What Apache 2.0 MetricFlow Changes for dbt Semantic Layer Users

---

If your team runs MetricFlow only for its own business, the Apache 2.0 relicense changes almost nothing: the Business Source License already allowed internal production use. Atlan reads the semantic models, metrics and lineage in your dbt project. The Semantic Layer’s APIs stay proprietary and need a dbt Starter, Enterprise, or Enterprise+ plan.

What moved on October 14, 2025 is that dbt Labs relicensed MetricFlow to Apache 2.0 from v0.209.0, so anyone may now run it as a service for others. Older releases keep a clock of their own: the BSL converts each release to Apache 2.0 on its third anniversary.

Fact	Detail	Source
License change date	October 14, 2025: pull request merged and v0.209.0 released the same day	PR #1905; v0.209.0 release; dbt Labs press release
Previous license	Business Source License, MetricFlow 0.150.0 to 0.208.2 per the README; last BSL release 0.208.1, September 18, 2025	MetricFlow README; LICENSE at v0.208.1; v0.208.1 release
Older releases	Each BSL release converts to Apache 2.0 on its third anniversary	LICENSE at v0.208.1
Current license	Apache License 2.0, MetricFlow 0.209.0 and later	MetricFlow README; About MetricFlow
Latest MetricFlow release	0.213.0, September 10, 2026	v0.213.0 release
Still proprietary	Service layer (including the Gateway) and Semantic Layer APIs (GraphQL, JDBC)	dbt Semantic Layer architecture
Plans that include the APIs	dbt Starter, Enterprise, Enterprise+	dbt Semantic Layer architecture
dbt compatibility	dbt 1.6 and higher	About MetricFlow

---

Who does the MetricFlow license change affect?

The answer turns on who uses what you build with MetricFlow, and most teams need no license action.

On the dbt platform, the service, APIs and integrations are unchanged, and MetricFlow commands says the platform manages MetricFlow versioning for you.

On self-hosted dbt, you keep the same workflow (define, validate, mf query) behind most semantic layer for analytics setups. From 0.209.0 the terms add an express patent license and drop the production carve-out.

A platform team running an internal metrics service was already covered. Offering it to customers or partners is covered by the open license from 0.209.0; under the BSL it needed a commercial license from dbt Labs.

A vendor shipping MetricFlow inside a hosted product sees the biggest change. What it embeds is a compiler from metric name to SQL, the approach weighed in deterministic SQL compared with text-to-SQL and in work on agent accuracy.

You are	What changed for you	What to do	Source
On the dbt platform (dbt sl)	Nothing in the service or APIs	No license action	dbt Semantic Layer architecture; MetricFlow commands
On self-hosted dbt, internal use only	Same rights; Apache 2.0 adds an express patent license	Upgrade to 0.209.0 or later if you want Apache terms now	LICENSE at v0.208.1; Apache License 2.0; MetricFlow README
A platform team offering MetricFlow to customers or partners as a managed service	Covered by Apache 2.0 from 0.209.0; under the BSL it needed a commercial license	Upgrade to 0.209.0 or later; meet Apache notice duties if you redistribute	LICENSE at v0.208.1; Apache License 2.0
A vendor shipping MetricFlow in a product	Covered by Apache 2.0, including as SaaS	License copy, change notices, NOTICE; check trademark use	dbt Labs announcement; Apache License 2.0

---

What changed when dbt Labs relicensed MetricFlow?

MetricFlow has carried three licenses. According to the MetricFlow README, versions up to 0.140.0 were AGPL, 0.150.0 to 0.208.2 were BSL, and 0.209.0 onward is Apache 2.0. The BSL arrived in April 2023; the BSL releases published on GitHub run from 0.200.0 to 0.208.1.

dbt Labs’ announcement (2025) says the Apache code covers defining metrics, calculating them in multi-dialect SQL, and the metadata representation of semantic constructs. That is the core of any dbt Semantic Layer deployment, the metrics layer itself.

dbt Labs also says it hopes to donate MetricFlow to an open-source foundation; the repository still sits under dbt Labs on GitHub. The dbt Licensing FAQ removed its BSL references on June 1, 2026, though the Semantic Layer FAQ (last updated October 19, 2026) still calls MetricFlow “a source-available component”.

