What problem does it solve? Operating Databricks Model Serving endpoints involves many error-prone details: correct CLI syntax, readiness states, traffic routing for zero-downtime version swaps, and AI Gateway configuration. This Skill provides the exact commands, JSON payloads, and gotchas needed to manage the full endpoint lifecycle without guessing. ## Core Features & Use Cases - Endpoint Lifecycle Management: Create, query, update, and delete serving endpoints via the Databricks CLI or the MLflow Deployments Python client, with correct readiness polling on both state.ready and state.config_update. - Traffic Routing & Version Swaps: Configure A/B and canary deployments with traffic percentages, and perform zero-downtime model version swaps by updating both the UC alias and the endpoint config. - Observability & Governance: Retrieve OpenAPI schemas, build and runtime logs, Prometheus metrics, permissions, and AI Gateway rate limits; discover Foundation Model API endpoints at runtime. - Use Case: You need to roll a production chatbot endpoint to a new Llama model version without downtime. The Skill guides you to repoint the alias, call update_endpoint with matching traffic_config, and poll both state fields until the swap completes. ## Quick Start Ask the agent to create a Databricks serving endpoint for a Foundation Model and wait until it reports READY before querying it.
