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

Databricks announced the general availability of Unity AI Gateway in August 2026 [12]. The platform provided administrators with a centralized control plane within the Databricks workspace to govern proprietary models, open-weight models, coding agents, and Model Context Protocol servers [2], [8], [11]. Key capabilities included end-to-end observability through execution traces, runtime guardrails to block sensitive data such as personally identifiable information, and budget controls that allowed teams to set token spending limits by user, workspace, or group [4], [5], [8], [11], [12].

Later in the month, Databricks introduced Smart Routing within the gateway to optimize cost and performance for coding workloads [6]. This system evaluated available models and harnesses to route tasks automatically, cutting task costs by more than 30 percent while matching frontier-level model output [6]. Databricks also expanded its catalog on the Foundation Model API by introducing Moonshot AI's Kimi K3 open-weight model, hosted in the United States under zero data retention terms and covered by gateway governance [9].

Practitioners focused on operationalizing these tools within agent architectures. Tutorials and demonstrations showed teams using Unity AI Gateway alongside Databricks Apps, Omnigent, and CodeX to schedule agent tasks, trace tool calls, and manage MCP server interactions [7], [10], [13]. Meanwhile, community discussions examined real-world deployments of the gateway's policy tools, evaluating how effectively the safeguards caught unauthorized prompts and prevented data leaks across enterprise environments [1], [3], [4].