Semantic Layer YAML Specification

Coming Soon

The Semantic Catalog is currently in development. This reference documents the planned YAML format. We'll update these docs when the feature launches.

Reference for the semantic layer YAML format.

Overview

Semantic layers are defined in YAML with three main sections:

models:
  - name: model_name
    table: schema.table_name
    dimensions: [...]
    measures: [...]

relationships:
  - name: relationship_name
    from: model_a
    to: model_b
    type: left
    sql: "model_a.id = model_b.model_a_id"

Models

A model maps to a database table or view.

models:
  - name: accounts
    table: public.accounts
    description: "Customer accounts"

    dimensions:
      - name: segment
        type: string
        sql: segment
        description: "Account segment: Enterprise, Mid-Market, SMB"

      - name: created_date
        type: date
        sql: created_at::date
        description: "Account creation date"

    measures:
      - name: total_arr
        type: sum
        sql: arr
        description: "Total annual recurring revenue"

Dimensions

Dimensions are attributes used to group or filter data.

Property	Required	Description
name	Yes	Unique identifier
type	Yes	string, number, date, datetime, boolean
sql	Yes	SQL expression
description	No	Human-readable description
primary_key	No	Set true for primary key

Dimension Types

dimensions:
  # String dimension
  - name: status
    type: string
    sql: status

  # Date dimension
  - name: signup_date
    type: date
    sql: created_at::date

  # Boolean dimension
  - name: is_active
    type: boolean
    sql: "status = 'active'"

Measures

Measures are aggregated values.

Property	Required	Description
name	Yes	Unique identifier
type	Yes	count, count_distinct, sum, avg, min, max, number
sql	Yes	SQL expression (column or formula)
description	No	Human-readable description
filters	No	Array of filter conditions

Measure Types

measures:
  # Count all rows
  - name: total_accounts
    type: count
    sql: id

  # Count distinct values
  - name: unique_users
    type: count_distinct
    sql: user_id

  # Sum a column
  - name: total_revenue
    type: sum
    sql: amount

  # Average
  - name: avg_deal_size
    type: average
    sql: amount

  # Custom formula
  - name: conversion_rate
    type: number
    sql: "COUNT(CASE WHEN converted THEN 1 END)::float / NULLIF(COUNT(*), 0)"

Filtered Measures

Apply filters to restrict which rows are aggregated:

measures:
  - name: active_arr
    type: sum
    sql: arr
    filters:
      - sql: "status = 'active'"

  - name: enterprise_count
    type: count
    sql: id
    filters:
      - sql: "segment = 'Enterprise'"

Relationships

Define how models join together.

Property	Required	Description
name	Yes	Unique identifier
from	Yes	Source model name
to	Yes	Target model name
type	Yes	inner, left, right, full
sql	Yes	Join condition

relationships:
  - name: account_users
    from: accounts
    to: users
    type: left
    sql: "accounts.id = users.account_id"

  - name: user_events
    from: users
    to: events
    type: left
    sql: "users.id = events.user_id"

Complete Example

models:
  - name: accounts
    table: public.accounts
    description: "Customer accounts"

    dimensions:
      - name: id
        type: string
        sql: id
        primary_key: true
      - name: segment
        type: string
        sql: segment
      - name: status
        type: string
        sql: status
      - name: created_date
        type: date
        sql: created_at::date

    measures:
      - name: count
        type: count
        sql: id
      - name: total_arr
        type: sum
        sql: arr
      - name: avg_arr
        type: average
        sql: arr

  - name: users
    table: public.users
    description: "User accounts"

    dimensions:
      - name: id
        type: string
        sql: id
        primary_key: true
      - name: account_id
        type: string
        sql: account_id
      - name: role
        type: string
        sql: role

    measures:
      - name: count
        type: count
        sql: id
      - name: unique_accounts
        type: count_distinct
        sql: account_id

relationships:
  - name: account_users
    from: accounts
    to: users
    type: left
    sql: "accounts.id = users.account_id"

Validation

The semantic layer validates:

• Table and column existence
• Type compatibility
• Circular relationship detection
• Unique names within a model

Validation errors appear when saving a draft or publishing a version.