MetricFlow: Semantic Layer and Metric SQL Engine
github.com · checked 15 Oct 2026
Open-source semantic layer that defines metrics and generates warehouse-specific SQL.
How it plugs in
Signal on Semantic Layer Software7.7#1 of 28
MetricFlow is a dbt Labs-maintained open-source semantic layer. It keeps metric and semantic-model definitions in YAML, plans joins across data sources, and produces SQL for supported warehouses. You can use it locally with dbt or access hosted execution through the dbt Semantic Layer. It supports several metric types, command-line queries, exports, and multiple integration interfaces.

Review
In a line
MetricFlow defines metric logic in a semantic layer, then produces SQL for querying those metrics across data dimensions.
What it does
Within a dbt project, teams write metric and semantic-model definitions as YAML. Supported metric types include conversion, cumulative, derived, ratio, and simple metrics.
Its engine builds and optimizes a query plan, then renders SQL for the selected warehouse. Entity types and parameters guide join construction while helping avoid fan-out and chasm joins. It can combine multiple fact and dimension sources, aggregate by different time levels, and group results by time dimensions.
The command line can list metrics, dimensions, dimension values, entities, and saved queries. It can also query metrics, check configurations, and run exports.
How it connects
The documented warehouse targets are Snowflake, BigQuery, Databricks, Redshift, and Postgres for dbt v1. MetricFlow works with dbt 1.6 and later and depends on a working dbt project with a dbt adapter.
You can install the self-hosted package from PyPI with `pip install dbt-metricflow`. Self-hosted installations run locally with the `mf` command prefix on Linux, macOS, or Windows. Hosted dbt Platform execution uses the `dbt sl` command.
The hosted dbt Semantic Layer lists connections for Power BI, Tableau, Google Sheets, Excel, Omni, Hex, Mode, and Sigma. Custom integrations can use JDBC, ADBC, GraphQL, or the Python SDK. See the wider semantic layer software category for related tools.
Good fit
MetricFlow suits teams that want one shared definition for metrics used in business intelligence tools, notebooks, spreadsheets, or other clients. Its semantic graph links entities, dimensions, and measures so users can navigate related data.
It is a fit for dbt projects that need reusable, warehouse-specific SQL and multi-step joins. The dbt Semantic Layer can also let Claude and ChatGPT query governed metrics through the dbt MCP server.
Watch for
MetricFlow requires a correctly configured dbt project, adapter, keys, and data grains. Setting up that model is an upfront effort.
dbt built-in functions and packages are not supported yet. Tableau connections do not support extracts, custom SQL, cross-database joins, or several Tableau calculated-field functions.
MetricFlow 0.209.0 and later use Apache 2.0; earlier releases used AGPL or BSL licenses.
Pricing
MetricFlow itself is free, open source, and has a free plan. The Developer plan is free for one developer seat, one project, and 3,000 successful models per month. The Apache 2.0 MetricFlow plan is also free, but it requires a working dbt project and adapter.
Hosted dbt Semantic Layer access is listed for dbt Starter, Enterprise, and Enterprise+ plans. Starter costs $100 a month, billed per user, with five developer seats, one project, 15,000 successful models per month, and 5,000 queried metrics per month. Enterprise and Enterprise+ have custom terms and price on request.
Verdict
Choose MetricFlow when your organization already uses dbt and wants a central place to define metrics before different tools query them. It is especially suitable for self-hosted teams that value an Apache-licensed project and need warehouse-aware SQL, multiple join paths, or command-line validation.
Look elsewhere if your work depends on dbt packages, built-in functions, or Tableau features excluded by its connector. Hosted use also requires weighing per-user Starter pricing or a custom Enterprise agreement.
Our take
MetricFlow makes the most sense for dbt teams that need shared metric definitions across several data tools and warehouses. Its free, self-hosted option keeps the software accessible, while hosted plans add managed execution and usage limits. The tradeoff is setup: teams must model entities, keys, and grains carefully, and some dbt and Tableau capabilities are unavailable. Consider it when consistent metric logic matters more than a quick, no-modeling setup.
- Free and open source under Apache 2.0 for version 0.209.0 and newer
- Produces SQL for Snowflake, BigQuery, Databricks, Redshift, and Postgres when used with dbt v1
- Supports multi-hop joins, time aggregation, and several metric types
- Provides JDBC, ADBC, GraphQL, and Python SDK access
- Runs self-hosted or through hosted dbt Semantic Layer execution
- Needs a working dbt project and dbt adapter
- dbt built-in functions and packages are not supported yet
- Tableau connector excludes extracts, custom SQL, and cross-database joins
- Correct keys and data grains require upfront semantic modeling
- Pick it ifdbt teams standardizing metrics across BI and notebook tools; Self-hosted projects needing warehouse-specific SQL; Organizations connecting governed metrics to AI clients
- Skip it ifYou depend on dbt packages or built-in functions; You need Tableau extracts, custom SQL, or cross-database joins; You want metrics without first modeling keys, entities, and grains
Pricing
The core MetricFlow plans are free. Developer includes one seat and one project; the Apache 2.0 MetricFlow plan requires dbt. Hosted Starter is $100 a month per user, while Enterprise and Enterprise+ are price on request.
- Developer1 developer seat · 1 project · 3,000 successful models/monthFree
- MetricFlowApache 2.0 licensed · requires a working dbt project and dbt adapterFree
- Starterper user/month · 5 developer seats · 1 project · 15,000 successful models/month · 5,000 queried metrics/month$100/mo
- EnterpriseCustom seats · 30 projects · 100,000 successful models/month · 20,000 queried metrics/monthOn request
- Enterprise+Custom seats and projects · 100,000 successful models/month · 20,000 queried metrics/monthOn request
As getdbt.com lists them, checked 15 Oct 2026.
Questions
Is MetricFlow free?
Yes. The Developer plan costs nothing, and the MetricFlow plan is free under Apache 2.0. Hosted Starter costs $100 a month per user; Enterprise plans are price on request.
What does MetricFlow require?
A working dbt project and a dbt adapter are required. MetricFlow supports dbt 1.6 and later.
Which warehouses does it support?
For dbt v1, the documented warehouses are Snowflake, BigQuery, Databricks, Redshift, and Postgres.
How can other tools connect to MetricFlow?
The hosted Semantic Layer provides GraphQL and JDBC interfaces. Custom integrations may also use ADBC or the Python SDK.
Can Tableau use MetricFlow?
Yes, but its connector lacks support for extracts, custom SQL, cross-database joins, and several Tableau calculated-field functions.
Can AI assistants query MetricFlow metrics?
Claude and ChatGPT can use the dbt MCP server to query governed metrics.