Date	What changed	Version	Source
April 28, 2023	MetricFlow moves from AGPL to the Business Source License	0.150.0 onward per the README	PR #462; MetricFlow README
September 18, 2025	v0.208.1 released, the last BSL release; its LICENSE file carries the BSL terms	v0.208.1	v0.208.1 release; LICENSE at v0.208.1
October 14, 2025	“Change License to Apache 2.0” merged; v0.209.0 released; press release at Coalesce 2025	v0.209.0	PR #1905; v0.209.0 release; dbt Labs press release
October 28, 2025	Ossie project page: MetricFlow contributed as a reference implementation for OSI	n/a	Ossie project update
April 24, 2026	Breaking change: MetricFlowEngine.get_measures_for_metrics removed; semantic interfaces move into MetricFlow	0.210.0	MetricFlow CHANGELOG; dbt-semantic-interfaces README
June 1, 2026	dbt Licensing FAQ removes BSL references	n/a	dbt Licensing FAQ
September 10, 2026	Latest release at the time of writing	0.213.0	v0.213.0 release

A pinned 0.208.x install did not change license on October 14. Its own clock still runs: 0.208.1, released September 18, 2025, reaches its third anniversary on September 18, 2028.

---

What did the Business Source License allow, and what does Apache 2.0 add?

The BSL already let your team run MetricFlow in production for its own business, so Apache 2.0 changes one right. The v0.208.1 LICENSE granted copying, modification, derivative works, redistribution and non-production use. Its Additional Use Grant then allowed production use for “your internal business purposes, but not for the purpose of providing managed services or software-as-a-service to others.”

That clause is the whole story for most teams. Internal dashboards, agents and metric services were already licensed. Serving others was possible too: the LICENSE says other use requires “a commercial license from the Licensor”.

The BSL also had a clock. Each release’s Change Date is “the 3rd anniversary of the initial release date of that release”, and rights convert to Apache 2.0 then, or on the fourth anniversary of first distribution, “whichever comes first”. Checked on November 3, 2026: by that clock, 0.200.0 to 0.202.0, released in August and September 2023 per the MetricFlow CHANGELOG, have already converted. The relicense removed the wait for 0.209.0 onward.

Apache License 2.0 grants a perpetual, worldwide, royalty-free copyright license, plus an express patent license from contributors that ends if you start patent litigation claiming the Work infringes. It grants no trademark rights, and it ships the software “AS IS”. Per dbt Labs’ announcement, other vendors can now build MetricFlow into their products, including a semantic layer for AI agents.

Right	BSL (releases 0.200.0 to 0.208.1)	Apache 2.0 (0.209.0 and later)	Source
Read, copy, modify, redistribute the code	Yes	Yes	LICENSE at v0.208.1; Apache License 2.0, section 2
Non-production use	Yes	Yes	LICENSE at v0.208.1; Apache License 2.0, section 2
Production use for your own business	Yes, under the Additional Use Grant	Yes	LICENSE at v0.208.1; Apache License 2.0, section 2
Offer MetricFlow to others as a managed service or SaaS	Not under the BSL grant; needed a commercial license from dbt Labs	Yes	LICENSE at v0.208.1; Apache License 2.0, section 2; dbt Labs announcement
Express patent license from contributors	No express patent grant in the BSL text	Yes; ends if you start patent litigation over the Work	LICENSE at v0.208.1; Apache License 2.0, section 3
Conversion to Apache 2.0	Each release on its third anniversary, at the latest its fourth	Already Apache	LICENSE at v0.208.1

This is a reading of the license text, not legal advice; check the LICENSE at the tag you run.

---

Which parts of the dbt Semantic Layer are open source and which still need a dbt plan?

dbt splits the Semantic Layer into four components, two under Apache 2.0 and two proprietary, and a BI tool or agent talks to the proprietary two. dbt Semantic Layer architecture (2026) lists MetricFlow and the semantic interfaces as Apache 2.0 on local dbt and every plan. The service layer, with its Gateway for executing SQL, and the GraphQL and JDBC APIs are proprietary, on Starter, Enterprise and Enterprise+.

