Announcing open source MetricFlow: Governed metrics to power trustworthy AI and agents

Today, at Coalesce 2025, we announced our clear commitment to open, portable semantics for everyone.

• We are open sourcing MetricFlow, the technology that powers the dbt Semantic Layer, by moving it to the Apache 2.0 license.
• We are building it in public with partners like Snowflake and Salesforce, so metric interoperability becomes the default across tools and clouds.
• We are aligning MetricFlow with the goals of Open Semantic Interchange (OSI) so any vendor can participate and any team can adopt without lock-in.

We believe that open standards for semantic technology will power trusted AI systems for many years to come. MetricFlow is our contribution to that future.

Why open semantics now?

The semantic layer, a layer that defines business logic and its associated computations, has felt important to us for many years. We announced in 2021 that we were making investments in building our own semantic layer. In 2023, we acquired Transform, the authors of MetricFlow, the leading technology in the space. In the years since then, we’ve seen consistent adoption of MetricFlow for BI and embedded analytics use cases.

But as useful as semantic technology has been as a part of the BI stack, what no one saw coming was how critical it was about to become for AI. As it turns out, the semantic layer is the critical component to build a bridge between AI and structured data. It allows AI and agents to apply correct business logic and return reliable results every time. And the recent investments being made by major players in the space make it clear that this is now widely understood.

Now in 2025, companies are racing to put trustworthy “chat with your data” in the hands of every employee. Semantic layers make that possible. Without them, AI emits raw SQL, guesses at joins, filters, time grains, and windows, and each model guesses differently. Numbers don’t align. Trust erodes. Adoption slows.

Metrics should not be probabilistic or depend on an LLM guessing each calculation.

They should be deterministic.

MetricFlow makes this possible. It's the single most advanced way to take natural language business concepts and map them to code. This enables every LLM call—agent or a human, GPT5 or Claude 4—to measure everything exactly the same way.

We feel so strongly about the industry standardizing on a shared approach that we’ve made some significant changes to the way that MetricFlow is both licensed and governed.

1. Apache 2.0: As of today, MetricFlow is now fully available under an Apache 2.0 license. This includes all of the code required to define metrics and calculate them in multi-dialect SQL, as well as its metadata representation of semantic constructs. Other vendors can now build MetricFlow into their products in a first class way, building a high-quality bridge from business metrics to models and agents.
2. Aligned with OSI: MetricFlow will be governed and maintained with OSI partner organizations like Snowflake and Salesforce. Our shared goal is for MetricFlow to power semantic interoperability between partner platforms. We will work collectively with OSI partners and with the community to evolve MetricFlow’s metadata representation of semantics to serve the needs of this initiative.

We hope to eventually donate this technology to a leading open source foundation to ensure that semantic interoperability be powered by a truly open standard, maintained by the community in perpetuity.

An open standard for the AI age

MetricFlow gives the community an open standard for semantic metadata and an extensible engine that turns semantic intent into fast, warehouse-specific SQL for your AI systems.

How MetricFlow powers trustworthy AI:

• Built with the ecosystem. Vendors and open source contributors co-maintain the metadata spec under OSI and in public to power metric + semantic interoperability across tools and clouds.
• Plan once, run anywhere. You define your metrics and dimensions once. When a tool integrated with MetricFlow asks for a metric, the engine plans from that spec and compiles optimized, dialect-specific SQL.
• Explainability by design. Every query is inspectable. You can see how joins, filters, time grains, and policies are applied, which makes reviews and audits faster.
• Performance aware. The engine optimizes queries so teams do not trade correctness for speed.
• Made for complex calculations. MetricFlow models joins, windows, cohorts, and semi-additive measures so hard problems are correct by default.

Opening the engine means agents can ask for a metric by name and receive proven SQL. Then the dbt Semantic Layer governs how definitions are authored, versioned, and accessed, so agents return the same answer everywhere. This is a step toward a world where AI and BI systems share one language for metrics so teams get the same answer everywhere and AI can be trusted at scale.

Have a complex calculation? Here's an example of how MetricFlow works:

Ask your AI: “Gross margin % by month for North America last quarter (net of discounts and returns, on the fiscal calendar.)”

MetricFlow pulls the right joins across orders/discounts/returns and COGS, applies the region filter, aligns to fiscal months, and computes the ratio with matching numerator/denominator populations. It compiles readable, warehouse-specific SQL and returns results with lineage; change the definition, and every connected tool picks it up, no rewrites.

Unlocking the next chapter of data and AI

Open source MetricFlow and the dbt Semantic Layer turn shared semantics into everyday wins.

Here’s what this means for your business:

• Less rework, fewer tickets. A single shared definition cuts duplicate dashboarding and ticket escalations. In a dbt Labs replication, AI answered 83 percent of addressable natural language questions correctly via the dbt Semantic Layer, with several answered at 100% accuracy.
• Portability savings. Interoperable semantics mean tool or warehouse changes without rebuilding metric logic, reducing migration costs and vendor lock-in premiums.
• Production-ready AI. Replace prompt-only guesses with governed queries that respect joins, filters, time grains, and policies. Findings reinforce that structured semantics materially improve accuracy over prompting alone.
• Faster resolution time. Clear lineage and definitions shorten investigations, which cuts on-call and incident costs.
• Reduce compute. MetricFlow plans efficient, warehouse-specific SQL, which reduces wasted scans and long-running queries.

What comes next

We will continue to invest in MetricFlow with the community and our partners to add optimizations, expand warehouse coverage, and improve explainability so every team can trust and verify each answer.

In parallel, we are deepening our investment in enterprise-ready metric consumption through the dbt Semantic Layer. In the future, this includes stronger governance and security, role-based access, versioned changes with review, auditability, lineage you can trace, and reliable APIs and connectors so metrics flow safely into the tools your teams use.

Together, MetricFlow and the dbt Semantic Layer accelerate AI adoption, ensuring every answer is governed, explainable, and consistent across tools and clouds.

We are grateful to our customers, partners, and community members who pushed for a simpler path. This is one step on a longer journey, and we invite you to build it with us.

• Learn more and stay up to date on MetricFlow.
• Learn more about the Open Semantic Interchange initiative.
• Learn how the dbt Semantic Layer brings these definitions to AI systems and tools.