Question

You are building a Mosaic AI agent that needs to calculate shipping costs by calling proprietary business logic. The logic is Python code that queries a Delta table and applies weight-based tiers. You want this callable as a tool by your LLM agent, governed by Unity Catalog permissions, and reusable across multiple agents and SQL queries. What is the recommended way to expose this logic as an agent tool on Databricks?

Accepted Answer

Register the logic as a Unity Catalog function (SQL/Python UDF) and expose it to the agent via a UCFunctionToolkit so the LLM can invoke it with governed permissions

Answer

Embed the Python function inline in the agent's system prompt so the LLM can read the code and simulate its output during reasoning

Answer

Deploy the logic as a separate Model Serving endpoint and have the agent query it through a vector search retriever

Answer

Store the shipping tiers as documents in a Vector Search index and let the agent retrieve them with semantic search at query time