The dbt Semantic Layer page says the Semantic Layer product needs a Starter or Enterprise-tier account; the MetricFlow CLI does not.

Exports need a Starter or Enterprise-tier plan, the same tier as the APIs. Two features sit higher: caching and multiple credentials per project are Enterprise and Enterprise+ only.

That split decides how a BI tool or agent reaches dbt-served metrics: through the APIs that BI tools and AI agents share, or over MCP. Where dbt sits among semantic layer tools depends on which half you count.

Component	License	Self-hosted dbt	Developer	Starter	Enterprise, Enterprise+	Source
MetricFlow engine	Apache 2.0	Yes	Yes	Yes	Yes	dbt Semantic Layer architecture
Semantic interfaces	Apache 2.0	Yes	Yes	Yes	Yes	dbt Semantic Layer architecture
Service layer (Gateway)	Proprietary	No	No	Yes	Yes	dbt Semantic Layer architecture
Semantic Layer APIs (GraphQL, JDBC)	Proprietary	No	No	Yes	Yes	dbt Semantic Layer architecture
Downstream integrations	Platform feature	No	No	Yes	Yes	dbt Semantic Layer architecture; dbt Semantic Layer FAQs
Exports	Platform feature	No	No	Yes	Yes	dbt Semantic Layer architecture; Write queries with exports
Caching	Platform feature	No	No	No	Yes	dbt Semantic Layer FAQs; Cache common queries
Multiple credentials per project	Platform feature	No	No	No	Yes	dbt Semantic Layer FAQs; Administer the Semantic Layer

Rows 1 to 4 follow dbt’s architecture table. Rows 5 to 8 combine the architecture feature table with dbt’s FAQ, Exports, caching and Semantic Layer setup pages. Read November 3, 2026.

The license moved the engine; the serving path stayed where it was. Recording what each metric means and who owns it is a separate job, the one a semantic layer and a data catalog split differently.

---

How to run MetricFlow on self-hosted dbt under Apache 2.0

Five steps take self-hosted dbt from install to a metric query.

1. Create or activate a virtual environment, then install dbt-metricflow with your adapter as an extra. Per MetricFlow commands (2026), you manage versions yourself: dbt, the adapter and MetricFlow on v1, or dbt and MetricFlow on v2.

python -m venv venv
python -m pip install "dbt-metricflow[adapter_package_name]"

2. Check the version. A metricflow package at 0.209.0 or later shipped under Apache 2.0. An older BSL release runs under its own LICENSE until its third anniversary.

3. Define a metric in the model’s YAML, where semantic models and dbt models meet. About MetricFlow shows the latest YAML spec, which dbt supports in dbt v1.12, dbt v2 and the platform’s v1 Latest release track; older versions use the legacy spec, and moving is a separate migration. The orders model from the docs’ revenue example, with comments removed:

models:
  - name: orders
    semantic_model:
      enabled: true
      name: orders_semantic_model

    agg_time_dimension: metric_time

    columns:
      - name: order_id
        description: "Primary key for orders table"
        entity:
          type: primary
          name: order_id
          label: "Order ID"

      - name: customer_id
        description: "Foreign key linking to customers"
        entity:
          type: foreign
          name: customer
          label: "Customer"

      - name: ordered_at
        granularity: day
        dimension:
          type: time
          label: "Order Date"
          description: "Date when the order was placed"

    metrics:
      - name: order_total
        description: "Total revenue from orders"
        label: "Order Total Revenue"
        type: simple
        agg: sum
        expr: order_total

4. Validate and query. Run mf validate-configs, then mf query. Adding --explain shows the generated SQL on dbt v1 (--compile on the dbt platform and dbt v2), and --csv writes results to a file. On the dbt platform, the prefix is dbt sl instead.

mf validate-configs
mf query --metrics <metric_name> --group-by <dimension_name>

5. Add validations to CI. dbt documents installing MetricFlow in your Git provider’s pipeline so the checks run on pull requests.

