In June 2026, the Apache Incubator accepted Apache Ossie, formerly known as the Open Semantic Interchange (OSI). Ossie is an open specification for defining business semantics, the metrics, dimensions, and relationships that give your data meaning, in a vendor-neutral format any tool can read.
We think this is one of the most important standards efforts in data right now. Dremio is one of the project's core developers alongside Snowflake and dbt Labs, and the incubation proposal was championed by Jean-Baptiste Onofré, an Apache Software Foundation member and Dremio engineer who helped shepherd Apache Polaris through the same journey. Dremio co-created Apache Arrow and Apache Polaris. Ossie is the next layer of that same open-standards story.
Here's what the project is, why it matters for AI, and how to follow along.
Picture a Monday leadership meeting. Marketing reports monthly active users up 12 percent. Product shows them flat. Finance has them down 3 percent. Nobody is wrong. Each team defined "active user" differently, and no shared, machine-readable definition exists to settle it.
The industry calls this semantic drift: the same business concept defined inconsistently across an organization's systems. Every company has it. Your metric definitions live scattered across BI tools, dbt projects, spreadsheets, and tribal memory, each written in a proprietary dialect that no other tool can read.
For decades, teams tolerated the cost. Analysts reconciled numbers by hand. Engineers rebuilt the same logic in every new tool. Migrations stalled because business definitions were trapped in one vendor's format.
Then AI agents arrived, and the cost stopped being tolerable. Ask an agent to calculate churn and it finds three candidate tables, several plausible formulas, and no way to know which one your business endorses. A human analyst would ask a colleague. The agent just picks one. The result is a confident, plausible, wrong answer, and confident wrongness is the failure mode that kills enterprise AI projects.
Text-to-SQL was never the hard part. Text-to-the-right-SQL is a semantics problem. That's the problem Ossie exists to solve.
Apache Ossie is a specification plus supporting tools, not a product you install. The repository at github.com/apache/ossie contains three main pieces.
The core specification. A vendor-neutral, YAML and JSON based format for writing down semantic models: metrics with their exact formulas and exclusions, dimensions for slicing them, and the entities and relationships that connect your data. Definitions become plain text files that live in version control, get reviewed like code, and validate against a published schema.
Converters. Reference translators between Ossie and existing semantic dialects. Converters for dbt, GoodData, Salesforce, and Apache Polaris are already merged, with a Spark converter in review. Nobody adopts a standard they can't migrate to, so the converters are the adoption path.
Ontology and validation tooling. Shared vocabulary work and schema validation, plus a complete example model for the TPC-DS benchmark schema. If you want to learn the format, that example file is the fastest education available.
The architecture matters as much as the format. Ossie is decentralized by design: systems read semantic metadata directly from the source that owns it, instead of maintaining brittle point-to-point mappings between every pair of tools. Define once, understood everywhere.
The speed of this project tells you how much demand was waiting for it.
Snowflake launched the Open Semantic Interchange initiative in September 2025 with Salesforce, dbt Labs, and a founding coalition of data tooling companies. The repository opened in November. In January 2026, the v0.1 specification went live under the Apache 2.0 license. The partner roster grew from 17 launch organizations to more than 50, pulling in Databricks, ThoughtSpot, Collibra, AtScale, Atlan, and much of the analytics ecosystem, including direct competitors of the founders. That's always the tell that a standard is real.
In June 2026, JB Onofré brought the proposal to the Apache Incubator, and the vote passed with binding support from across the Incubator PMC. The project entered the ASF with more than 100 commits, contributors from a dozen companies, and five active working groups covering the metric language, composability, catalog integration, ontology, and a sync API.
Why the new name? The OSI acronym collides with the Open Source Initiative, one of the most established names in open source. The community chose Ossie, keeping the phonetic echo while clearing the confusion.
And why the ASF? Because a standard's entire value is neutrality. Companies will only pour their business logic into an interchange format if no single vendor can tilt or capture it. Apache incubation converts "trust the founding vendors" into "trust a foundation whose entire constitution prevents capture." It's the same reason Apache Iceberg and Apache Polaris won their layers.
INTERNAL LINK: Dremio's Apache Polaris deep dive coverage
At Dremio, we describe the platform as the Agentic Lakehouse, and Ossie fills a gap we've been vocal about: agents need three things from a data platform, and most platforms only deliver two.
