🤖 React Chat AI Widget

React Chat AI Widget is a plug-and-play React component that lets you embed a fully customizable, intelligent chatbot into your app — with zero backend setup required.

It supports Hugging Face or Replicate APIs for real-time AI responses, making it ideal for customer support, onboarding, FAQs, smart documentation bots, and more.

✨ Features

• 🤖 AI chatbot powered by Hugging Face or Replicate
• ⚡ No backend needed – fully client-side with secure token proxy
• 🎯 Behavior-aware via use cases (e.g. support, documentation, etc.)
• 🧩 Built with React + TypeScript
• 🎨 Fully customizable UI and styles
• 🔐 Tokens are secured — never exposed to the client
• 🧪 Ready for testing with Vitest and Cypress

📦 Installation

npm install react-chat-ai-widget

import { ChatWidget } from "react-chat-ai-widget";

<ChatWidget
  title="InsightFlow"
  direction="left" // or "right" (default)
  huggingface="YOUR_HUGGINGFACE_TOKEN"
  Loader={<Loader />}
  initialQuestions={[{ question: "What is the price of this product?" }]}
  // replicate="YOUR_REPLICATE_TOKEN" (use one or the other)
  config={{
    temperature: 0.7,
    max_tokens: 200,
    top_p: 0.9,
    frequency_penalty: 0.5,
    model: "microsoft/DialoGPT-large",
    stop_sequences: ["\n\n", "Human:", "Bot:"],
  }}
  //replicate and huggingFace has different config
  data={{
    useCase: "customer-support",
    questions: [
      {
        question: "What are your business hours?",
        answer: "We’re open Monday to Friday from 10am to 4pm.",
        category: "general",
        confidence: 0.98,
      },
      {
        question: "Who wrote Don Quixote?",
        answer: "Miguel de Cervantes.",
        category: "literature",
        confidence: 0.92,
      },
    ],
  }}
  chatStyles={{}}
  formStyles={{
    inputStyles: {},
    buttonStyles: {},
    formStyles: {},
  }}
>
  <div>open chat</div>
</ChatWidget>;

🎨 Styling

To ensure the ChatWidget renders properly, import the style file in your main entry point:

import "react-chat-ai-widget/style.css";

You can customize the layout using either class names or inline style objects.

🧩 Component Props

initialQuestions

An array of objects containing the initial questions to be displayed in the chat interface.

🧠 Data Format

Define a list of questions and answers the chatbot can use, with an optional useCase to control tone or behavior.

{
  useCase: "qa-bot", // or "customer-support", "documentation", "strict-compliance"
  questions: [
    {
      question: "Your question here?",
      answer: "The chatbot response.",
      category: "general",
      confidence: 0.85 // Value between 0 and 1
    },
    // Add as many as needed
  ]
}

Supported useCase values

• "customer-support": Friendly, helpful support tone
• "documentation": Technical, concise answers
• "qa-bot": Default Q&A with open logic
• "strict-compliance": Controlled, policy-aware responses

🔐 Token Security

Your API tokens are never exposed to the client.

• ✅ Hugging Face or Replicate tokens are sent securely through a proxy
• ✅ No direct client-to-model communication
• ✅ Fully encrypted and safe

🖼️ Chat Preview

Chat aligned left	Chat aligned right

🔧 Advanced Model Configuration

Take full control of your AI's behavior with comprehensive model customization options:

🎛️ Core Parameters

• temperature (0.0 - 2.0) - Controls response creativity and randomness

• 0.1 = Focused, deterministic responses

• 0.7 = Balanced creativity (recommended)

• 1.5 = Highly creative, varied outputs

• max_tokens (1 - 4096) - Maximum response length

• 50 = Short, concise answers

• 150 = Standard responses

• 500+ = Detailed, comprehensive replies

🎯 Fine-Tuning Options

• top_p (0.0 - 1.0) - Nucleus sampling for response diversity
• frequency_penalty (-2.0 - 2.0) - Reduces repetitive content
• presence_penalty (-2.0 - 2.0) - Encourages topic exploration
• stop_sequences - Custom stop words to control response endings
• model - Choose specific AI models for different capabilities

💡 Example Configuration

config={{
 temperature: 0.7,
 max_tokens: 200,
 top_p: 0.9,
 frequency_penalty: 0.5,
 model: "microsoft/DialoGPT-large",
 stop_sequences: ["\n\n", "Human:", "Bot:"]
}}

⚠️ Notes

• Do not use both huggingface and replicate props at the same time.

• You must import the CSS file for the widget to render correctly.

• The chatStyles and headerStyles props override any class-based styles (chatClassName, headerClassName).

  ⚠️ Important Note for Hugging Face Users If you're using this chatbot with Hugging Face and plan to enable mode (or similar streaming indicators), make sure to set max_tokens to at least 500 in your configuration.

  is necessary because consumes a significant number of tokens during generation. If max_tokens is too low, the assistant's response might get cut off or fail to generate properly.

🧪 Testing

• Includes support for unit and UI testing using Vitest and Cypress.
• Example test commands:

npm run test        # Vitest
npm run dev         # Local dev mode
npm run lint        # ESLint
npm run build       # Production build

🚧 Upcoming Features

🛍️ Product Recommendation Engine

• The chatbot will soon be able to recommend products based on user questions, behavior, and preferences in real time.

• 🌐 Multi-language support

• 🌎 Internationalization (i18n)

• 📊 Analytics Dashboard (Q3 2024) Conversation insights and performance metrics.

• 🔗 Integration Plugins (Q4 2024) Direct integrations with popular platforms (Shopify, WordPress, etc.).

📄 License

MIT © Jorge Ernesto