Two limits apply. MetricFlow doesn’t support dbt built-in functions or packages at this time; dbt says support is planned. And this path can’t serve the metric through dbt’s GraphQL or JDBC APIs, which dbt’s FAQ reserves for platform plans.

---

What do you take on if you serve MetricFlow yourself?

Under Apache 2.0, a self-built serving tier may also serve customers and partners. dbt’s Starter and Enterprise plans list what it would replace: the Gateway that executes SQL, the GraphQL and JDBC interfaces, caching, multiple credentials, and Exports for scheduled tables. Your workload decides which you need.

Redistribution carries duties. Section 4 of the Apache License asks you to give recipients a copy of the license, mark files you changed, keep existing notices, and carry any NOTICE text. Section 6 grants no permission to use dbt Labs’ trademarks or product names, beyond describing where the code came from. Section 7 ships the code with no warranty.

Embedding the Python engine means planning for breaking changes. MetricFlow 0.210.0 removed MetricFlowEngine.get_measures_for_metrics, and the dbt-semantic-interfaces package was superseded by metricflow-semantic-interfaces inside MetricFlow from 0.210.0.

For agent access, dbt’s MCP server comes self-hosted or remote, the pattern behind any MCP server for dbt. Its self-hosted server lists “Optional integration with dbt platform APIs for metadata discovery and Semantic Layer access”, so its Semantic Layer tools still run through dbt’s APIs, which vary by plan. An AI-ready semantic layer also needs agents to see where a metric came from, and MCP for data lineage is one way to give them that view.

---

Common mistakes when reading the MetricFlow license change

Most confusion comes from treating the engine, the hosted Semantic Layer, the YAML spec and Apache Ossie as one change. The engine is Apache 2.0; the service layer and APIs a BI tool calls are proprietary and need a Starter plan or above.

The latest YAML spec arrives with dbt v1.12 (and dbt v2), where measures are replaced by simple metrics defined within the model. It is a separate migration.

The Ossie project, formerly Open Semantic Interchange, received MetricFlow as a reference implementation in October 2025 and announced its new name, Apache Ossie (Incubating), on July 10, 2026. dbt v1.12 reads Ossie documents at version 0.1.0 or 0.1.1, and dbt’s docs list support in dbt v2 as coming soon.

A spec for a universal semantic layer differs from the engine that computes it, as an ontology and a semantic layer differ. Arguments that semantic layers failed and questions about where metric ownership lives sit outside the license.

Mistake	What is true	Fix	Source
“MetricFlow is open source, so the dbt Semantic Layer is free”	The service layer and APIs are proprietary	Budget for Starter or above if BI tools or agents need dbt’s APIs	dbt Semantic Layer architecture; dbt Semantic Layer FAQs
Running 0.208.x and assuming Apache terms	Terms follow the release you run and its BSL Change Date	Require metricflow>=0.209.0, or check the release’s third anniversary	MetricFlow README; LICENSE at v0.208.1
Self-serving without the service work	Apache permits it; dbt lists caching and multiple credentials as Enterprise and Enterprise+ features	Scope auth, caching and access before building	Apache License 2.0; Cache common queries; Administer the Semantic Layer
Treating the license change as the YAML spec change	The latest spec belongs to dbt v1.12 and dbt v2	Plan the spec migration separately	Upgrading to v1.12; Migrate to the latest YAML spec
Treating Ossie as MetricFlow	Ossie is a specification in the Apache Incubator; MetricFlow is a reference implementation	Check which Ossie versions dbt reads	Ossie project pages; Apache Ossie semantic layer documents
Using dbt or MetricFlow marks for a fork	Apache 2.0 grants no trademark rights	Check trademark use before naming a fork or product	Apache License 2.0, section 6
Expecting mf query --csv to replace Exports	--csv writes a local file for self-hosted users; Exports are scheduled tables or views on the platform	Use Exports on Starter or above, or schedule your own job	MetricFlow commands; Write queries with exports

---

Checklist: does the MetricFlow license change require action from your team?

