dbt Semantic Layer

Overview

The dbt Semantic Layer moves metric definitions out of BI tools and into the place the tables are already built. Metrics are defined on top of existing dbt models, and every tool that asks for "revenue" gets the same definition, refreshed everywhere when it changes.

Underneath it is MetricFlow, a SQL generation engine that works out the joins. Semantic models declare entities, dimensions and a time dimension; metrics come in five types, simple, cumulative, derived, ratio and conversion; and saved queries bundle metrics, dimensions and filters into reusable nodes in the dbt graph.

Join handling is the engineering worth knowing about. MetricFlow picks the join type from the entity types and avoids fan-out and chasm joins, with multi-hop joins limited to two hops, which is what stops a metric quietly double counting.

Consumers connect through GraphQL, JDBC or a Python SDK, with ADBC and Arrow Flight SQL for custom work, and through an MCP server for AI clients. First-class integrations include Power BI, Tableau, Excel, Google Sheets, Hex, Mode and Sigma; Looker is not on the published list.

Licensing needs care when comparing. dbt Core and MetricFlow are both Apache 2.0, MetricFlow having been relicensed from BSL in October 2025. But the service layer and the Semantic Layer APIs are proprietary and available only on paid dbt plans, so the open-source pieces alone do not give you a running semantic layer.

Supported platforms are Snowflake, BigQuery, Databricks, Redshift, PostgreSQL and Trino, and explicitly not Microsoft Fabric. Caching, which pre-builds a table from a saved query and invalidates when upstream data changes, is enterprise-only.

Key features and capabilities

The same headings are used for every semantic layer & metrics entry, so two tools can be read side by side.

How metrics are defined

• Semantic models declare entities, dimensions and a time dimension over existing dbt models
• Five metric types: simple, cumulative, derived, ratio and conversion
• Dimensions are categorical or time-based, with granularities from nanosecond to year
• Saved queries bundle metrics, dimensions and filters as nodes in the dbt graph

Query interfaces

• GraphQL API, JDBC API and a Python SDK
• ADBC and Arrow Flight SQL for custom integrations
• MCP server exposing metrics, dimensions and queries to AI clients
• Exports write saved query results into the warehouse on dbt's scheduler

Performance and caching

• Declarative caching pre-builds a table from a saved query; enterprise tiers only
• The cache invalidates automatically when upstream models refresh
• Queries are rewritten to reduce compute, with pattern matching to optimise joins
• Joins chosen by entity type, avoiding fan-out and chasm joins; two-hop limit

Governance and testing

• Access follows the configured warehouse credential, so platform policies apply
• Starter allows one credential per project; enterprise tiers allow several
• Role-based access, SSO and SCIM on enterprise tiers
• Metric definitions live in version control with the rest of the dbt project

BI and tool integrations

• BI: Power BI, Tableau, Google Sheets, Excel, Hex, Mode, Omni, Sigma (preview) and others
• Looker is not on the published integrations list
• Tools without an integration connect over JDBC, GraphQL or the Python SDK
• AI clients through the dbt MCP server

How it runs

• Warehouses: Snowflake, BigQuery, Databricks, Redshift, PostgreSQL and Trino; not Microsoft Fabric
• Multi-tenant or single-tenant dbt accounts
• MetricFlow needs dbt 1.6 or later
• The service layer and APIs are proprietary and paid-plan only

Pricing

SubscriptionIncluded from Starter, $100 per user a month

The semantic layer needs a paid dbt plan. Starter is $100 per user a month and includes defining and querying metrics, with 15,000 models built and 5,000 queried metrics a month. Enterprise and Enterprise+ are quoted and raise those to 100,000 and 20,000, and add caching and multiple credentials. The free Developer plan does not include it. A 14-day trial is offered. Note how usage counts: each metric in a query counts separately, and unused allowance does not roll over.

Vendor pricing page →

Demos and videos

• dbt product spotlight: the dbt Semantic Layer Product demo · YouTube
• The new-look dbt Semantic Layer, powered by MetricFlow Webinar or talk · YouTube

About dbt Labs

dbt Labs, formerly Fishtown Analytics, has been building dbt since 2016 and raised a $150m Series C in 2021 led by Altimeter, Sequoia and Andreessen Horowitz. The semantic layer's engine came from acquiring Transform, the company behind MetricFlow, founded by former leaders of Airbnb's data platform team, in February 2023. In June 2026 dbt Labs completed a merger with Fivetran, with Fivetran's George Fraser as chief executive and dbt's Tristan Handy as president, which is worth noting when weighing vendor independence.

Founded 2016 · getdbt.com