Databricks Unity AI Gateway Now GA; Centralizes Cost Controls, Governance Across Models, Agents, MCPs

Databricks announced August 4, 2026 that Unity AI Gateway is now generally available. The product serves as a centralized control plane for enterprise AI governance across models, agents, MCPs (Model Context Protocol services), skills, and tools. More than a quadrillion tokens have passed through the gateway in the past year, with early customers including Rivian, Asana, and Edmunds. The platform consolidates cost observability, security enforcement, and developer choice into one place.

Key capabilities include: (1) End-to-end observability and granular cost attribution across models, providers, teams, and applications; (2) Hard spend caps and budget enforcement to prevent runaway token costs; (3) Smart Routing (beta) to dynamically route requests to the right model based on quality, cost, performance, and budget; (4) Contextual Service Policies to govern what agents can do at runtime (e.g., allow/deny writing to sensitive folders); (5) Unified agent tracing to capture model and MCP tool activity in one governed telemetry layer, with integration to Lakewatch (Databricks' SIEM) for security incident investigation.

For architects: Unity AI Gateway shifts governance from static access control to runtime action control. As agent sprawl accelerates, enterprises need spend visibility across frontier models, internal models, and custom agents. The platform's integration with Unity Catalog (data governance) means agents inherit data-level policies. Early adoption signals (Rivian, Asana, quadrillion tokens YoY) suggest this is moving beyond preview to production critical path for large organizations scaling AI fleets.