Run these ten checks in order. Each one cites the page that settles it.

ID	Check	If yes	Source
MFL-01	Do you run MetricFlow on the dbt platform (dbt sl)?	Versions are managed for you; no license action	MetricFlow commands
MFL-02	On self-hosted dbt, is metricflow 0.209.0 or later?	You are on Apache 2.0 terms; if lower, a BSL release keeps BSL terms until its third anniversary (0.140.0 and earlier were AGPL)	MetricFlow README; LICENSE at v0.208.1
MFL-03	Do you offer MetricFlow to others as a managed service or SaaS, or ship it in a product?	Allowed under Apache 2.0 from 0.209.0; earlier releases needed a commercial license until their Change Date	LICENSE at v0.208.1; Apache License 2.0
MFL-04	Do you redistribute MetricFlow or a modified copy?	Include the license, mark changes, keep notices and NOTICE	Apache License 2.0, section 4
MFL-05	Does a fork or product name use “dbt” or “MetricFlow”?	Apache 2.0 grants no trademark rights	Apache License 2.0, section 6
MFL-06	Do BI tools or agents need to query metrics over an API?	dbt’s APIs need Starter or above; otherwise build your own serving tier	dbt Semantic Layer; dbt Semantic Layer architecture
MFL-07	Do you need caching or multiple credentials?	Enterprise or Enterprise+	dbt Semantic Layer FAQs; Cache common queries
MFL-08	Do you embed the Python engine?	Review 0.210.0’s breaking change and the interfaces move before upgrading	MetricFlow CHANGELOG; dbt-semantic-interfaces README
MFL-09	Do metric definitions rely on dbt built-in functions or packages?	Not supported by MetricFlow at this time; dbt lists support as planned	About MetricFlow
MFL-10	Are you planning the v1.12 YAML spec or Ossie at the same time?	Treat them as separate decisions	Upgrading to v1.12; Apache Ossie semantic layer documents

---

Where does a context layer fit once MetricFlow can run in more places?

The license decides who may run the engine that calculates a metric. It says nothing about who owns the definition, whether anyone certified it, or what breaks downstream when it changes.

dbt Labs’ announcement says the hosted Semantic Layer “governs how definitions are authored, versioned, and accessed.” That covers teams that query through dbt’s service. If your metrics are computed by more than one engine, say dbt’s service plus MetricFlow embedded in an internal agent, the definition needs a named owner above both: the gap between context engineering and a semantic layer, and the job that separates a context layer, a data catalog and a semantic layer.

Atlan reads the dbt side of that record: its crawl reference for dbt Core and for dbt Cloud list semantic models, measures, dimensions, entities and metrics, with lineage from model to semantic model to metric. For dbt Core, Atlan processes the manifest.json and run_results.json files for each job, files a self-hosted project produces without the dbt platform.

An enterprise context layer is where you record who owns a metric definition and whether anyone certified it, next to KPI definitions and a business glossary and data lineage for AI. It is the same problem as metric conflicts across analytics teams, and the input to context engineering that a semantic layer alone does not hold. Agents reach that context through the Atlan MCP server.

For most teams on the dbt platform, the license change asks for nothing. The older question stays open: which metric in your project has more than one definition today, and who would settle it. That list is a practical first input when you implement an enterprise context layer.

---

FAQs about MetricFlow’s Apache 2.0 license

1. Is the dbt Semantic Layer open source now?

Partly. dbt lists MetricFlow and the semantic interfaces under Apache 2.0, and the service layer and Semantic Layer APIs as proprietary. dbt’s FAQ calls the Semantic Layer proprietary, with some open-source components.

2. Which MetricFlow version is the first under Apache 2.0?

Version 0.209.0, released October 14, 2025. The README lists 0.150.0 to 0.208.2 under the Business Source License (the last BSL release published is 0.208.1), and each of those releases converts to Apache 2.0 on its third anniversary.

3. Can I use MetricFlow without a dbt platform plan?

Yes. You can define metrics and query them locally from the command line without a plan. dbt’s APIs and integrations still need one.

