How do I monitor AI usage, cost, and security in one place?

Summary

• Unity Gateway gives you a central dashboard to monitor who accessed which AI models and MCP servers, when they were used, and what data they touched, so you can detect suspicious activity and enforce governance policies.
• Log every AI request (model, user, data, timestamp, cost, tokens) for compliance audits and incident investigation, with policies that automatically block unauthorized access.
• Set rate limits and quotas per user or team to prevent runaway spending and resource exhaustion across AI tools.
• Integrate logs with your data warehouse to build cost allocation models, security alerts, and compliance dashboards.

How do I monitor AI usage, cost, and security in one place?

AI models and tools are proliferating across your organization, often without visibility or control. Developers might use various LLMs, code-generation tools, and MCP servers without proper authorization, potentially exposing sensitive data. Unity Gateway centralizes monitoring and governance, letting you see all AI access in one dashboard, enforce policies, and track cost.

What is Unity Gateway?

Unity Gateway is Databricks' unified AI governance platform. It sits between your organization's AI tools and resources, intercepting all requests to apply policies, audit access, and enforce security controls. It works across multiple LLMs (Claude, internal models, open-source frameworks) and MCP servers, giving you a single control plane for AI governance.

Key capabilities to evaluate

Capability	Why it matters
Centralized access logging	Every AI request is logged with context: who, what model, what data accessed, when. This audit trail is essential for compliance and incident response.
Real-time usage monitoring	See which teams are using which models, at what frequency. Spot unusual patterns or runaway usage quickly.
Cost tracking by model/user	Route costs back to teams and projects so you can incentivize efficient usage and budget accordingly.
Policy enforcement	Define who can use which models, on which data. Policies are enforced at the gateway, not relying on individual developers to follow guidelines.
Rate limiting and quotas	Prevent a single user or team from consuming all available capacity. Set per-user or per-team rate limits and quota.
Data lineage integration	Correlate which data was accessed during each AI request, so you can prove compliance (e.g., for HIPAA, SOC 2) or debug data exposure.

How Databricks approaches AI governance

Unity Gateway operates as a reverse proxy for AI requests. When a developer (or automated system) calls an LLM or MCP server, the request passes through the gateway, which logs it, checks the user against policies, applies rate limits, and forwards the request.
You configure policies declaratively: allow user@company.com to access LLMs on tables in the sales schema, but not customer PII; allow data engineers to use code-generation models on internal codebases, but block external APIs. Policies are inherited from Unity Catalog permissions, so your existing governance model extends to AI.
The gateway logs every request: model called, tokens consumed, latency, data accessed, and result (allowed or denied). These logs flow into a Databricks SQL warehouse, where you can query usage trends, cost per team, and security alerts. You can also export logs to a SIEM for enterprise monitoring.
Cost is tracked per request and rolled up by team, project, or cost center. This lets you run chargeback models or budget enforcement.

Use cases

Preventing unauthorized data access: Your organization needs to allow data scientists to use AI models for analysis, but not on customer PII. Unity Gateway enforces this automatically. If a user tries to query PII via an AI model, the gateway blocks it and logs the attempt.
Cost allocation and efficiency: A growing team of engineers relies on code-generation models for productivity. Without visibility, spending grows unchecked. Unity Gateway shows cost per engineer and per project, helping managers optimize tool spend.
Compliance and auditability: Your organization handles regulated data (health, financial). You must prove that only authorized personnel accessed AI, and that no sensitive data was exposed. Gateway logs provide this audit trail.
Incident response: An employee's account is compromised and used to query sensitive data via an AI model. You need to know what was accessed. Gateway logs show every request by that user, on which data, and when.

FAQs

Can I use my own LLMs or just Databricks models?

Unity Gateway works with any LLM: open-source models, internal models you've fine-tuned, and Databricks-hosted models. You configure which models are available to which users.

How do I set up cost chargeback?

Gateway logs cost per request. You can export these logs to your finance or billing system and allocate cost to teams, projects, or cost centers. Some organizations use the logs to build internal chargeback models in SQL.

What happens if someone tries to violate a policy?

The request is denied at the gateway and logged as a security event. You can set up alerts to notify security or compliance teams, and review the attempt in the audit log.

Does Unity Gateway work with MCP servers?

Yes. MCP servers are governed the same way as AI models: you can control who can use which servers, on which data, and log every interaction.

Can I integrate these logs into my SIEM or data warehouse?

Yes. Gateway logs are exported to a Databricks SQL warehouse and can also be streamed to external monitoring and security systems for centralized monitoring and alerting.