rasa-webchat

Rasa Webchat

A simple webchat widget to connect with a chatbot 💬platform. Originally forked from react-chat-widget and optimized for Rasa and Botfront.

Features

• Text Messages
• Quick Replies
• Images and Videos
• Markdown support
• Easy to import in a script tag or as a React Component
• Persistent sessions
• Typing indications
• Smart delay between messages

Setup

In a <script> tag

In your <body/>:

<div id="webchat"/>
<script src="https://storage.googleapis.com/mrbot-cdn/webchat-latest.js"></script>
// Or you can replace latest with a specific version
<script>
  WebChat.default.init({
    selector: "#webchat",
    initPayload: "/get_started",
    customData: {"userId": "123"}, // arbitrary custom data. Stay minimal as this will be added to the socket
    socketUrl: "http://localhost:5500",
    socketPath: "/socket.io/",
    title: "Title",
    subtitle: "Subtitle",
    inputTextFieldHint: "Type a message...",
    connectingText: "Waiting for server...",
    hideWhenNotConnected: true,
    fullScreenMode: false,
    showFullScreenButton: false,
    profileAvatar: "http://to.avat.ar",
    openLauncherImage: 'myCustomOpenImage.png',
    closeLauncherImage: 'myCustomCloseImage.png',
    displayUnreadCount: true, // --> [view](./assets/unread_count_pastille.png)
    showMessageDate: false,
    tooltipPayload: '/get_tooltip',
    tooltipDelay: 1000,
    customMessageDelay: (message) => {
      if (message.length > 100) return 2000;
      return 1000;
    },
    params: {
      images: {
        dims: {
          width: 300,
          height: 200,
        }
      },
      storage: "local"
    },
  })
</script>

About images: width and height define the size in pixels that images in messages are crop-scaled to. If not present, the image will scale to the maximum width of the container and the image.

It is recommended to use a particular version (i.e. "webchat-.js") however the file "webchat-latest.js" is also available and is updated continuously with the latest version.

As a React component

Install the package from GitHub by running:

npm install rasa-webchat

Then once it is installed it can be implemented as follows.

import { Widget } from 'rasa-webchat';

function CustomWidget = () => {
  return (
    <Widget
      initPayload={"/get_started"}
      socketUrl={"http://localhost:5500"}
      socketPath={"/socket.io/"}
      customData={{"userId": "123"}} // arbitrary custom data. Stay minimal as this will be added to the socket
      title={"Title"}
      inputTextFieldHint={"Type a message..."}
      connectingText={"Waiting for server..."}
      hideWhenNotConnected
      connectOn={"mount"}
      embedded={true}
      showFullScreenButton={false}
      openLauncherImage="myCustomOpenImage.png"
      closeLauncherImage="myCustomCloseImage.png"
      displayUnreadCount={true} // --> [view](./assets/unread_count_pastille.png)
      showMessageDate={false} // display message date, can use fonction as (timestamp, message) => return 'my custom date'
      tooltipPayload='/get_tooltip'
      tooltipDelay={1000}
      customMessageDelay{(message) => {
        if (message.length > 100) return 2000;
        return 1000;
      }}
      params={{
        images: {
          dims: {
            width: 300,
            height: 200
          }
        },
        storage: "local"
      }}
      customComponent={ (messageData) => (<div>Custom React component</div>) }
    />
  )
}

• Make sure to have the prop embedded set to true if you don't want to see the launcher.

Backend

Rasa Core

Use the SocketIOInput channel: See instructions in the Rasa Core documentation

If you want to process customData in Rasa Core you have to create a new channel. Use channel rasa_core.channels.socketio as a template for your new channel. In such channel customData can be retrieved via data['customData']. Then you can modify sender_id, save customData to the database, fill slots or whatever you need to with your additional data.

