ML Endpoint Governance

Deploy Models Freely
With Platform Guardrails

Databricks doesn't have built-in policies for ML endpoints. Kostavo fills that gap: enforce sizing limits, lifecycle rules, and environment policies that Databricks doesn't offer natively.

Governance Databricks Doesn't Offer

Missing Native Policies

Databricks has no built-in way to enforce endpoint sizing limits or lifecycle policies. Kostavo adds that layer.

Uncontrolled Endpoint Sizes

Anyone can deploy any workload size or enable provisioned throughput. There are no native guardrails to prevent oversized deployments.

No Lifecycle Visibility

Failed deployments, empty Vector Search endpoints, and abandoned configurations go unnoticed. Databricks won't surface them for you.

What Kostavo Scans

Endpoint Size & Config Policies

Kostavo scans endpoint workload sizes, concurrency settings, and provisioned throughput against configurable limits.

  • Workload size policies
  • Concurrency limit policies
  • Provisioned throughput detection

Failed & Stale Endpoint Detection

Kostavo identifies endpoints in failed or aborted deployment states, and Vector Search endpoints with no indexes. Findings can trigger notifications or automatic deletion.

  • Failed deployment detection
  • Empty Vector Search endpoints
  • Configurable action modes

Serving findings for the ML Serving Guardrails profile: scale-to-zero, provisioned throughput, and workload size.

Endpoint Types We Support

Model Serving

MLflow models and custom serving endpoints

Vector Search

Vector search endpoints for RAG applications

Feature Serving

Feature serving endpoints covered by all serving policies

Add the Policies Databricks Lacks

Find the endpoints running without scale-to-zero on your first scan.

Start Free Trial