NVIDIA Auto Ontology: Open Semantic Layer for AI Agent SQL

NVIDIA officially released Auto Ontology on November 1, 2026, an open-source library that bridges the gap between enterprise metadata systems and AI agents generating SQL queries. Released under the Apache 2.0 license, the tool automatically constructs semantic layers from existing metadata repositories, ensuring AI agents query databases using certified business definitions rather than raw schema interpretations.

What Auto Ontology Solves

Enterprise databases often contain thousands of tables with cryptic column names and undocumented relationships. When AI agents attempt to generate SQL queries against these systems, they frequently misinterpret schema elements or fabricate table relationships that do not exist. Auto Ontology addresses this problem by extracting business context from metadata catalogs and transforming it into a machine-readable semantic layer. This layer acts as a controlled vocabulary that grounds agent queries in verified organizational knowledge.

The library ingests metadata from enterprise data catalogs, documentation systems, and governance platforms. It then generates standardized semantic definitions that describe what each table represents, how columns relate to business concepts, and which joins are valid. AI agents query this semantic layer instead of raw schemas, reducing hallucination rates and improving query accuracy across complex database environments.

Release Date and Availability

Officially launched on November 1, 2026, Auto Ontology is available now through GitHub under Apache 2.0 licensing. The open-source release includes full source code, deployment documentation, and integration examples for common enterprise metadata platforms. NVIDIA designed the library for immediate production use in organizations already managing metadata through established governance workflows.

Technical Architecture and Interfaces

Auto Ontology operates through three primary interfaces: the Model Context Protocol (MCP), a command-line interface (CLI), and a REST API. The MCP integration allows AI agents to request semantic context dynamically during query generation, pulling only the definitions relevant to the current task. The CLI supports batch processing for organizations that need to regenerate semantic layers on scheduled intervals, while the REST API enables custom integration patterns for proprietary agent architectures.

The library supports multiple metadata formats and can federate definitions across disparate systems. Organizations running hybrid data environments can unify semantic context from cloud data warehouses, on-premises databases, and legacy systems into a single coherent layer. This federation capability prevents agents from generating queries that incorrectly span incompatible data sources.

Enterprise Metadata Standards

Auto Ontology enforces semantic consistency by mapping enterprise metadata to standardized ontology formats. The library validates that business definitions align with documented data governance policies before exposing them to AI agents. This validation step ensures agents cannot exploit ambiguous or contradictory metadata to justify incorrect queries.

• Automatic extraction from metadata catalogs and documentation systems
• Validation against data governance policies and business glossaries
• Real-time context serving through MCP for dynamic agent workflows
• Multi-source federation for hybrid cloud and on-premises environments

What This Means

NVIDIA's Auto Ontology represents a significant infrastructure layer for enterprise AI agent deployments. By grounding SQL generation in certified metadata rather than schema inference, organizations can deploy agents against production databases with measurably lower error rates. The Apache 2.0 license removes adoption barriers for enterprises already investing in metadata management platforms, while the MCP integration positions the tool as a standard component in emerging agent orchestration stacks. Expect metadata governance teams to evaluate Auto Ontology as a required control layer before authorizing agent access to sensitive data environments.