Ataccama will open-source a converter that carries governed business definitions and live data quality signals into Apache Ossie, the incubating semantic standard Snowflake originally launched as Open Semantic Interchange. The goal is to have AI agents check whether the data they read is actually fit for use.
The converter, announced today, pulls both business context and continuously evaluated quality checks out of Ataccama ONE and into Ossie-compliant YAML. Agents and analytics tools can then read beyond just the immediate meaning of whatever data they’re looking at, but also whether it is based on sensible, consistent and reliable information.
This tackles a key issue at the core of blindly sending data to agents. A semantic layer tells an agent what revenue means and which records represent a customer. It says nothing about whether the revenue table is complete, late, or failing a check. The agent will still produce a confident answer. In there lies a cause for structural distrust of AI systems, which may make the technology toxic to those that could have used it in a more cleaned-up context.
What Ossie is, and isn’t
Apache Ossie is the Apache Incubator project formerly known as Open Semantic Interchange. It is a specification, so not a product, in the form of a declarative YAML format for metrics, dimensions, joins and relationships that different platforms can read and write. AI agents are explicitly treated as “first-class consumers” of those definitions.
The project entered the incubator earlier this year. Its ecosystem pages list vendors including Snowflake, Databricks, Kyvos, Honeydew, Strategy and Veezoo, with 35 contributors on GitHub. Ataccama says its converter joins contributions from Snowflake, Databricks and dbt.
Quality signals that travel
Because Ataccama evaluates quality in source systems as well as on cloud data platforms, the company says its signals move without forcing everything onto one platform. Agents get an immediate indicator through the semantic model and can dig deeper via Ataccama’s MCP Server, retrieving which checks failed and which records are affected.
The connection to Snowflake is to be expected, not just because its platform is popular but also for financial reasons. Snowflake Ventures invested in Ataccama to tighten technical integration between the two. Ataccama also moved agentic data observability into its platform core.
“AI exacerbates the impact of incorrect data,” said Jessica Smith, Chief Product Officer at Ataccama. In the BI era, she noted, bad data usually got caught because it clashed with a report or someone’s memory of the last board deck.