Prerequisites for AI agent Demo

Before we start, generate Open AI API Key, you can checkout step "Pre-Requisties Before making a Fine-tune Model" in this blog here to know how to generate Open AI API Key - Fine Tune Model Blog

Now for this session, we will be using Jupyter Notebook, so install Anaconda on your local Machine - Anaconda Download

Once Downloaded and Installed, You will be able to see various tools offered by Anaconda

click on Jupyter Notebook - Launch

This would open Jupyter Notebook in your local browser

Now lets us create a new folder for our AI Agents here

Now once you are inside folder, let us create Jupyter Notebooks, we will create env file first to save our OpenAI API key

Now here we will save our generated OpenAIAPI key in this format

and also select python as kernel

click on Run icon or "Shift+Enter" for cell to execute code

here this code generally creates .env file and "w" stands for write, which actually erases any existing data of .env file and writes new data, if you want to append data, you can use "a" instead of w

AI Agent

Now let us create a new notebook for us to create AI agents and call this file there, we will create translator AI agent, you can create any AI agent of your choice

create a new notebook

And now let us call our OpenAI API key first before we start building our Agent

from dotenv import load_dotenv
load_dotenv()

if the result returned is true, this means we were able to successfully import our OpenAI API key and it would be used, Python checks the env file whenever required automatically and will be used by our AI agent

Now install Open-ai agents

pip install openai-agents

once agents are installed, let us check if we are able to successfully import agents and runners

from agents import Agent, Runner

since it executed cell without any errors, we are able to successfully import agents

Now let us build our first AI Agent

Build your own AI Agent

agent= Agent (

    name= "Portugal language translator",

    instructions= "you are a portugal langugage translator, you need to take user inputs in english and translate it in portugese" )

Here i had build translator agent, but you can built any agent of your choice like Spanish Translator, French Translator or even any other AI agent like detective agent for fun but make sure you give right instructions for agent to know its expected Task

Now let us have chat_history as well included for us to have logs

chat_history = []

Now let us use this AI agent, let us build user input prompt

while True:

    userInput = input("Please provide word or sentence you want to translate: ")



    if userInput.lower() in ["exit", "bye"]:

        print("Sayonara Amigo!")

        break



    elif userInput.lower() == "history":

        print("\n---- Chat History ----")

        for i, (u, b) in enumerate(chat_history):

            print(f"{i+1}. You: {u}\n   ChatBot: {b}")

        print("----------------------\n")

        continue



    result = await Runner.run(agent, userInput)

    response = result.final_output



    chat_history.append((userInput, response))



    print("\nChatBot:", response)

    print("To end chat, please type 'bye' or 'exit'. Type 'history' to view past conversations.\n")

since now our AI agent is ready to use, let us check if our AI agent is able to successfully do our job or not, since our AI agent is translator here, it takes inputs in English and gives response to us in Portugese

As seen, it could translate our inputs to english, Now since this Agent runs in loop, and we have specified condition to include history, you can type history to check conversation

Now since this agent runs in a loop, and to exit loop , type "exit" or "bye" to close AI agent

HandOff Agents

Now let us consider you have multiple AI agents, and based on user Input you need that particular Agent to be picked up, that is when handoff plays a major role, you can check for more in OpenAI Python SDK

let us here take a use case where an application has multiple language users, based on language user uses, it must respond in same language of user, this is where Handoff can be used

from agents import Agent, Runner

import asyncio

# Define the agents

spanish_agent = Agent(

    name="Spanish agent",
    instructions="You only speak Spanish.",
)

english_agent = Agent(
    name="English agent",
    instructions="You only speak English.",
)


triage_agent = Agent(
    name="Triage agent",
    instructions="Handoff to the appropriate agent based on the language of the request.",
    handoffs=[spanish_agent, english_agent],
)

