On this tutorial, we’ll stroll by find out how to construct a customized Chatbot utility that can permit us to ask questions and obtain high-quality solutions. The bot will keep in mind earlier prompts, simulating context-aware dialog.

A GIF animation showing our finished bot in action

Chatbots have develop into indispensable instruments for companies and builders in search of to enhance buyer interactions and streamline consumer experiences in at this time’s quickly evolving digital panorama.

OpenAI’s ChatGPT has reworked from a cutting-edge experiment right into a powerhouse in chatbot growth. Its meteoric rise to success is nothing wanting outstanding, charming customers worldwide.

The demo code of this mission is obtainable on CodeSandbox. You’ll have to offer your personal OpenAI API key within the .env file to check it reside. To get one, create an account on the OpenAI, log in, navigate to the API keys and generate a brand new API key.

Desk of Contents

Planning Options and UI

Our utility will likely be primarily based on React, and we’ll use OpenAI API to entry the information and use CSS modules for styling.

Using React will permit us to create a dynamic and responsive consumer interface, enhancing the general consumer expertise.

The OpenAI API will allow us to acquire entry to superior language processing capabilities, offering information for creating insightful interactions.

Moreover, CSS modules will permit us to keep up a modular design, facilitating environment friendly growth and customization of the app.

The options we’ll be implementing embrace:

  • A chosen enter space the place customers will be capable to craft prompts, inviting contextually related inquiries.
  • A Submit button that can permit customers to submit their prompts to the API, initiating the dialog course of.
  • Message objects that will likely be showcased as chat-style messages inside the dialog window, enhancing the interactive chat expertise.
  • Message objects to show ChatGPT replies that can present a conversational stream.
  • A Historical past function that can record all the consumer’s latest prompts. This may also permit customers to revisit earlier conversations.
  • A Clear button that can permit the removing of generated content material, providing a clear slate for brand spanking new conversations.

The picture under exhibits our component-based wireframe.

A wireframe of the app's interface

The entire utility will likely be wrapped in the primary container, which can maintain all the components collectively. It is going to be additional divided right into a two-column structure.

The primary column will embrace all the messages from the consumer and ChatGPT. On the backside of the column, there will likely be an enter space and a button for submitting the immediate.

The second column will maintain the historical past of all the latest prompts. On the backside of the column, there will likely be a Clear button that can permit the consumer to wipe the generated content material.

Choosing a Colour Scheme

The applying design will prioritize the convenience of content material notion. It will permit us to offer a few vital advantages:

  • Customers will be capable to rapidly comprehend the introduced info, resulting in a extra intuitive and user-friendly expertise.
  • It should additionally improve accessibility, making certain that people of various backgrounds and skills will be capable to simply navigate and have interaction with the content material.

The picture under exhibits our shade scheme.

Our five-color scheme: black, dark gray, lime-green, peach and white

The background of the applying will likely be black, whereas the messages, historical past objects, and enter kind will likely be darkish grey.

The textual content on the messages and enter backgrounds will likely be white, offering a pleasant distinction and make textual content simple to learn.

To provide the app some highlights, the column titles, Submit button, and response message avatars will use a brilliant, lime-green tone.

To accent the Clear button, a gentle purple tone will likely be used. This may also assist customers keep away from clicking the button by accident.

Setting Up the React App

We’ll use create-react-app to create our utility. Run npx create-react-app react-chatgpt to create a brand new React mission.

Watch for a minute for the setup to finish, after which change the working listing to the newly created folder by cd react-chatgpt and run npm begin to start out the developer server.

This could open up our mission in our default browser. If not, navigate to http://localhost:3000 to open it manually. We needs to be introduced with the React welcome display, as pictured under.

React welcome screen

Including International Kinds

We’ll add world styling to ascertain a constant and unified visible look throughout all elements of the applying.

Open index.css and embrace the next styling guidelines:

@import url("https://fonts.googleapis.com/css2?household=Varela+Spherical&show=swap");

* {
  margin: 0;
  padding: 0;
  box-sizing: border-box;
  font-family: "Varela Spherical", sans-serif;
}

physique {
  background-color: #121212;
}

First, we import the Varela Round font and set the entire app to make use of it.

