept with LookML, but it is proprietary, expensive, and locks you into the Looker ecosystem. Cube Core replicates much of the same capability at no licensing cost while exposing standard APIs rather than Looker’s proprietary endpoints. As one Cube user at Alcon described it: “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.”

vs. Embedded BI alternatives (Logi Symphony, Bold BI, Qrvey)

These products are full BI platforms — they include dashboards, charts, and user-facing interfaces. Cube Core is not a BI platform; it is the infrastructure layer underneath one. If you are evaluating whether to use Cube vs. Bold BI, the answer depends on whether you want to build a custom analytics experience (Cube + your own frontend) or deploy a complete out-of-the-box BI solution (Bold BI).

vs. Custom SQL APIs

The most common “alternative” is simply writing SQL endpoints yourself. Teams outgrow this when metric definitions diverge between teams — when marketing’s “active users” no longer matches engineering’s. Cube’s core value is eliminating that divergence rather than replacing a specific product.

---

Features: what it actually does

Semantic modeling

• Define measures, dimensions, segments, and joins in YAML or JavaScript
• Computed measures, rolling windows, percentage-of-total calculations
• Support for partitioned data, incremental builds, and pre-aggregations
• Multi-tenancy via row-level security in the data model

API layer

• REST API for most embedded analytics use cases
• GraphQL API for flexible query composition
• SQL API (PostgreSQL wire protocol) for BI tools that speak SQL, including Tableau, Power BI, Metabase, and Superset
• AI API for integrating LLMs with governed metric definitions

Caching and performance

• In-memory and external (Redis) query result caching
• Pre-aggregation engine: materialized views managed automatically by Cube
• Claimed sub-second latency for cached queries

Connectivity

• 20+ database connectors: Postgres, MySQL, BigQuery, Snowflake, Databricks, ClickHouse, Athena, Redshift, and more
• Downstream connectors to BI tools: Tableau, Power BI, Looker, Metabase, Superset, Excel, Google Sheets

Security

• JWT-based authentication for multi-tenant deployments
• Row-level security in the data model
• No user-facing authentication UI in Cube Core — that is the application’s responsibility

AI integration

• AI API designed to ground LLM responses in governed metric definitions
• Integrations with Anthropic Claude and other LLM providers
• Positioned as infrastructure for “agentic analytics”

---

Pricing math

Tier	Cost	Notes
Cube Core (self-hosted)	Free	Headless, no UI, community support
VPS to run Cube Core	~$20–40/month	Adequate for most mid-size deployments
Cube Cloud Starter	Contact sales	Managed, includes UI and support
Cube Cloud Enterprise	Contact sales	SSO, audit logs, SLA, dedicated support
Looker (Google)	~$5,000+/month	Full BI platform, proprietary
Logi Symphony	Custom enterprise	Comparable embedded analytics
Bold BI	From ~$495/month	Self-hosted option available

The primary cost argument for Cube Core self-hosting is avoiding middleware licensing fees at scale. A team building embedded analytics into a SaaS product would otherwise pay per-seat BI licensing (Looker, Sigma) or per-query fees on managed data platforms. Cube Core eliminates that layer entirely. The hidden cost is engineering time: someone needs to build and maintain the frontend visualization layer, which Cube Core does not provide.

---

Deployment reality

Cube Core deploys via Docker. The one-liner from the README is:

docker run -p 4000:4000 \
  -p 15432:15432 \
  -v ${PWD}:/cube/conf \
  -e CUBEJS_DEV_MODE=true \
  cubejs/cube

Then open http://localhost:4000. In development mode, Cube introspects your database and generates a starter schema. That part works well.

The harder part is production configuration. You need to write or refine the data model schema (YAML/JS files defining your metrics), configure database credentials, caching backend (Redis for production), and API authentication, set up a reverse proxy with HTTPS, and decide on pre-aggregation strategy for performance.

Schema authoring is where most teams spend time. The model language is straightforward but requires understanding your data relationships well. Complex joins, many-to-many relationships, and incremental pre-aggregations have a learning curve.

The SQL API (port 15432) is the fastest path to connecting existing BI tools: point Metabase or Tableau at Cube’s PostgreSQL wire protocol endpoint and your existing reports gain a semantic layer.

For production, Docker Compose with Redis is the recommended self-hosted setup. Kubernetes deployments are supported and documented for larger teams.

---

Who should use Cube

Best fit

• Engineering teams building embedded analytics into a SaaS application
• Data teams that need a single source of truth for metric definitions across multiple BI tools
• Organizations deploying AI agents that need grounded, governed data access
• Teams migrating away from proprietary BI infrastructure (Looker, SSAS) who want an open-source alternative
• Companies with existing SQL warehouses (Snowflake, BigQuery, Postgres) that want a query performance layer without changing their storage architecture

Not the right tool if

• You need a ready-to-use dashboard and chart builder — use Metabase, Grafana, or Superset instead
• Your team has no frontend engineering capacity to build visualization on top of Cube’s APIs
• You are a small team with a single BI tool and no need for a shared metric layer
• Your data stack is 100% within a single platform where a built-in governance layer already exists

---

Alternatives worth considering

• Metabase — Full self-hosted BI with dashboards and charts. No semantic layer abstraction, but much faster to deploy and use for teams that want visualization included.
• Apache Superset — Open-source BI platform with chart builder and SQL editor. Similar deployment complexity to Cube but includes a full UI.
• dbt Semantic Layer — Built into dbt Cloud. Best if your team already uses dbt for transformation and wants metrics to flow naturally from the same workflow.
• Lightdash — Open-source BI tool built on top of dbt. Combines dbt’s transformation layer with a Looker-like explore interface.
• Grafana — The dominant choice for infrastructure and application metrics visualization. Right choice when your use case is operational dashboards rather than business intelligence.

Sources

This review synthesizes 5 independent third-party articles along with primary sources from the project itself. Inline references throughout the review map to the numbered list below.

1. [1] google.co.id (2026) — “Google’s products and services - About Google” — general-products (link)
2. [2] capterra.com (2026) — “Just a moment…” — user-reviews (link)
3. [3] capterra.com (2026) — “Just a moment…” — user-reviews (link)
4. [4] selecthub.com (2026) — “Cube Alternatives: Key Findings” — comparison (link)
5. [5] golimelight.com (2026) — “Cube FP&A Alternatives and Competitors” — critical (link)
6. [6] GitHub repository — official source code, README, releases, and issue tracker (https://github.com/cube-js/cube)
7. [7] Official website — Cube project homepage and docs (https://cube.dev)

References [1]–[7] above were used to cross-check claims about features, pricing, deployment, and limitations in this review.