Malloy

Malloy — Semantic Data Language

Overview

Malloy, the experimental data language from Google that replaces SQL for analytics with a composable, reusable, and more readable syntax. Helps developers write Malloy models, build nested queries, and explore data with Malloy's VS Code extension and notebook interface.

Instructions

Source Definition

Define reusable data models:

// models/ecommerce.malloy — Ecommerce data model

source: orders is duckdb.table('orders.parquet') extend {
  // Dimensions (attributes to group by)
  dimension:
    order_date is created_at::date
    order_month is created_at.month
    order_year is created_at.year
    is_high_value is amount > 100
    order_size is pick
      'small' when items_count < 3
      'medium' when items_count < 10
      'large'

  // Measures (aggregations)
  measure:
    order_count is count()
    total_revenue is sum(amount)
    avg_order_value is avg(amount)
    unique_customers is count(distinct customer_id)
    revenue_per_customer is total_revenue / unique_customers

  // Reusable named queries (views)
  view: revenue_by_month is {
    group_by: order_month
    aggregate:
      total_revenue
      order_count
      avg_order_value
    order_by: order_month
  }

  view: top_customers is {
    group_by: customer_id
    aggregate:
      total_revenue
      order_count
    order_by: total_revenue desc
    limit: 20
  }

  view: daily_dashboard is {
    group_by: order_date
    aggregate:
      total_revenue
      order_count
      unique_customers
      avg_order_value
    order_by: order_date desc
    limit: 30
  }
}

Queries

Write composable, readable analytics queries:

// queries/analysis.malloy — Analytics queries using the model

import "models/ecommerce.malloy"

// Simple aggregation
run: orders -> {
  aggregate:
    total_revenue
    order_count
    avg_order_value
}

// Group by with filters
run: orders -> {
  where: order_year = 2026
  group_by: order_month
  aggregate:
    total_revenue
    order_count
  order_by: order_month
}

// Nested queries — multiple levels of aggregation in one query
run: orders -> {
  group_by: order_size
  aggregate:
    total_revenue
    order_count
    avg_order_value
  // Nested: for each order_size, show monthly breakdown
  nest: monthly_trend is {
    group_by: order_month
    aggregate: total_revenue
    order_by: order_month
  }
  // Nested: for each order_size, show top customers
  nest: top_customers is {
    group_by: customer_id
    aggregate: total_revenue, order_count
    order_by: total_revenue desc
    limit: 5
  }
}

// Pipeline: chain transformations
run: orders
  -> { where: status = 'completed' }
  -> revenue_by_month                   // Reuse named view
  -> { where: total_revenue > 10000 }   // Filter the result

Joins and Relationships

// models/full_model.malloy — Multi-table model with joins

source: customers is duckdb.table('customers.parquet') extend {
  dimension: signup_month is created_at.month
  measure:
    customer_count is count()
    avg_lifetime_value is avg(lifetime_value)
}

source: products is duckdb.table('products.parquet') extend {
  dimension: price_tier is pick
    'budget' when price < 25
    'mid-range' when price < 100
    'premium'
  measure: product_count is count()
}

source: order_items is duckdb.table('order_items.parquet') extend {
  // Join to related tables
  join_one: orders on order_id = orders.id
  join_one: products on product_id = products.id
  join_one: customers is orders.customer_id = customers.id

  measure:
    total_quantity is sum(quantity)
    item_revenue is sum(quantity * unit_price)

  // Query across joined tables
  view: revenue_by_category is {
    group_by: products.category
    aggregate:
      item_revenue
      total_quantity
    order_by: item_revenue desc
  }

  view: customer_product_matrix is {
    group_by: customers.signup_month
    aggregate: item_revenue
    nest: by_category is {
      group_by: products.category
      aggregate: item_revenue
    }
  }
}

Notebooks and Visualization

// In Malloy notebook (.malloynb) or VS Code extension

// Malloy auto-renders results as charts when appropriate

// Bar chart — group by with single measure
run: orders -> {
  group_by: status
  aggregate: order_count
}
// # bar_chart

// Line chart — time series
run: orders -> revenue_by_month
// # line_chart

// Dashboard — multiple visualizations from one query
run: orders -> {
  group_by: order_size
  aggregate: total_revenue, order_count
  nest: trend is {
    group_by: order_month
    aggregate: total_revenue
  }
}
// # dashboard

DuckDB and BigQuery Connections

// Connection configuration
// DuckDB (local files)
connection: duckdb is duckdb [
  parquet_path: "./data/"
]

// BigQuery
connection: bq is bigquery [
  project_id: "my-gcp-project"
  dataset: "analytics"
]

// Use BigQuery tables in models
source: events is bq.table('analytics.events') extend {
  measure: event_count is count()
}

Installation

# VS Code Extension (recommended)
# Install "Malloy" from VS Code Marketplace

# CLI
npm install -g @malloydata/malloy-cli

# Python package
pip install malloy

# Run a Malloy file
malloy run analysis.malloy

Examples

Example 1: Integrating Malloy into an existing application

User request:

Add Malloy to my Next.js app for the AI chat feature. I want streaming responses.

The agent installs the SDK, creates an API route that initializes the Malloy client, configures streaming, selects an appropriate model, and wires up the frontend to consume the stream. It handles error cases and sets up proper environment variable management for the API key.

Example 2: Optimizing queries performance

User request:

My Malloy calls are slow and expensive. Help me optimize the setup.

The agent reviews the current implementation, identifies issues (wrong model selection, missing caching, inefficient prompting, no batching), and applies optimizations specific to Malloy's capabilities — adjusting model parameters, adding response caching, and implementing retry logic with exponential backoff.

Guidelines

1. Models separate from queries — Define sources and views in model files; write queries in separate files or notebooks
2. Name your views — Reusable views (named queries) are Malloy's superpower; define common analyses once, use everywhere
3. Nested queries for rich analysis — Instead of multiple separate queries, nest related analyses into a single query
4. Use pick for categorization — The pick expression replaces SQL's verbose CASE WHEN for creating dimensions
5. Pipeline for progressive filtering — Chain queries with -> to progressively refine results; each step is readable
6. DuckDB for local analysis — Use DuckDB connection with Parquet files for fast local analytics; switch to BigQuery for production
7. Malloy notebooks for exploration — Use .malloynb files for iterative data exploration with inline visualization
8. Version your models — Malloy models are code; store in Git alongside your data pipelines