We additionally take away any pre-defined margins and paddings, in addition to set box-sizing to border-box so the app seems the identical on completely different browsers.

Lastly, we set the background of the physique to a darkish tone, which permits us to focus on the content material of the applying.

We’ll want a few avatars to symbolize the authors of the messages from the consumer and OpenAI API. This fashion, they’ll be simpler to differentiate.

Create a brand new icons folder contained in the src listing and embrace the bot.png and consumer.png icons.

You possibly can obtain samples from icons listing here, or you need to use customized ones from websites like FlatIcon or Icons8, so long as you retain the above file names.

Constructing the Elements

First, we’d like a well-organized file construction that matches the wireframe design.

We’ll use the terminal to create the required folder and part information. Every part may have its personal JavaScript file for performance and CSS file for styling.

Change the working listing within the src folder by working cd src after which run the next command:

mkdir elements && cd elements && contact Message.js Message.module.css Enter.js Enter.module.css Historical past.js Historical past.module.css Clear.js Clear.module.css

The command above will first create a /elements/ folder, then change the working listing to it, and create all the required information inside it.

The Message part

The Message part will show consumer prompts and API responses inside the dialog, facilitating the real-time trade of knowledge between the consumer and the chatbot.

Open the Message.js file and embrace the next code:

import bot from "../icons/bot.png";
import consumer from "../icons/consumer.png";

import kinds from "./Message.module.css";

export default perform Message({ function, content material }) {
  return (
    <div className={kinds.wrapper}>
      <div>
        <img
          src={function === "assistant" ? bot : consumer}
          className={kinds.avatar}
          alt="profile avatar"
        />
      </div>
      <div>
        <p>{content material}</p>
      </div>
    </div>
  );
}

First, we import the downloaded icons for avatars after which import the exterior CSS guidelines for styling.

After that, we create the wrapper for the Message part, which can include each icons and textual content content material.

We use the function prop within the conditional to show the suitable avatar because the picture src.

We additionally use the content material prop, which will likely be handed in because the textual content response from the OpenAI API and consumer enter immediate.

Now let’s model the part so it seems like a chat message! Open the Message.module.css file and embrace the next guidelines:

.wrapper {
  show: grid;
  grid-template-columns: 60px auto;
  min-height: 60px;
  padding: 20px;
  margin-bottom: 20px;
  border-radius: 10px;
  background-color: #1b1b1d;
}

.avatar {
  width: 40px;
  top: 40px;
}

We divide the structure into two columns, with the avatars proven within the fixed-width container on the precise and the textual content on the left.

Subsequent, we add some padding and margin to the underside of the message. We additionally model the message to have spherical borders and set the background to darkish grey.

Lastly, we set the avatar icon to a set width and top.

The Enter part

The Enter part will likely be an interface component designed to seize consumer queries, serving because the means by which customers work together and have interaction with the chatbot.

Open the Enter.js file and embrace the next code:

import kinds from "./Enter.module.css";

export default perform Enter({ worth, onChange, onClick }) {
  return (
    <div className={kinds.wrapper}>
      <enter
        className={kinds.textual content}
        placeholder="Your immediate right here..."
        worth={worth}
        onChange={onChange}
      />
      <button className={kinds.btn} onClick={onClick}>
        Go
      </button>
    </div>
  );
}

We first import the exterior stylesheet to model the part.

We return the part wrapper that features the enter subject for the consumer prompts and the button to submit it to the API.

We set the placeholder worth to be displayed when the enter kind is empty, and create the worth prop to carry the entered immediate, in addition to the onChange prop that will likely be known as as soon as the enter worth adjustments.

For the button, the onClick prop will likely be known as as soon as the consumer clicks on the button.

