foundation-model-agent-loop

To build an AI agent with tool calling using Databricks Foundation Models, implement a canonical agentic loop that sends HTTP requests to model endpoints, handles message history, executes tools, and feeds results back for iterative behavior.

Yes, you can implement function calling with foundation models without complex frameworks by orchestrating interactions directly through HTTP requests to OpenAI-compatible endpoints, eliminating the need for heavy framework dependencies.

An agentic loop handles tool execution by sending message history to a foundation model, invoking defined tools based on the model's function calling output, and appending the tool results back into the conversation for the next iteration.

The best way to create a customer support chatbot that queries databases and calculates prices is to define specific tools for these actions and use an agent loop to let the foundation model invoke them based on user requests.

No, you do not need LangGraph to build tool-calling agents on Databricks; you can leverage Foundation Models through direct HTTP requests to handle the agentic loop, tool execution, and message history natively.