Agents need discovery, a way to find what data exists. They need governance, enforcement of what they may touch. And they need semantics, machine-readable knowledge of what things mean. An agent with the first two but not the third can find your revenue table and query it safely, then compute revenue with a formula your CFO has never seen.
The emerging agent standards stack makes the gap concrete. The Model Context Protocol (MCP) standardized how agents reach tools. Agent2Agent (A2A) standardized how agents coordinate. Ossie is the strongest candidate for the layer that decides whether agents can be trusted with meaning, and it now has neutral governance to match the layers around it.
Here's the end-to-end picture. An agent connects to your lakehouse through an MCP server. The catalog verifies the agent's role grants and vends short-lived, scoped credentials. Alongside the tables, the agent retrieves the Ossie-formatted definition of the metric it's about to compute. Its SQL encodes your company's definition of churn, not the model's best guess, and its answer matches the CFO's dashboard because both came from the same source of truth.
That's not a diagram of the future. Every piece of it is shipping or incubating today.
Dremio's involvement in Ossie follows the same playbook as our open source history. We co-created Apache Arrow to standardize in-memory data. We co-created Apache Polaris to standardize the catalog. We've been a key Apache Iceberg contributor for years. Semantics was the last major layer of the stack without an open standard, so we showed up to help build one.
The integration story is already concrete. An Ossie converter for Apache Polaris is merged in the project's repository, catalog integration has its own working group, and the Polaris community voted this year to accept an OSI-aligned semantic model API specification. The architecture taking shape puts semantic models where they belong: governed in the open catalog, right next to the tables they describe, discoverable and permissioned like any other asset.
That maps directly onto how Dremio works today. Dremio's Open Catalog is powered by Apache Polaris, and the AI Semantic Layer already gives your data the business context that agents and analysts need, with AI-powered wikis, labels, and curated views acting as a living encyclopedia for the business. Dremio's MCP Server lets external agents from tools like Claude and LangChain query that governed, contextualized data directly. Ossie gives this whole approach an open interchange format, so the semantics you build are portable by standard rather than by promise.
No lock-in at the storage layer, the table layer, the catalog layer, or the semantic layer. That's the point of the open lakehouse, and it's why we're investing engineering time in this project.
INTERNAL LINK: Dremio AI Semantic Layer overview INTERNAL LINK: Dremio MCP Server documentation
Ossie is a young podling with a v0.1 specification, so the right posture for most teams is align, experiment, and contribute rather than migrate. Three practical moves.
Start treating semantic definitions as an exportable asset. Inventory where your metric definitions actually live. Write down the twenty that matter most, precisely, and put them under version control. Semantic debt compounds daily, and paying it down is valuable under every possible future.
Get hands-on with the format. Clone the repository, read the core spec, and walk the TPC-DS example model. If you run dbt, the merged converters and MetricFlow's ability to consume OSI models mean you can experiment with a neutral semantic model today.
Contribute. Podlings are the best entry point into the Apache world. The project has committed to the same community playbook that worked for Polaris: curated first issues, onboarding sprints, and open working groups. The dev list is [email protected].
One honest caveat: core development today is concentrated in three companies, and the specification's hardest problems, time semantics, composability, dialect translation, are still being designed in the open. The ASF's graduation bar exists to force contributor diversity before the project earns top-level status. Watch the community growth over the next year. That's the real story, and we'll be part of it.
The semantic layer is becoming the deciding factor between AI agents that produce trustworthy answers and agents that produce confident guesses. Apache Ossie is the open standard aiming to settle it, and Dremio will be contributing, integrating, and reporting on the project throughout its incubation.
Want to stay on top of Apache Ossie milestones, releases, and Dremio's semantic layer work as it happens? Follow the Dremio LinkedIn page for ongoing updates on Apache Ossie and the open lakehouse ecosystem.
Deploy agentic analytics directly on Apache Iceberg data with no pipelines and no added overhead.
We're always looking for ways to better handle and save money on our data. That's why the "data lakehouse" is becoming so popular. It offers a mix of the flexibility of data lakes and the ease of use and performance of data warehouses. The goal? Make data handling easier and cheaper. So, how do we […]
With its capabilities in on-prem to cloud migration, data warehouse offload, data virtualization, upgrading data lakes and lakehouses, and building customer-facing analytics applications, Dremio provides the tools and functionalities to streamline operations and unlock the full potential of data assets.
Dremio Arctic bring new features to Dremio Cloud, including Apache Iceberg table optimization and Data as Code.
