supervisor-api

What problem does it solve?

It removes the burden of managing tool-selection, looping, and synthesis logic on the client by delegating the full agent loop to the Databricks Supervisor API.

Core Features & Use Cases

• Server-side tool orchestration: Declare hosted tools and let responses.create() handle the iterative loop and final synthesis.
• Hosted + client-side function tools in one request: Mix Databricks-hosted tools (e.g., Genie space, UC functions, knowledge assistants) with your own Python-callable function tools.
• Production integration patterns: Drop-in changes for agent_server/agent.py, support for model selection at runtime, and optional MLflow distributed tracing.

Quick Start

Use the Databricks Supervisor API by defining your hosted and function tools, then replace your existing agent invoke/stream handlers so they call responses.create() to let Databricks run the tool loop server-side.