Web-Dev-For-Beginners/translations/en/9-chat-project/README.md

Chat Project

This chat project demonstrates how to build a Chat Assistant using GitHub Models.

Here's what the completed project looks like:

To provide some context, building Chat Assistants with generative AI is an excellent way to start learning about AI. In this lesson, you'll learn how to integrate generative AI into a web app. Let's get started.

Connecting to Generative AI

For the backend, we're using GitHub Models. It's a fantastic service that allows you to use AI for free. Visit its playground and grab the code corresponding to your preferred backend language. Here's what it looks like at GitHub Models Playground.

As mentioned, select the "Code" tab and your preferred runtime.

Using Python

In this example, we select Python, which means we use the following code:

"""Run this model in Python

> pip install openai
"""
import os
from openai import OpenAI

# To authenticate with the model you will need to generate a personal access token (PAT) in your GitHub settings. 
# Create your PAT token by following instructions here: https://docs.github.com/en/authentication/keeping-your-account-and-data-secure/managing-your-personal-access-tokens
client = OpenAI(
    base_url="https://models.github.ai/inference",
    api_key=os.environ["GITHUB_TOKEN"],
)

response = client.chat.completions.create(
    messages=[
        {
            "role": "system",
            "content": "",
        },
        {
            "role": "user",
            "content": "What is the capital of France?",
        }
    ],
    model="openai/gpt-4o-mini",
    temperature=1,
    max_tokens=4096,
    top_p=1
)

print(response.choices[0].message.content)

Let's clean up this code a bit to make it reusable:

def call_llm(prompt: str, system_message: str):
    response = client.chat.completions.create(
        messages=[
            {
                "role": "system",
                "content": system_message,
            },
            {
                "role": "user",
                "content": prompt,
            }
        ],
        model="openai/gpt-4o-mini",
        temperature=1,
        max_tokens=4096,
        top_p=1
    )

    return response.choices[0].message.content

With this call_llm function, we can now pass a prompt and a system prompt, and the function will return the result.

Customizing the AI Assistant

To customize the AI assistant, you can define its behavior by modifying the system prompt like this:

call_llm("Tell me about you", "You're Albert Einstein, you only know of things in the time you were alive")

Exposing It via a Web API

Great, we've completed the AI part. Now, let's see how to integrate it into a Web API. For the Web API, we'll use Flask, but any web framework should work. Here's the code:

Using Python

# api.py
from flask import Flask, request, jsonify
from llm import call_llm
from flask_cors import CORS

app = Flask(__name__)
CORS(app)   # *   example.com

@app.route("/", methods=["GET"])
def index():
    return "Welcome to this API. Call POST /hello with 'message': 'my message' as JSON payload"


@app.route("/hello", methods=["POST"])
def hello():
    # get message from request body  { "message": "do this taks for me" }
    data = request.get_json()
    message = data.get("message", "")

    response = call_llm(message, "You are a helpful assistant.")
    return jsonify({
        "response": response
    })

if __name__ == "__main__":
    app.run(host="0.0.0.0", port=5000)

In this code, we create a Flask API and define two routes: the default route "/" and "/chat". The "/chat" route is intended for the frontend to send questions to the backend.

To integrate llm.py, here's what we need to do:

• Import the call_llm function:

  from llm import call_llm
  from flask import Flask, request
  

• Use it in the "/chat" route:

  @app.route("/hello", methods=["POST"])
  def hello():
     # get message from request body  { "message": "do this taks for me" }
     data = request.get_json()
     message = data.get("message", "")
  
     response = call_llm(message, "You are a helpful assistant.")
     return jsonify({
        "response": response
     })
  

  In this step, we parse the incoming request to extract the message property from the JSON body. Then, we call the LLM with this:

  response = call_llm(message, "You are a helpful assistant")
  
  # return the response as JSON
  return jsonify({
     "response": response 
  })
  

Great, now we've completed the necessary steps.

Configuring CORS

It's important to set up CORS (Cross-Origin Resource Sharing). Since our backend and frontend will run on different ports, we need to allow the frontend to communicate with the backend.

Using Python

Here's a piece of code in api.py that sets this up:

from flask_cors import CORS

app = Flask(__name__)
CORS(app)   # *   example.com

Currently, it's configured to allow all origins ("*"), which is not secure. This should be restricted when moving to production.

Running Your Project

To run your project, start the backend first, followed by the frontend.