Now let’s model the part in order that the enter space seems lovely and the consumer is inspired to offer prompts! Open the Enter.module.css file and embrace the next guidelines:

.wrapper {
  show: grid;
  grid-template-columns: auto 100px;
  top: 60px;
  border-radius: 10px;
  background-color: #323236;
}

.textual content {
  border: none;
  define: none;
  background: none;
  padding: 20px;
  shade: white;
  font-size: 16px;
}

.btn {
  border: none;
  border-radius: 0 10px 10px 0;
  font-size: 16px;
  font-weight: daring;
  background-color: rgb(218, 255, 170);
}

.btn:hover {
  cursor: pointer;
  background-color: rgb(200, 253, 130);
}

We set the wrapper to be divided into two columns, with a set width for the button and the remainder of the accessible width devoted to the enter space.

We additionally outline the particular top of the part, set the rounded borders for it, and set the background to darkish grey.

For the enter space, we take away the default border, define, background and add some padding. We set the textual content shade to white and set a particular font measurement.

The Historical past part

The Historical past part will show the sequence of previous consumer and chatbot interactions, offering customers with a contextual reference of their dialog.

Open the Historical past.js file and embrace the next code:

import kinds from "./Historical past.module.css";

export default perform Historical past({ query, onClick }) {
  return (
    <div className={kinds.wrapper} onClick={onClick}>
      <p>{query.substring(0, 15)}...</p>
    </div>
  );
}

We first import the exterior model guidelines for the part. Then we return the wrapper that can embrace the textual content.

The textual content worth will likely be handed in as a query prop from the consumer immediate, and solely the primary 15 characters of the textual content string will likely be displayed.

Customers will likely be allowed to click on on the historical past objects, and we’ll cross the onClick prop to manage the press habits.

Now let’s model the part to make sure it’s visually interesting and matches properly within the sidebar! Open the Historical past.module.css file and embrace the next guidelines:

.wrapper {
  padding: 20px;
  margin-bottom: 20px;
  border-radius: 10px;
  background-color: #1b1b1d;
}

.wrapper:hover {
  cursor: pointer;
  background-color: #323236;
}

We set some padding, add the margin to the underside, and set the rounded corners for the historical past objects. We additionally set the background shade to darkish grey.

As soon as the consumer hovers over the merchandise, the cursor will change to a pointer and the background shade will change to a lighter shade of grey.

The Clear part

The Clear part will likely be a UI component designed to reset or clear the continuing dialog, offering customers with a fast strategy to begin a brand new interplay with out navigating away from the present interface.

Open the Clear.js file and embrace the next code:

import kinds from "./Clear.module.css";

export default perform Clear({ onClick }) {
  return (
    <button className={kinds.wrapper} onClick={onClick}>
      Clear
    </button>
  );
}

We first import the exterior stylesheet to model the part.

We return the button that can permit customers to clear the content material of the applying. We’ll cross the onClick prop to attain the specified habits.

Now let’s model the part to make it stand out and scale back the possibilities of customers urgent it by accident! Open the Clear.module.css file and embrace the next guidelines:

.wrapper {
  width: 100%;
  top: 60px;
  background-color: #ff9d84;
  border: none;
  border-radius: 10px;
  font-size: 16px;
  font-weight: daring;
}

.wrapper:hover {
  cursor: pointer;
  background-color: #ff886b;
}

We set the button to fill the accessible width of the column, set the particular top, and set the background shade to gentle purple.

We additionally take away the default border, set the rounded corners, set a particular font measurement, and make it daring.

On hover, the cursor will change to a pointer and the background shade will change to a darker shade of purple.

Constructing the Person Interface

Within the earlier part, we constructed all the needed elements. Now let’s put them collectively and construct the consumer interface for the applying.

We’ll configure their performance to create a purposeful and interactive chatbot interface with organized and reusable code.

