Sharing metadata on data semantics is not an easy problem. Open Semantic Interchange (OSI) is a new international initiative to create a standard for sharing semantic metadata in a machine-readable format across tools such as Snowflake, Databricks, dbt, Tableau, and more. In its own words:
The Open Semantic Interchange is an industry-wide specification effort to standardize how we exchange semantic metadata across analytics, AI and BI platforms, providing a vendor neutral, single source of truth for semantic data.
OSI uses YAML1 to define:
- the semantic model (top-level container),
- datasets (fact and dimension tables),
- relationships between datasets,
- fields (row-level attributes), and
- metrics (aggregations of, and computations on, fields).
These entities support custom extensions where tool vendors can add platform-specific metadata without breaking compatibility as well as AI context structure encompassing instructions, synonyms, and examples to also help AI models understand the semantics of the data they are working with.
The specification and documentation is licensed under the Creative Commons Attribution license (CC BY). The website has more info.
Reality check (per Gael Grosch):
Nothing executes against OSI yet. No vendor has shipped native import or export tooling. And translation between existing YAML semantic formats is already fairly straightforward with AI code generation tools.
What makes it worth following is the longer-term direction it points toward. A vendor-neutral semantic layer that any tool can consume, from your data warehouse to your BI platform to your AI agent, is a meaningful shift from how semantic layers have worked so far. Whether OSI becomes that layer or fades into a forgotten spec will depend on how much native tooling platforms actually ship through 2026.
The working group features some interesting names.
One that is immediately recognisable in our industry: CARTO. See, for example, here. and here.