@llamaindex/chat-ui

Chat UI components for LLM apps

Overview

@llamaindex/chat-ui is a React component library that provides ready-to-use UI elements for building chat interfaces in LLM (Large Language Model) applications. This package is designed to streamline the development of chat-based user interfaces for AI-powered applications.

Installation

To install the package, run the following command in your project directory:

npm install @llamaindex/chat-ui

Features

• Pre-built chat components (e.g., message bubbles, input fields)
• Minimal styling, fully customizable with Tailwind CSS
• TypeScript support for type safety
• Easy integration with LLM backends

Usage

1. Install the package

npm install @llamaindex/chat-ui

2. Configure your tailwind.config.ts to include the chat-ui components

module.exports = {
  content: [
    'app/**/*.{ts,tsx}',
    'node_modules/@llamaindex/chat-ui/**/*.{ts,tsx}',
  ],
  // ...
}

3. Import the components and use them

The easiest way to get started is to connect the whole ChatSection component with useChat hook from vercel/ai:

import { ChatSection } from '@llamaindex/chat-ui'
import { useChat } from 'ai/react'

const ChatExample = () => {
  const handler = useChat()
  return <ChatSection handler={handler} />
}

Component Composition

Components are designed to be composable. You can use them as is:

import { ChatSection } from '@llamaindex/chat-ui'
import { useChat } from 'ai/react'

const ChatExample = () => {
  const handler = useChat()
  return <ChatSection handler={handler} />
}

Or you can extend them with your own children components:

import { ChatSection, ChatMessages, ChatInput } from '@llamaindex/chat-ui'
import LlamaCloudSelector from './components/LlamaCloudSelector' // your custom component
import { useChat } from 'ai/react'

const ChatExample = () => {
  const handler = useChat()
  return (
    <ChatSection handler={handler}>
      <ChatMessages />
      <ChatInput>
        <ChatInput.Preview />
        <ChatInput.Form className="bg-lime-500">
          <ChatInput.Field type="textarea" />
          <ChatInput.Upload />
          <LlamaCloudSelector /> {/* custom component */}
          <ChatInput.Submit />
        </ChatInput.Form>
      </ChatInput>
    </ChatSection>
  )
}

Your custom component can use the useChatUI hook to send additional data to the chat API endpoint:

import { useChatInput } from '@llamaindex/chat-ui'

const LlamaCloudSelector = () => {
  const { requestData, setRequestData } = useChatUI()
  return (
    <div>
      <select
        value={requestData?.model}
        onChange={e => setRequestData({ model: e.target.value })}
      >
        <option value="llama-3.1-70b-instruct">Pipeline 1</option>
        <option value="llama-3.1-8b-instruct">Pipeline 2</option>
      </select>
    </div>
  )
}

Styling

chat-ui components are based on shadcn components using Tailwind CSS.

You can override the default styles by changing CSS variables in the globals.css file of your Tailwind CSS configuration. For example, to change the background and foreground colors:

:root {
  --background: 0 0% 100%;
  --foreground: 222.2 84% 4.9%;
}

For a list of all available CSS variables, please refer to the Shadcn Theme Config.

Additionally, you can also override each component's styles by setting custom classes in the className prop. For example, setting the background color of the ChatInput.Form component:

import { ChatSection, ChatMessages, ChatInput } from '@llamaindex/chat-ui'
import { useChat } from 'ai/react'

const ChatExample = () => {
  const handler = useChat()
  return (
    <ChatSection handler={handler}>
      <ChatMessages />
      <ChatInput>
        <ChatInput.Preview />
        <ChatInput.Form className="bg-lime-500">
          <ChatInput.Field type="textarea" />
          <ChatInput.Upload />
          <ChatInput.Submit />
        </ChatInput.Form>
      </ChatInput>
    </ChatSection>
  )
}

Example

See the example app for a complete example.

License

@llamaindex/chat-ui is released under the MIT License.

Support

If you encounter any issues or have questions, please file an issue on our GitHub repository.