# Chat history for better context (optional)

chat_history = []


async def chat_with_agent():

    print("👋 Hellowww! Type 'exit' or 'quit' to stop chatting.\n")

    while True:

        user_input = input("You: ")


        if user_input.lower() in ["exit", "quit"]:

            print("Sayonara! 👋")

            break


        elif user_input.lower() == "history":

            print("\n---- Chat History ----")

            for i, (u, b) in enumerate(chat_history):

                print(f"{i+1}. You: {u}\n   Agent: {b}")

            print("----------------------\n")

            continue

        # Send the message to the triage agent

        result = await Runner.run(triage_agent, input=user_input)

        response = result.final_output

        # Save to chat history


        chat_history.append((user_input, response))

        print(f"Agent: {response}\n")

# For Jupyter or IPython:

await chat_with_agent()

here in the code, we have used 3 AI agents , one agent which speaks only English, other Agent which speaks only Spanish and Triagent which actually checks user input and routes to particular agent. In other words, if user speaks english, it responds in English, if user speaks spanish, it responds in spanish

let us run this Agent now to check if our handoff is working fine or not

As seen from above screenshot, it responded exactly way we wanted , hence proving that handoff was successful

AI agent with Tools

Now before we start, understand that this AI Agents far till now were able to use as chatbots, here you can attach Tools to your AI agents, it could be from websearch tool to check real-time data like news, weather forecast, stock analysis, or even image generation tools - you can check list of tools openAI could use here- Open AI Agent tools

Let us use a WebsearchTool for our demo here, we will use weather forecast tool to know real time weather of particular city or place

AI Agent without Tools

from agents import Agent, Runner
import asyncio

# Define the agents
weather_agent = Agent(
    name="Weather agent",
    instructions="You search the internet about current temparature of the city asked by the user and give answer to the user.",

)

# Chat history for better context (optional)
chat_history = []

async def chat_with_agent():
    print("👋 Enter the city you want to find temparature of.\n")

    while True:
        user_input = input("You: ")

        if user_input.lower() in ["exit", "quit"]:
            print("Goodbye! 👋")
            break

        elif user_input.lower() == "history":
            print("\n---- Chat History ----")
            for i, (u, b) in enumerate(chat_history):
                print(f"{i+1}. You: {u}\n   Agent: {b}")
            print("----------------------\n")
            continue

        # Send the message to weather forecast agent
        result = await Runner.run(weather_agent, input=user_input)
        response = result.final_output

        # Save to chat history
        chat_history.append((user_input, response))

        print(f"Agent: {response}\n")

# For Jupyter or IPython:
await chat_with_agent()

and now since it does not have tools, it cant browse the internet , it can give me hallucinations but not accurate response

As you could see, agent says, it cant actively browse the internet and it gave a predicted answer but not accurate

AI Agent with Tools

Now let us attach WebSearchTool to our AI agent and find out if it is able to give us accurate response, and for arguments required, you can use this blog by openAI - OpenAI Websearch Tool

from agents import Agent, Runner, WebSearchTool
import asyncio

# Define the agents
weather_agent = Agent(
    name="Weather agent",
    instructions="You search the internet about current temparature of the city asked by the user and give answer to the user.",
    tools=[WebSearchTool(user_location={"type": "approximate","country" : "IN","city": "Visakhapatnam"})]
)

# Chat history for better context (optional)
chat_history = []

async def chat_with_agent():
    print("👋 Enter the city you want to find temparature of.\n")

    while True:
        user_input = input("You: ")

        if user_input.lower() in ["exit", "quit"]:
            print("Sayonara! 👋")
            break

        elif user_input.lower() == "history":
            print("\n---- Chat History ----")
            for i, (u, b) in enumerate(chat_history):
                print(f"{i+1}. You: {u}\n   Agent: {b}")
            print("----------------------\n")
            continue

        # Send the message to weather forecast agent
        result = await Runner.run(weather_agent, input=user_input)
        response = result.final_output

        # Save to chat history
        chat_history.append((user_input, response))

        print(f"Agent: {response}\n")

# For Jupyter or IPython:
await chat_with_agent()

As you could see, it responded that Porto weather is 10 degree celsius, Now let us verify this in our browser

As you could see, results were accurate and AI agent was able to successfully browse internet for real-time data.

Sagar Kakkala's World

Search This Blog

AI agents - Build AI Agents, Learn Handoffs, Learn to Integrate tools with AI agents | Sagar Kakkala´s World