Cube — Open Source Semantic Layer for Data Apps

What it is

Cube (formerly Cube.js) is an open-source semantic layer that sits between your data warehouse and your data consumers. You define metrics, dimensions, and joins in a data model, and Cube exposes them via SQL, REST, GraphQL, and MDX APIs. Any BI tool, embedded dashboard, or AI application queries Cube instead of writing raw SQL against the warehouse.

Cube is for data engineers, analytics engineers, and product teams who need consistent metrics across multiple consumers. Instead of duplicating metric logic in every dashboard and notebook, you define it once in Cube and serve it everywhere.

How it saves time or tokens

Without a semantic layer, every consumer writes its own SQL. Two dashboards computing 'monthly active users' with slightly different WHERE clauses produce conflicting numbers. Cube eliminates this by centralizing metric definitions.

For AI applications, Cube's API is simpler than raw SQL. An LLM can query Orders.count through the Cube API rather than constructing complex JOIN statements. This reduces token usage and hallucination risk because the LLM operates on a curated, validated data model.

How to use

1. Create a new Cube project:

npx cubejs-cli create my-cube -d postgres
cd my-cube

1. Define a data model in schema/Orders.js:

cube('Orders', {
  sql_table: 'public.orders',
  measures: {
    count: { type: 'count' },
    total_revenue: { type: 'sum', sql: 'amount' },
  },
  dimensions: {
    status: { sql: 'status', type: 'string' },
    created_at: { sql: 'created_at', type: 'time' },
  },
});

1. Start Cube and query via the API:

docker compose up -d
curl 'http://localhost:4000/cubejs-api/v1/load?query={"measures":["Orders.count"]}'

Example

Querying Cube from a Python AI pipeline:

import requests

response = requests.get(
    'http://localhost:4000/cubejs-api/v1/load',
    params={'query': '{"measures":["Orders.total_revenue"],"timeDimensions":[{"dimension":"Orders.created_at","granularity":"month"}]}'}
)
data = response.json()['data']
for row in data:
    print(f"{row['Orders.created_at.month']}: ${row['Orders.total_revenue']}")

Related on TokRepo

• Database AI tools -- tools for database management and analytics
• RAG tools -- retrieval-augmented generation with structured data

Common pitfalls

• Cube's pre-aggregation system caches query results for performance, but stale caches can serve outdated data. Configure refresh keys to match your data freshness requirements.
• The data model syntax differs between Cube's JavaScript and YAML formats. Pick one format for consistency across your team.
• Cube's free tier has limitations on concurrency and query volume. For production workloads with many concurrent users, evaluate Cube Cloud pricing.