Unity Gateway

What it does

Unity Gateway is Databricks' governance layer for AI model access. It takes in requests from applications, coding agents, MCP servers and tools, and routes them through one central control point. The output is governed model responses, plus cost, rate-limit, routing and audit records. Built on Unity Catalog, it extends governance to runtime interactions between models, agents, MCP servers and tools.

Teams use it to attribute token spend by user, team or application, set budgets, spend limits and rate limits, and balance load across providers with automatic fallbacks. Smart Routing can select a model and harness for each coding task, and service policies can filter PII or PHI. Audit logging and tracing cover prompts, responses and tool activity. Developers connect existing coding agents such as Claude Code, Codex and Gemini CLI through the Unity Gateway CLI without changing their workflows.

It suits platform, engineering and security teams governing AI adoption at enterprise scale. Pricing is usage-based, billed per second; request a quote or see the Databricks pricing page for rates. Coverage for MCP servers, tools and models depends on Unity Catalog registration and Databricks workspace setup.

What it can do

• Governed access to proprietary and open models through one access layer
• Cost management with budgets, spend limits and rate limits
• Smart Routing of requests to a model and harness per task
• Traffic management with load balancing and automatic fallbacks
• Service policies and AI guardrails for PII and PHI filtering
• AI audit logging and tracing of prompts, responses and tool activity

Works with

• Unity Catalog
• MLflow
• Databricks Lakewatch
• Claude Code
• OpenAI Codex
• Gemini CLI

Example workflows

Attribute and cap coding agent spend

Illustrative example. Input: a team's coding agents, such as Claude Code, routed through the CLI. Steps: connect the agent, set per-user budgets and rate limits, then review attributed token usage. Output: spend reports by user and team with enforced limits.

Govern an MCP tool connection

Illustrative example. Input: an agent that calls an MCP server registered as a Unity Catalog securable. Steps: grant or revoke access with Unity Catalog privileges, attach a service policy, then review audit logs. Output: an audit trail of tool calls and policy decisions.

Before you choose

• Pricing is usage-based and billed per second; request a quote or check the Databricks pricing page for per-unit rates.
• Governance of MCP servers, tools and models depends on Unity Catalog registration and Databricks workspace setup.
• The documentation is labeled beta, so confirm availability status with Databricks before relying on it.

Listed under

• Security

Get started

Documentation is the first step; a product demo is also offered on the official page.

1. Review the Unity Gateway documentation at the docs link, or watch a product demo.
2. Connect an existing coding agent through the Unity Gateway CLI.
3. Configure model access, routing, budgets and rate limits centrally from the platform.

Costs depend on the underlying Databricks services, compute, cloud and region. Use the product’s current SKU rates and calculator; general platform trial credits are temporary.

We couldn’t find published prices for Unity Gateway. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Unity Gateway: Databricks AI gateway for cost and security ↗www.databricks.com
• Databricks Pricing: Flexible Plans for Data and AI Solutions | Databricks ↗www.databricks.com
• AI governance with Unity Gateway | Databricks on AWS ↗docs.databricks.com

Company website

databricks.com: product, pricing and policy pages

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.