Open the App.js file and embrace the next code:

import { useState } from "react";

import Message from "./elements/Message";
import Enter from "./elements/Enter";
import Historical past from "./elements/Historical past";
import Clear from "./elements/Clear";

import "./kinds.css";

export default perform App() {
  const [input, setInput] = useState("");
  const [messages, setMessages] = useState([]);
  const [history, setHistory] = useState([]);

  return (
    <div className="App">
      <div className="Column">
        <h3 className="Title">Chat Messages</h3>
        <div className="Content material">
          {messages.map((el, i) => {
            return <Message key={i} function={el.function} content material={el.content material} />;
          })}
        </div>
        <Enter
          worth={enter}
          onChange={(e) => setInput(e.goal.worth)}
          onClick={enter ? handleSubmit : undefined}
        />
      </div>
      <div className="Column">
        <h3 className="Title">Historical past</h3>
        <div className="Content material">
          {historical past.map((el, i) => {
            return (
              <Historical past
                key={i}
                query={el.query}
                onClick={() =>
                  setMessages([
                    { role: "user", content: history[i].query },
                    { function: "assistant", content material: historical past[i].reply },
                  ])
                }
              />
            );
          })}
        </div>
        <Clear onClick={clear} />
      </div>
    </div>
  );
}

First, we import the useState hook that we’ll use to trace the information state for the applying. Then we import all of the elements we constructed and the exterior stylesheet for styling.

Then we create the enter state variable to retailer the consumer immediate enter, messages to retailer the dialog between the consumer and ChatGPT, and historical past to retailer the historical past of consumer prompts.

We additionally create the primary wrapper for the entire app that can maintain two columns.

Every column may have a title and content material wrapper that can embrace the dialog messages, enter space, and Submit button for the primary column and historical past objects and the Clear button for the second column.

The dialog messages will likely be generated by mapping by the messages state variable and the historical past objects — by mapping by the historical past state variable.

We set the enter onChange prop to replace the enter state variable every time consumer enters any worth within the enter kind.

As soon as the consumer clicks the Ship button, the consumer immediate will likely be despatched to the OpenAI API to course of and obtain the reply.

For the historical past objects, we set the onClick prop in order that the messages state variable will get up to date to the particular immediate and reply.

Lastly, for the Clear button, we cross the onClick prop a perform that can clear each the message and historical past values, clearing the applying information.

Creating the App Format

On this part, we’ll prepare the consumer interface elements to create an intuitive construction for efficient consumer interplay.

Open App.css and embrace the next styling guidelines:

.App {
  show: grid;
  grid-template-columns: auto 200px;
  hole: 20px;
  max-width: 1000px;
  margin: 0 auto;
  min-height: 100vh;
  padding: 20px;
}

.Column {
  shade: white;
}

.Title {
  padding: 20px;
  margin-bottom: 20px;
  border-radius: 10px;
  shade: black;
  background-color: rgb(218, 255, 170);
}

.Content material {
  top: calc(100vh - 200px);
  overflow-y: scroll;
  margin-bottom: 20px;
}

::-webkit-scrollbar {
  show: none;
}

We cut up the primary app wrapper into two columns, separated by a gap through the use of CSS grid layout, and we set the left column for historical past objects to a set width.

Subsequent, we set the wrapper to by no means exceed a sure width, middle it on the display, make it use all the display viewport top, and add some padding inside it.

For every column’s contents, we set the textual content shade to white.

For the column titles, we set some padding, add the underside margin, and set the rounded corners. We additionally set the title component background shade to lime-green and set the textual content shade to black.

We additionally model the columns themselves by setting the rule that the content material shouldn’t exceed a sure top and set the content material to be scrollable if it reaches exterior the peak. We additionally add a margin to the underside.

We additionally disguise the scrollbars, in order that we don’t need to model them to override the default values for every browser. This rule is optionally available and we may go away it out.

Getting the API Key from OpenAI

When you haven’t already arrange your personal API key for the Sandbox within the introduction of this tutorial, ensure to create an account on the OpenAI web site.

Subsequent, log in and navigate to the API keys and generate a brand new API key.

setting up an api key

Copy the important thing to the clipboard and open your mission.

Create a brand new .env file in your mission root and paste the worth for the next key like so:

REACT_APP_OPENAI_API_KEY=paste-your-code-here

Making ready the Request Name to OpenAI API

By the OpenAI API, our chatbot will be capable to ship textual prompts to the OpenAI server, which can then course of the enter and generate human-like responses.

