Agents and tools

Agent runs a chat generator in a loop with tools: when the model asks for a tool, the agent runs it and sends the result back, until the model answers.

from haystack.components.agents import Agent
from haystack.dataclasses import ChatMessage
from haystack.tools import tool

from shop_chat import ShopChat

ORDERS = {"A-1001": "shipped on 12 March", "A-1002": "waiting for stock"}


@tool
def lookup_order(order_id: str) -> str:
    """Look up the status of an order by its id, like A-1001."""
    return ORDERS.get(order_id, "no such order")

The agent gets the stand-in generator from lesson 11 and the tool:

agent = Agent(chat_generator=ShopChat(), tools=[lookup_order])

result = agent.run(messages=[ChatMessage.from_user("Where is my order A-1001?")])
for message in result["messages"]:
    detail = message.text or message.tool_call or message.tool_call_result.result
    print(f"{message.role.value:9} {detail}")

@tool turns a function into a Tool: its name, docstring and type hints become what the model sees. Agent checked that ShopChat.run accepts tools, then looped: the generator returned a tool call, the agent ran lookup_order and added a tool message with the result, and the generator answered from it.

print(lookup_order.name)
print(lookup_order.description)
print(lookup_order.parameters)

This is the tool definition a real model receives. Write the docstring for the model: what the tool does and what its arguments look like.

Pipelines as tools

ComponentTool wraps any component, and PipelineTool a whole pipeline, as a tool. The RAG pipeline from lesson 12 could become a search_policies tool, so the agent decides when to search the policies and when to look up an order.