Using Python

Now that we have llm.py and api.py, how do we make the backend work? There are two steps:

• Install dependencies:

  cd backend
  python -m venv venv
  source ./venv/bin/activate
  
  pip install openai flask flask-cors openai
  

• Start the API:

  python api.py
  

  If you're using Codespaces, go to the Ports section at the bottom of the editor, right-click on it, select "Port Visibility," and choose "Public."

Working on the Frontend

Now that the API is up and running, let's create a frontend for it. We'll start with a basic frontend and improve it step by step. In a frontend folder, create the following:

backend/
frontend/
index.html
app.js
styles.css

Let's begin with index.html:

<html>
    <head>
        <link rel="stylesheet" href="styles.css">
    </head>
    <body>
      <form>
        <textarea id="messages"></textarea>
        <input id="input" type="text" />
        <button type="submit" id="sendBtn">Send</button>  
      </form>  
      <script src="app.js" />
    </body>
</html>    

The above code is the bare minimum needed to support a chat window. It includes a textarea for displaying messages, an input field for typing messages, and a button to send messages to the backend. Next, let's look at the JavaScript in app.js.

app.js

// app.js

(function(){
  // 1. set up elements  
  const messages = document.getElementById("messages");
  const form = document.getElementById("form");
  const input = document.getElementById("input");

  const BASE_URL = "change this";
  const API_ENDPOINT = `${BASE_URL}/hello`;

  // 2. create a function that talks to our backend
  async function callApi(text) {
    const response = await fetch(API_ENDPOINT, {
      method: "POST",
      headers: { "Content-Type": "application/json" },
      body: JSON.stringify({ message: text })
    });
    let json = await response.json();
    return json.response;
  }

  // 3. add response to our textarea
  function appendMessage(text, role) {
    const el = document.createElement("div");
    el.className = `message ${role}`;
    el.innerHTML = text;
    messages.appendChild(el);
  }

  // 4. listen to submit events
  form.addEventListener("submit", async(e) => {
    e.preventDefault();
   // someone clicked the button in the form
   
   // get input
   const text = input.value.trim();

   appendMessage(text, "user")

   // reset it
   input.value = '';

   const reply = await callApi(text);

   // add to messages
   appendMessage(reply, "assistant");

  })
})();

Here's a breakdown of the code:

1. We get references to all the elements we'll use later in the code.
2. This section defines a function that uses the built-in fetch method to call our backend.
3. The appendMessage function adds both the user's messages and the assistant's responses to the textarea.
4. We listen for the submit event, read the input field, display the user's message in the textarea, call the API, and render the response in the textarea.

Next, let's look at styling. You can get creative here, but here are some suggestions:

styles.css

.message {
    background: #222;
    box-shadow: 0 0 0 10px orange;
    padding: 10px:
    margin: 5px;
}

.message.user {
    background: blue;
}

.message.assistant {
    background: grey;
} 

These three classes style messages differently depending on whether they come from the assistant or the user. For inspiration, check out the solution/frontend/styles.css folder.

Changing the Base URL

One thing we haven't set yet is the BASE_URL. This depends on where your backend is running. To set it:

• If you're running the API locally, it should be something like http://localhost:5000.
• If you're using Codespaces, it will look something like "[name]app.github.dev".

Assignment

Create your own folder named project with the following structure:

project/
  frontend/
    index.html
    app.js
    styles.css
  backend/
    ...

Copy the content from the instructions above, but feel free to customize it to your liking.

Solution

Solution

Bonus

Try changing the personality of the AI assistant.

For Python

When calling call_llm in api.py, you can modify the second argument to customize the assistant's behavior. For example:

call_llm(message, "You are Captain Picard")

Frontend

You can also customize the CSS and text to your liking by editing index.html and styles.css.

Summary

Congratulations! You've learned how to create a personal assistant using AI from scratch. We accomplished this using GitHub Models, a Python backend, and a frontend built with HTML, CSS, and JavaScript.

Setting Up with Codespaces

• Navigate to: Web Dev For Beginners repo.

• Create a new repository from the template (make sure you're logged into GitHub) by clicking the button in the top-right corner:

  

• Once you're in your repository, create a Codespace:

  

  This will launch an environment where you can start working.

---

Disclaimer:
This document has been translated using the AI translation service Co-op Translator. While we aim for accuracy, please note that automated translations may include errors or inaccuracies. The original document in its native language should be regarded as the authoritative source. For critical information, professional human translation is advised. We are not responsible for any misunderstandings or misinterpretations resulting from the use of this translation.