Others

Your backend must expose a socket with socket.io

Receiving messages from the chat

@socketio.on('user_uttered')
    def handle_message(message):
        # do something

Sending messages from the backend to the chat widget

sending plain text

emit('bot_uttered', {"text": "hello"}, room=session_id)

sending quick replies

message = {
  "text": "Happy?",
  "quick_replies":[
    {"title":"Yes", "payload":"/affirm"},
    {"title":"No", "payload":"/deny"}
  ]}
emit('bot_uttered', message, room=socket_id)

sending a link Snippet

Admittedly a bit far fetched, thinking that Snippets would evolve to carousels of generic templates :)

message = {
  "attachment":{
    "type":"template",
    "payload":{
      "template_type":"generic",
      "elements":[
        {
          "title":"Title",
          "buttons":[ {
            "title":"Link name",
            "url": "http://link.url"
          }
        ]
      }
    ]
  }
}
}
emit('bot_uttered', message, room=socket_id)

sending a Video Message

message = {
  "attachment":{
    "type":"video",
    "payload":{
      "title":"Link name",
      "src": "https://www.youtube.com/watch?v=f3EbDbm8XqY"
    }
  }
}
emit('bot_uttered', message, room=socket_id)

sending an Image Message

message = {
      "attachment":{
        "type":"image",
        "payload":{
          "title":"Link name",
          "src": "https://encrypted-tbn0.gstatic.com/images?q=tbn:ANd9GcT_IX5FSDQLrwm9qvuXu_g7R9t_-3yBSycZ8OxpRXvMDaTAeBEW"
        }
      }
    }
emit('bot_uttered', message, room=socket_id)

sending a message with custom data

message = {
      "data":{
        "customField1": 'anything you want',
        "customField2": 'other custom data, 
      }
    }
emit('bot_uttered', message, room=socket_id)

sending a message to be displayed as a tooltip

You first need to set a tooltipPayload in the props of the component, then, for the answer to that payload, you should define a response with a

object and a property tooltip = true. This message will then be displayed as a tooltip before the widget is opened. This works with Botfront, but not yet with vanilla Rasa.

The prop tooltipDelay lets you set a delay before calling the payload. It default to 500ms.

message = {
  "text": "Hi!",
  "metadata":{
    "tooltip": true
   }
 }
emit('bot_uttered', message, room=socket_id)

Usage

Session Persistence

storage specifies the location where the the conversation and state of the WebChat is stored in the browser's storage.

storage: "session" defines the state to be stored in the session storage. The session storage persists on reload of the page, and is cleared after the browser or tab is closed, or when sessionStorage.clear()is called.

storage: "local" defines the state to be stored in the local stoage. The local storage persists after the the browser is closed, and is cleared when the cookies of the browser are cleared, or when localStorage.clear()is called.

Sending a message on page load

When reconnecting to an existing chat session, the bot will send a message contained in the localStorage key specified by the NEXT_MESSAGE constant. The message should be stringified JSON with a message property describing the message and an expiry property set to a UNIX timestamp in milliseconds after which this message should not be sent. This is useful if you would like your bot to be able to offer your user to navigate around the site.

docViewer

Note : this is an experimental feature

If you add this prop to the component or to the init script, docViewer=true , this will treat links in received messages as links to a document ( .pdf .doc .xlsx etc. ) and will open them in a popup using https://docs.google.com/viewer service

connectOn

This prop lets you choose when the widget will try connecting to the server. By default, it tries connecting as soon as it mounts. If you select connectOn='open' it will only attempt connection when the widget is opened. it can only take the values mount and open.

onSocketEvent

This prop lets you call custom code on a specific socket event. Here is what it should look like.

onSocketEvent={{
  'bot_uttered': () => console.log('the bot said something'),
  'connect': () => console.log('connection established'),
  'disconnect': () => doSomeCleanup(),
}}

tooltip payload

This feature lets you set a tooltipPayload in the props of the component, then, for the answer to that payload, you should define a response with a metada object and a property tooltip = true. This message will then be displayed as a tooltip before the widget is opened. Disclaimer: This works with Botfront, but not yet with vanilla Rasa. