Abstract

Ossie comes from Open Semantic Interchange. It is an open specification that defines a vendor-neutral format for expressing business metrics, dimensions, and their relationships. It enables any tool to consume and produce semantic definitions without loss of meaning.

Open Semantic Interchange solves the problem of "semantic drift": the idea that the same business concept (e.g., "Monthly Active Users") being defined inconsistently across an organization's systems. By providing a shared, machine-readable format for metric definitions and their underlying logic, Ossie allows systems to exchange not just data but the intent behind that data. Unlike rigid schema-mapping approaches, Ossie achieves this through standardized ontologies and a decentralized architecture where systems read semantic metadata directly from the source rather than relying on point-to-point field translations.

Background

The evolution of data management has reached a critical juncture where the sheer volume of information is no longer the primary hurdle; rather, the challenge lies in the fragmentation of meaning. Historically, the industry has excelled at moving data (syntactic interoperability) but has struggled to ensure that the data retains its "truth" as it travels across different platforms. For example, a column “id” in one table may refer to “purchase_number” in another, “profit” in one context could be “overhead” in another. Solving for this “semantic drift” is the foundational problem that the Ossie was designed to solve.

The Pre-Ossie Landscape: The "Babelization" of Data

Before the formalization of OSI, the data ecosystem was characterized by several persistent bottlenecks:

Metric Duplication: A metric, defined as a named calculation or relationship, often are defined differently across their CRM, data warehouse, and BI tools. Without being able to trace the context in which a value is produced, or being able to define how it should be calculated, leads to conflicting reports. For example a user attempting to calculate “daily active users” may have several tables with “user” and “events” columns but without additional context, there are any number of seemingly correct calculations which could produce the metric.
The Third System: Integrating two systems required significant manual effort to map fields. For example, System A’s Customer_ID and System B’s Account_UID might represent the same entity, but this logical connection doesn’t exist within either system. Neither system knows the other exists, or has a way of referring to entities outside of it’s own control. This meant that engineers needed to create manually curated documents explaining how the various components fit together.
The Rise of Agentic AI: As Large Language Models (LLMs) began to act as autonomous agents, the lack of structured business context became a liability. Agents have no way to disambiguate similarly named metrics, or have automated access to the manual catalog of the relationships between entities leading to hallucinations and non-deterministic results. As queries become involved more and more entities, these problems are only compounding.

The Birth of the OSI Initiative (2025–2026)

In September 2025, a coalition of industry leaders—including Snowflake, Salesforce, dbt Labs—launched the Open Semantic Interchange initiative. The goal was to move away from proprietary silos toward a vendor-neutral standard.

The momentum accelerated rapidly:

January 2026: The OSI v0.1 Specification was officially released under the Apache 2.0 license. This provided the first YAML-based standard for defining metrics, dimensions, and relationships in a format that any tool could ingest.
The Ecosystem Expansion: Within months, dozens of partners—including Databricks, Thoughtspot, Strategy (MicroStrategy), and Collibra—joined the working group, signaling an industry-wide shift toward a "write once, query anywhere" philosophy.

Current State: A Rosetta Stone for the AI Era

Today, OSI serves as the "Rosetta Stone" for modern enterprise. By providing a shared language that bridges the gap between raw data storage and natural language understanding, it has effectively ended the era of isolated data silos. It ensures that whether a query is run by a human analyst in a dashboard or an AI agent in a chat interface, the underlying math and meaning remain consistent.

Rationale

Adoption of Ossie (OSI) is not merely a technical upgrade; it is a strategic response to the systemic "meaning debt" accumulated over decades of siloed digital growth. As data ecosystems move from human-centric reporting to AI-driven automation, the cost of ambiguity has become unsustainable.

The rationale for adopting OSI is built upon three critical pillars: The Death of Manual Mapping, The Requirement for Deterministic AI, and The Democratization of Data Sovereignty.

1. The Death of Manual Mapping (Scalability)

The relationship between entities grows quadratically as new systems are added, and each of these relationships must be annotated manually. The sales database may use "Customer_Code" for the customer database's "Id". The "expenses" table for customer accounts may be income, while the "expenses" table of the RD database might be loss. With each new entity or service the number of connections to define grows.

The Problem: Manual mapping is slow, error-prone, and breaks the moment a source schema changes.
The OSI Solution: By attaching a persistent semantic layer to the data at the source, the data becomes self-describing. Systems no longer need to "guess" what a field means; they simply read the OSI metadata. This shifts the integration burden from human translation to machine-led discovery.

2. The Requirement for Deterministic AI (Accuracy)

Generative AI and Large Language Models (LLMs) are exceptionally good at syntax but struggle with specialized logic. When an AI agent is asked to "Calculate the churn rate for Q3," it may look at a database and find three different tables labeled cancellations, churn_final, and user_dropouts. Without a semantic standard, the AI is forced to guess, leading to "hallucinated" business logic.

The Rationale: For AI to be useful in an enterprise context, it must be deterministic.
The OSI Impact: OSI provides a "Guardrail of Truth." It defines the mathematical formula for "Churn" within the metadata itself. When the AI accesses the data, it doesn't just see numbers; it sees the intent and the formula, ensuring that the output is consistent regardless of which model is processing the request.

3. Democratization and Data Sovereignty

Proprietary "walled gardens" have long used closed semantic formats to lock users into specific platforms. If your business logic is trapped inside a specific BI tool’s proprietary layer, moving to a new vendor becomes a multi-million dollar migration project.

The Rationale: Businesses require data portability. They should be able to move their intellectual property (their logic, metrics, and definitions) across cloud providers and tools without friction.
The OSI Impact: As an open-source standard, OSI acts as a "Universal Translator." It ensures that a metric defined in an AWS environment is perfectly understood by a tool running on Azure or a local private cloud. This breaks vendor lock-in and restores sovereignty to the data owner.

Future Plan

Though we expect most of the future roadmap to be decided by the community if this proposal is accepted, our intent is to continue pushing forward on the following:

Metrics Language Standard: Evolve the spec to include a formal, expressive language specification that supports advanced metric logic, windowing functions, and complex relationships.
More Converters: Expand ecosystem support by adding more converters for additional data platforms, BI tools, and modeling frameworks to ensure seamless interoperability.
Catalog (Polaris) Integration: Deepen the synergy with Apache Polaris, enabling direct discovery and management of semantic models within the catalog ecosystem.
Semantic Query Standard: Develop a standardized query specification that allows semantic definitions to be executed consistently across diverse query engines.
Reference Query Engines: Implement reference query engine implementation that demonstrate how OSI models can be interpreted and executed by various processing frameworks.
Industry-Specific OSI Models: Collaborate with industry verticals to develop pre-built, domain-specific semantic ontologies for common business use cases.

Current Status

Meritocracy

OSI is already community‑driven. The project is licensed under Apache 2.0, with a published CONTRIBUTING.md and a Code of Conduct that formalizes the Apache Way: contributors earn merit through sustained work of any kind and committership is earned by demonstrated contribution rather than employer. Specification changes follow a public process with a minimum seven‑day discussion window and a +1 / 0 / -1 vote, with binding votes reserved for the Technical Steering Committee. Under incubation we are committed to continuing this meritocracy‑driven model, to engaging the community and adjacent projects for support and contributions, and to ensuring that contributors and committers come from a broad mix of organizations through merit‑based decisions.

Ossie will follow this pattern and fully embracing the Apache Way.

Community

Ossie (OSI) is cultivating a diverse and vibrant community. Since the project repository opened in November 2025, more than 100 commits and 35 merged pull requests have landed from contributors at Snowflake, Salesforce, Databricks, dbt Labs, AtScale, Atlan, RelationalAI, ThoughtSpot, GoodData, Honeydew, Hex, and The Apache Software Foundation, with the broader participating coalition listed on open-semantic-interchange.org growing from 17 launch partners to more than 50 organizations.

Day‑to‑day collaboration happens entirely in the open: on GitHub (76+ active design discussions across spec semantics, ontology, time, dialects, and catalog integration), in the opensemantic Slack workspace, and through five working groups with dedicated leads (Metric Language, Composability, Catalog, Ontology, and Sync API) that hold their own meetings and channels. OSI is currently the specification, with first implementations already merged; more implementations and tools are expected to join the project soon. Bringing OSI to the ASF will let us consolidate these implementations and tools, grow the community further, and deepen connections within the project's ecosystem.

Core Developers

The core development team currently hails from three companies, Snowflake, Dremio, and dbt Labs, each with deep open source track records and active participation in ASF projects including Apache Beam, Apache Iceberg, Apache Polaris, and Apache Spark. Most individual contributors have direct experience with ASF governance and the Incubator process.

Alignment

OSI started with a specification, and its tooling already aligns closely with the ASF ecosystem. The first reference implementations (the osi-python core types and the OSI ↔ dbt Semantic Layer converters) are merged, with dbt Labs' MetricFlow able to consume OSI models directly as an alternative to its native YAML configuration. Connectors for several Apache projects are already in the repository or in active review: an Apache Polaris converter (merged), an Apache Spark converter (in review), and an Apache Iceberg expression‑language vendor proposal (under discussion). Joining the ASF should accelerate that alignment and foster deeper collaboration with these neighboring projects.

