Tutorial: Building a Chat Application with Function Calling


This tutorial uses Haystack 2.0. To learn more, read the Haystack 2.0 announcement or visit the Haystack 2.0 Documentation.

Overview

📚 Useful Sources:

OpenAI’s function calling connects large language models to external tools. By providing a tools list with functions and their specifications to the OpenAI API calls, you can easily build chat assistants that can answer questions by calling external APIs or extract structured information from text.

In this tutorial, you’ll learn how to convert your Haystack pipeline into a function-calling tool and how to implement applications using OpenAI’s Chat Completion API through OpenAIChatGenerator for agent-like behavior.

Setting up the Development Environment

Install Haystack 2.0 and sentence-transformers using pip:

%%bash

pip install haystack-ai
pip install "sentence-transformers>=3.0.0"

Enable Telemetry

Knowing you’re using this tutorial helps us decide where to invest our efforts to build a better product but you can always opt out by commenting the following line. See Telemetry for more details.

from haystack.telemetry import tutorial_running

tutorial_running(40)

Save your OpenAI API key as an environment variable:

import os
from getpass import getpass

if "OPENAI_API_KEY" not in os.environ:
    os.environ["OPENAI_API_KEY"] = getpass("Enter OpenAI API key:")

Learning about the OpenAIChatGenerator

OpenAIChatGenerator is a component that supports the function calling feature of OpenAI through Chat Completion API. In contrary to OpenAIGenerator, the way to communicate with OpenAIChatGenerator is through ChatMessage list. Read more about the difference between them in Generators vs Chat Generators.

To start working with the OpenAIChatGenerator, create a ChatMessage object with “SYSTEM” role using ChatMessage.from_system() and another ChatMessage with “USER” role using ChatMessage.from_user(). Then, pass this messages list to OpenAIChatGenerator and run:

from haystack.dataclasses import ChatMessage
from haystack.components.generators.chat import OpenAIChatGenerator

messages = [
    ChatMessage.from_system("Always respond in German even if some input data is in other languages."),
    ChatMessage.from_user("What's Natural Language Processing? Be brief."),
]

chat_generator = OpenAIChatGenerator(model="gpt-4o-mini")
chat_generator.run(messages=messages)

Basic Streaming

OpenAIChatGenerator supports streaming, provide a streaming_callback function and run the chat_generator again to see the difference.

from haystack.dataclasses import ChatMessage
from haystack.components.generators.chat import OpenAIChatGenerator
from haystack.components.generators.utils import print_streaming_chunk

chat_generator = OpenAIChatGenerator(model="gpt-4o-mini", streaming_callback=print_streaming_chunk)
response = chat_generator.run(messages=messages)

Creating a Function Calling Tool from a Haystack Pipeline

To use the function calling of OpenAI, you need to introduce tools to your OpenAIChatGenerator using its generation_kwargs param.

For this example, you’ll use a Haystack RAG pipeline as one of your tools. Therefore, you need to index documents to a document store and then build a RAG pipeline on top of it.

Index Documents with a Pipeline

Create a pipeline to store the small example dataset in the InMemoryDocumentStore with their embeddings. You will use SentenceTransformersDocumentEmbedder to generate embeddings for your Documents and write them to the document store with the DocumentWriter.

After adding these components to your pipeline, connect them and run the pipeline.

If you’d like to learn about preprocessing files before you index them to your document store, follow the Preprocessing Different File Types tutorial.

from haystack import Pipeline, Document
from haystack.document_stores.in_memory import InMemoryDocumentStore
from haystack.components.writers import DocumentWriter
from haystack.components.embedders import SentenceTransformersDocumentEmbedder

documents = [
    Document(content="My name is Jean and I live in Paris."),
    Document(content="My name is Mark and I live in Berlin."),
    Document(content="My name is Giorgio and I live in Rome."),
    Document(content="My name is Marta and I live in Madrid."),
    Document(content="My name is Harry and I live in London."),
]

document_store = InMemoryDocumentStore()

indexing_pipeline = Pipeline()
indexing_pipeline.add_component(
    instance=SentenceTransformersDocumentEmbedder(model="sentence-transformers/all-MiniLM-L6-v2"), name="doc_embedder"
)
indexing_pipeline.add_component(instance=DocumentWriter(document_store=document_store), name="doc_writer")

indexing_pipeline.connect("doc_embedder.documents", "doc_writer.documents")

indexing_pipeline.run({"doc_embedder": {"documents": documents}})

Build a RAG Pipeline

Build a basic retrieval augmented generative pipeline with SentenceTransformersTextEmbedder, InMemoryEmbeddingRetriever, PromptBuilder and OpenAIGenerator.

