Databricks made Unity AI Gateway generally available with smart routing and spend controls.

Databricks announced the general availability of Unity AI Gateway in August 2026 [14], [15], [16]. The platform centralized access control, observability, and governance across external model providers, coding agents, and Model Context Protocol (MCP) servers [2], [10], [16]. To support infrastructure-as-code workflows, Terraform provider version 1.124.0 introduced dedicated resources and data sources to configure gateway model services, model provider services, and MCP endpoints [19]. Databricks also expanded available foundation models by adding Moonshot AI's open-weight Kimi K3 model with zero data retention [11].

Cost and security governance formed the core of the month's functional updates. Databricks introduced Smart Routing, which automatically directed coding queries to optimal models across the price-to-performance frontier, reducing task costs by up to 30 percent [7], [16]. Administrators gained the ability to configure named spending budgets assigned to users, workspaces, or groups to curb token usage [6], [17]. For security, the gateway added input guardrail policies that detected and blocked personally identifiable information such as emails and social security numbers before prompts reached external endpoints [8], [10], [13].

Practitioners focused heavily on agent deployments and monitoring capabilities. Tutorials demonstrated how to connect custom applications and tools like Omnigent to MCP servers while routing requests through the gateway [9], [12]. Administrators used MLflow traces within the gateway interface to track token consumption, review agent tool execution, and audit queries for sensitive data leaks [3], [5]. Community members also shared practical experiences testing the new policy safeguards to enforce organizational boundaries on model interactions [4].