Skip to main content

Deploying Agentic Flows

Agents are a special form of orchestration flows. They iterate in a reasoning loop, decide which tools to call, process the tool results, and iterate until they are satisfied with the final answer.

Data Agent with Iceberg Cache​

This example demonstrates how to deploy and operate a data agent that uses business data for decision making. The agent answers stock price questions by:

  1. Checking if data exists in an Iceberg table cache
  2. Fetching missing data from Yahoo Finance via a Tower app
  3. Using an LLM (local or cloud) to reason about which tools to call

The agent uses a reasoning loop powered by a language model specialized in tool calling (such as xLAM or GPT-4). The LLM dynamically decides the best path for each ticker, minimizing external API calls by leveraging cached data when available.

Inference options:

  • Local inference: Use llama.cpp or ollama with models like xLAM-2
  • Cloud inference: Use OpenAI, DeepSeek, or other providers

Highlights: LangChain, Iceberg, tool calling, llama.cpp/ollama, OpenAI/DeepSeek