For a step-by-step guide to create a RAG pipeline with Haystack, follow the Creating Your First QA Pipeline with Retrieval-Augmentation tutorial.

from haystack.components.embedders import SentenceTransformersTextEmbedder
from haystack.components.retrievers.in_memory import InMemoryEmbeddingRetriever
from haystack.components.builders import PromptBuilder
from haystack.components.generators import OpenAIGenerator

template = """
Answer the questions based on the given context.

Context:
{% for document in documents %}
    {{ document.content }}
{% endfor %}
Question: {{ question }}
Answer:
"""
rag_pipe = Pipeline()
rag_pipe.add_component("embedder", SentenceTransformersTextEmbedder(model="sentence-transformers/all-MiniLM-L6-v2"))
rag_pipe.add_component("retriever", InMemoryEmbeddingRetriever(document_store=document_store))
rag_pipe.add_component("prompt_builder", PromptBuilder(template=template))
rag_pipe.add_component("llm", OpenAIGenerator(model="gpt-4o-mini"))

rag_pipe.connect("embedder.embedding", "retriever.query_embedding")
rag_pipe.connect("retriever", "prompt_builder.documents")
rag_pipe.connect("prompt_builder", "llm")

Run the Pipeline

Test this pipeline with a query and see if it works as expected before you start using it as a function calling tool.

query = "Where does Mark live?"
rag_pipe.run({"embedder": {"text": query}, "prompt_builder": {"question": query}})

Convert the Haystack Pipeline into a Tool

Wrap the rag_pipe.run call with a function called rag_pipeline_func. This rag_pipeline_func function will accept a query and return the response coming from the LLM of the RAG pipeline you built before. You will then introduce this function as a tool to your OpenAIChatGenerator.

def rag_pipeline_func(query: str):
    result = rag_pipe.run({"embedder": {"text": query}, "prompt_builder": {"question": query}})

    return {"reply": result["llm"]["replies"][0]}

Creating Your tools List

In addition to the rag_pipeline_func tool, create a new tool called get_current_weather to be used to get weather information of cities. For demonstration purposes, you can use hardcoded data in this function.

WEATHER_INFO = {
    "Berlin": {"weather": "mostly sunny", "temperature": 7, "unit": "celsius"},
    "Paris": {"weather": "mostly cloudy", "temperature": 8, "unit": "celsius"},
    "Rome": {"weather": "sunny", "temperature": 14, "unit": "celsius"},
    "Madrid": {"weather": "sunny", "temperature": 10, "unit": "celsius"},
    "London": {"weather": "cloudy", "temperature": 9, "unit": "celsius"},
}


def get_current_weather(location: str):
    if location in WEATHER_INFO:
        return WEATHER_INFO[location]

    # fallback data
    else:
        return {"weather": "sunny", "temperature": 21.8, "unit": "fahrenheit"}

Now, add function specifications for rag_pipeline_func and get_current_weather to your tools list by following OpenAI’s tool schema. Provide detailed descriptions about rag_pipeline_func and query so that OpenAI can generate the adaquate arguments for this tool.

tools = [
    {
        "type": "function",
        "function": {
            "name": "rag_pipeline_func",
            "description": "Get information about where people live",
            "parameters": {
                "type": "object",
                "properties": {
                    "query": {
                        "type": "string",
                        "description": "The query to use in the search. Infer this from the user's message. It should be a question or a statement",
                    }
                },
                "required": ["query"],
            },
        },
    },
    {
        "type": "function",
        "function": {
            "name": "get_current_weather",
            "description": "Get the current weather",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {"type": "string", "description": "The city and state, e.g. San Francisco, CA"}
                },
                "required": ["location"],
            },
        },
    },
]

Running OpenAIChatGenerator with Tools

To use the function calling feature, you need to pass the list of tools in the run() method of OpenAIChatGenerator as generation_kwargs.

Instruct the model to use provided tools with a system message and then provide a query that requires a function call as a user message:

from haystack.dataclasses import ChatMessage
from haystack.components.generators.chat import OpenAIChatGenerator
from haystack.components.generators.utils import print_streaming_chunk

messages = [
    ChatMessage.from_system(
        "Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous."
    ),
    ChatMessage.from_user("Can you tell me where Mark lives?"),
]

chat_generator = OpenAIChatGenerator(model="gpt-4o-mini", streaming_callback=print_streaming_chunk)
response = chat_generator.run(messages=messages, generation_kwargs={"tools": tools})

As a response, you’ll get a ChatMessage with information about the tool name and arguments in JSON format:

