Governed Context, Security and Spend Controls for AI Agents

A single dashboard, analyst report, and AI agent may each define 'net revenue' differently, creating disputes even with governed systems. Snowflake’s new Business Ontology, part of Horizon Context, provides a single governed repository for core business concepts, ensuring all teams reference the same definitions. Analysts can bootstrap definitions using existing Snowflake context via Cortex Sense, while owners approve and maintain them through a steward workflow. Approved definitions automatically propagate to semantic views, ensuring agents across finance, sales, and support operate from consistent meanings.

Cortex Sense also embeds governed context directly into agent interactions, resolving conflicts by prioritizing approved definitions and learning from corrections. Horizon Context aggregates metadata from tools like Tableau and dbt, enriching it with lineage, tags, and AI-generated descriptions to create a connected foundation for Business Ontology. The system supports intentional variations, such as region-specific metric definitions, while maintaining traceability and consistency.

Snowflake’s Cortex AI Gateway acts as a centralized, governed entry point for agents to access models and data, applying consistent policies across clouds and providers. It inherits existing identity and access controls, eliminating the need for separate agent credentials. New Cortex Sandboxes provide secure, private spaces for agents to execute custom code alongside enterprise data, launching and shutting down automatically to reduce exposure risks.

Observability features capture every model call, tool invocation, and guardrail decision in governed tables, enabling SQL-based analysis of agent behavior. Cost Governance attributes AI spend to teams or agents and enforces budget limits in near real time, while Dynamic Model Routing matches requests to approved models based on cost, accuracy, and latency, improving token efficiency by up to 3x in testing.