Known Risks

Project Name

Open Semantic Interchange name is already used by the projects (website, github). OSI is conflicting with other projects and initiatives (Open Source Initiative, …).

That’s why we are proposing to use Ossie to avoid any naming issue.

Orphaned Products

OSI is actively implemented in production by Snowflake, Salesforce, Dremio, and dbt Labs as part of their semantic layer infrastructure. This creates a strong commercial incentive for sustained investment independent of any single contributor's participation. Additionally, OSI is seeing increasing interest, development, and adoption from other organizations outside the founding group (Databricks, Collibra, Thoughtspot) and that provides a natural hedge against abandonment.

We strongly believe that Ossie is the natural move for OSI.

Inexperience with Open Source

Several of the proposed initial committers and PPMC members have direct experience in open source software, the Apache Incubator, and committerships and PMC positions in ASF projects. We do not anticipate governance or process gaps.

Homogeneous Developers

As noted above, the contributor base is currently concentrated across three organizations. We view diversification as a critical incubation goal, not a stretch objective.

To address this, we plan to follow a playbook proven in related ASF projects such as Apache Polaris, which successfully broadened its contributor base through community meetups and new-contributor sprints. Our approach will include maintaining a curated "good first issue" backlog, hosting onboarding workshops, and running focused contributor sprints to lower barriers and actively recruit participants from outside the founding organizations.

Reliance on Salaried Developers

While most of the contributors are affiliated with Snowflake, Dremio, dbt Labs, Salesforce, the commitments to open source ensure a genuine commitment to the Apache Way and open source principles.

Additionally, as the ecosystem matures and additional organizations build implementations, we expect the contributor base to naturally diversify beyond salaried employees of the founding companies. The contributor growth initiatives described above (sprints, onboarding workshops) are specifically designed to cultivate independent contributors alongside corporate ones.

Relationship with Other Apache Products

Relationship with Apache Data Processing and Query Engines

Ossie is currently a specification for semantic metadata. However, the long-term vision is broader: to evolve into a standard, open semantic metadata and query specification that can be implemented across multiple data processing engines.

Ossie does not aim to replace or compete with existing Apache engines. Instead, it is designed to complement them by providing:

A consistent semantic metadata model
A unified query specification aligned with that model
Reference implementations that integrate with existing engines

We envision close collaboration with Apache projects such as:

Apache Spark™
Apache Flink®
Apache Impala®
Apache Iceberg™
Apache Polaris™

Other SQL/query engines and execution frameworks within the ASF ecosystem

The goal is to enable these engines to interpret and execute queries expressed through a shared semantic layer, improving interoperability while preserving each engine’s independence and design principles.

Ossie will engage these communities through design discussions, shared contributors, and optional integration efforts, ensuring that any collaboration is driven by community interest and adheres to ASF governance principles.

Relationship with the Open Enterprise Data & Metadata Ecosystem

Beyond execution engines, Ossie aims to contribute to a broader open ecosystem for enterprise data, metadata, and semantic interoperability.

Many Apache projects operate in adjacent areas, including:

Data storage and table formats (Apache Iceberg™)
Metadata catalogs (Apache Hive™ and Apache Polaris™)

Ossie’s role in this ecosystem would be to:

Provide a standardized semantic metadata specification
Enable interoperability between metadata systems and execution engines
Serve as a bridge between physical data representation and business-level semantics

Excessive fascination with the Apache Brand

OssieI’s interest in joining the ASF is rooted in governance alignment, not brand leverage. A vendor-neutral interoperability standard requires a governance model built on consensus and multi-stakeholder collaboration, which is what the ASF provides.

Our existing relationships with Apache Iceberg, Polaris, and Spark communities make the ASF a natural operational home, not just a credibility signal.

Documentation

Ossie (OSI) documentation is available here: https://open-semantic-interchange.org/

Initial Source

The initial source for the OSI specification which we will submit to the Apache Software Foundation is available on GitHub: https://github.com/open-semantic-interchange/OSI

Apache 2.0 licensed.

Source and Intellectual Property Submission Plan

The code is licensed under the Apache 2.0 license, and all contributors have signed contributor license agreements based substantially on ASF CLAs
The project LICENSE file is similar to the ASF policies.

External Dependencies

Ossie specification doesn’t have dependencies from now.

Ossie converters and other implementations have external dependencies aligned with the ASF policy.

We don’t have any Category X (only A and B).

Required Resources

Mailing Lists

Git Repository

https://github.com/apache/ossie

Issue Tracking

The project will use GitHub for issue tracking.

Other Resources

Ossie will use GitHub Actions for CI/CD, in compliance with the ASF policies.

Page tree

OssieProposal