{'replies': [
    ChatMessage(
        content='[{"index": 0, "id": "call_3VnT0XQH0ye41g3Ip5CRz4ri", "function": {"arguments": "{\\"query\\":\\"Where does Mark live?\\"}", "name": "rag_pipeline_func"}, "type": "function"}]', role=<ChatRole.ASSISTANT: 'assistant'>, 
        name=None, 
        meta={'model': 'gpt-4o-mini-2024-07-18', 'index': 0, 'finish_reason': 'tool_calls', 'usage': {}}
        )
    ]
}

You can then parse the message content string into JSON and call the corresponding function with the provided arguments.

import json

## Parse function calling information
function_call = json.loads(response["replies"][0].content)[0]
function_name = function_call["function"]["name"]
function_args = json.loads(function_call["function"]["arguments"])
print("Function Name:", function_name)
print("Function Arguments:", function_args)

## Find the correspoding function and call it with the given arguments
available_functions = {"rag_pipeline_func": rag_pipeline_func, "get_current_weather": get_current_weather}
function_to_call = available_functions[function_name]
function_response = function_to_call(**function_args)
print("Function Response:", function_response)

As the last step, run the OpenAIChatGenerator again by appending the function response to the messages list as a new message with ChatMessage.from_function() and let the model summarize the results.

from haystack.dataclasses import ChatMessage

function_message = ChatMessage.from_function(content=json.dumps(function_response), name=function_name)
messages.append(function_message)

response = chat_generator.run(messages=messages, generation_kwargs={"tools": tools})

Building the Chat Application

As you notice above, OpenAI Chat Completions API does not call the function; instead, the model generates JSON that you can use to call the function in your code. That’s why, to build an end-to-end chat application, you need to check if the OpenAI response is a tool_calls for every message. If so, you need to call the corresponding function with the provided arguments and send the function response back to OpenAI. Otherwise, append both user and messages to the messages list to have a regular conversation with the model.

To build a nice UI for your application, you can use Gradio that comes with a chat interface. Install gradio, run the code cell below and use the input box to interact with the chat application that has access to two tools you’ve created above.

Example queries you can try:

  • What is the capital of Sweden?”: A basic query without any function calls
  • Can you tell me where Giorgio lives?”: A basic query with one function call
  • What’s the weather like in Berlin?”, “Is it sunny there?”: To see the messages are being recorded and sent
  • What’s the weather like where Jean lives?”: To force two function calls
  • What’s the weather like today?”: To force OpenAI to ask more clarification

Keep in mind that OpenAI models can sometimes hallucinate answers or tools and might not work as expected.

%%bash

pip install gradio
import gradio as gr
import json

from haystack.dataclasses import ChatMessage
from haystack.components.generators.chat import OpenAIChatGenerator

chat_generator = OpenAIChatGenerator(model="gpt-4o-mini")
response = None
messages = [
    ChatMessage.from_system(
        "Don't make assumptions about what values to plug into functions. Ask for clarification if a user request is ambiguous."
    )
]


def chatbot_with_fc(message, history):
    messages.append(ChatMessage.from_user(message))
    response = chat_generator.run(messages=messages, generation_kwargs={"tools": tools})

    while True:
        # if OpenAI response is a tool call
        if response and response["replies"][0].meta["finish_reason"] == "tool_calls":
            function_calls = json.loads(response["replies"][0].content)
            print(response["replies"][0])
            for function_call in function_calls:
                ## Parse function calling information
                function_name = function_call["function"]["name"]
                function_args = json.loads(function_call["function"]["arguments"])

                ## Find the correspoding function and call it with the given arguments
                function_to_call = available_functions[function_name]
                function_response = function_to_call(**function_args)

                ## Append function response to the messages list using `ChatMessage.from_function`
                messages.append(ChatMessage.from_function(content=json.dumps(function_response), name=function_name))
                response = chat_generator.run(messages=messages, generation_kwargs={"tools": tools})

        # Regular Conversation
        else:
            messages.append(response["replies"][0])
            break
    return response["replies"][0].content


demo = gr.ChatInterface(
    fn=chatbot_with_fc,
    examples=[
        "Can you tell me where Giorgio lives?",
        "What's the weather like in Madrid?",
        "Who lives in London?",
        "What's the weather like where Mark lives?",
    ],
    title="Ask me about weather or where people live!",
)

## Uncomment the line below to launch the chat app with UI
# demo.launch()

What’s next

🎉 Congratulations! You’ve learned how to build chat applications that demonstrate agent-like behavior using OpenAI function calling and Haystack Pipelines.

If you liked this tutorial, there’s more to learn about Haystack 2.0:

To stay up to date on the latest Haystack developments, you can sign up for our newsletter or join Haystack discord community.

Thanks for reading!