Databricks has expanded Unity AI Gateway with new beta capabilities for service policies, LLM guardrails, payload logging, and cost controls, giving teams one place to govern model calls, agent actions, and MCP interactions as AI moves into production.
Key highlights
- LLM guardrails for safer AI – Apply configurable guardrails to inputs, outputs, or both to block or sanitize content such as PII, unsafe content, jailbreak attempts, and hallucinations.
- Cost controls with better visibility – Track AI cost by endpoint, model, user, and tags, with budget alerts and controls that help teams catch runaway spend early.
- Payload logging for MCPs – Capture requests and responses so teams can audit agent activity, investigate issues, and better understand how tools are being used.
- Service policies for MCPs – Define rules for which MCP tools can be called based on identity and request context, with policy enforcement on every service call.
- One governance layer across AI – Unity AI Gateway brings permissions, observability, guardrails, and usage controls together across agents, LLMs, coding agents, and MCP servers.
In the full post, you’ll see how Unity AI Gateway helps teams move from ad hoc AI usage to more controlled, production-ready operations with consistent policies, clearer observability, and stronger cost and safety controls across models and tools.
👉 Read the full post here