4. Does my team need to change anything if it is on the dbt platform?

The relicense asks for no license action. The service layer and APIs your BI tools call are still proprietary dbt features, and the dbt platform manages MetricFlow versions for you.

5. Can a company build MetricFlow into a commercial product?

Yes, from version 0.209.0. Apache 2.0 asks you to pass on the license, mark changed files, and keep notices, and it grants no permission to use dbt Labs’ trademarks. Under the BSL, production use beyond its Additional Use Grant needed a commercial license from dbt Labs.

6. Is Apache Ossie the same thing as MetricFlow?

No. Ossie, formerly Open Semantic Interchange, is a specification project in the Apache Incubator. dbt Labs contributed MetricFlow as a reference implementation for it.

7. Has dbt Labs donated MetricFlow to a software foundation?

dbt Labs’ announcement says it hopes to eventually donate MetricFlow to a leading open-source foundation. As of November 3, 2026, the code is published at dbt-labs/metricflow on GitHub.

---

Sources

1. dbt Labs Affirms Commitment to Open Semantic Interchange by Open Sourcing MetricFlow, dbt Labs, October 14, 2025
2. Announcing open source MetricFlow: Governed metrics to power trustworthy AI and agents, dbt Labs, last edited December 18, 2025
3. Change License to Apache 2.0 by courtneyholcomb · Pull Request #1905 · dbt-labs/metricflow · GitHub, merged October 14, 2025
4. Release MetricFlow v0.209.0 · dbt-labs/metricflow · GitHub, October 14, 2025
5. Release v0.208.1 · dbt-labs/metricflow · GitHub, September 18, 2025
6. metricflow/LICENSE at v0.208.1 · dbt-labs/metricflow · GitHub
7. metricflow/README.md at main · dbt-labs/metricflow · GitHub
8. Update MetricFlow License to BSL by nhandel · Pull Request #462 · dbt-labs/metricflow · GitHub, merged April 28, 2023
9. metricflow/CHANGELOG.md at main · dbt-labs/metricflow · GitHub, 0.210.0 entry April 24, 2026
10. Release MetricFlow 0.213.0 · dbt-labs/metricflow · GitHub, September 10, 2026
11. GitHub - dbt-labs/dbt-semantic-interfaces: The shared semantic layer definitions that dbt-core and MetricFlow use. · GitHub
12. dbt Licensing FAQ, dbt Labs, change history June 1, 2026
13. About MetricFlow, dbt Developer Hub, last updated October 20, 2026
14. dbt Semantic Layer, dbt Developer Hub, last updated September 29, 2026
15. dbt Semantic Layer architecture, dbt Developer Hub, last updated September 10, 2026
16. dbt Semantic Layer FAQs, dbt Developer Hub, last updated October 19, 2026
17. MetricFlow commands, dbt Developer Hub, last updated September 10, 2026
18. Write queries with exports, dbt Developer Hub, last updated September 8, 2026
19. Cache common queries, dbt Developer Hub, last updated July 23, 2026
20. Administer the Semantic Layer, dbt Developer Hub, last updated August 18, 2026
21. Migrate to the latest YAML spec, dbt Developer Hub, last updated September 8, 2026
22. Upgrading to v1.12, dbt Developer Hub, last updated October 16, 2026
23. Apache Ossie semantic layer documents, dbt Developer Hub, last updated September 25, 2026
24. About dbt Model Context Protocol (MCP) server, dbt Developer Hub, last updated July 23, 2026
25. Apache License, Version 2.0 | Apache Software Foundation, Version 2.0, January 2004
26. Announcing Open Source MetricFlow: Governed Metrics for AI and Agents - Apache Ossie (incubating), October 28, 2025
27. Apache Ossie (Incubating): The New Name for Open Semantic Interchange - Apache Ossie (incubating), July 10, 2026
28. Home - Apache Ossie (incubating)
29. What does Atlan crawl from dbt Core?, Atlan Documentation
30. What does Atlan crawl from dbt Cloud?, Atlan Documentation
31. Set up dbt Core, Atlan Documentation