That is achieved by leveraging a strong language mannequin that’s been educated on various textual content sources. By offering the mannequin with a dialog historical past and the present consumer immediate, our chatbot will obtain context-aware responses from the API.

On this part, we’ll put together the request and implement the decision to the API to obtain the response and set the information to the state variable we outlined earlier.

Open the App.js once more and add the next code:



export default perform App() {
  

  const handleSubmit = async () => {
    const immediate = {
      function: "consumer",
      content material: enter,
    };

    setMessages([...messages, prompt]);

    await fetch("https://api.openai.com/v1/chat/completions", {
      methodology: "POST",
      headers: {
        Authorization: `Bearer ${course of.env.REACT_APP_OPENAI_API_KEY}`,
        "Content material-Kind": "utility/json",
      },
      physique: JSON.stringify({
        mannequin: "gpt-3.5-turbo",
        messages: [...messages, prompt],
      }),
    })
      .then((information) => information.json())
      .then((information) => {
        const res = information.decisions[0].message.content material;
        setMessages((messages) => [
          ...messages,
          {
            role: "assistant",
            content: res,
          },
        ]);
        setHistory((historical past) => [...history, { question: input, answer: res }]);
        setInput("");
      });
  };

  const clear = () => {
    setMessages([]);
    setHistory([]);
  };

  return <div className="App">
}

First, we create a separate handleSubmit perform, which will likely be executed as soon as the consumer has entered the immediate within the enter kind and clicks the Submit button.

Inside handleSubmit, we first create the immediate variable that can maintain the function consumer and the immediate itself as an object. The function is vital as a result of, when storing our messages, we’ll have to know which of them are consumer messages.

Then we replace the messages state variable with the consumer immediate.

Subsequent, we make an precise fetch name to the api.openai.com/v1/chat/completions endpoint to entry the information from the OpenAI API.

We specify that it’s a POST request, and set the headers with the authorization token and the content material sort. For the physique parameters, we specify which API mannequin to make use of, and we cross the messages variable because the content material from the consumer.

As soon as the response is acquired, we retailer it within the res variable. We add the thing consisting of the function assistant and the response itself to the message state variable.

We additionally replace the historical past state variable with the thing, with the query and corresponding reply because the keys.

After the response is acquired and state variables are up to date, we clear the enter state variable to arrange the enter kind for the subsequent consumer immediate.

Lastly, we create a easy clear perform to clear the messages and historical past state variables, permitting the consumer to clear the information of the applying.

Testing the Software

At this level, we should always have created a totally purposeful chat utility! The very last thing left to do is to check it.

First, let’s attempt to ask ChatGPT a single query.

A question asked via our new app

The animation above exhibits a query being submitted and a solution being acquired.

Now let’s attempt to create a dialog.

Submitting multiple questions

As proven within the animation above, the chatbot remembers the context from the earlier messages, so we will communicate with it whereas being totally context-aware.

Now let’s see what occurs as soon as we click on on the Historical past button.

Clicking on the History button

Discover how the chat switches to the respective consumer immediate and reply. This may very well be helpful if we wish to resume the dialog from a particular level.

Lastly, let’s click on on the Clear button.

Clicking on the Clear button

As anticipated, the contents of the app are cleared. This can be a helpful choice when there’s a variety of content material and the consumer needs to start out contemporary.

Conclusion

On this tutorial, we’ve realized find out how to create an easy-to-use consumer interface, find out how to construction our code by way of elements, find out how to work with states, find out how to make API calls, and find out how to course of the acquired information.

With the mix of superior pure language processing capabilities of the OpenIAI API and the flexibleness of React, you’ll now be capable to create refined chatbot functions that you could customise additional to your liking.

Discover that this tutorial shops the API key on the frontend, which could not be safe for manufacturing. If you wish to deploy the mission, it could be advisable to create an Express server and use the API key there.

Additionally, if you’d like the historical past prompts to be accessible after the subsequent preliminary launch, you possibly can retailer after which learn them from local storage, and even join a database to your app and retailer and skim information from there.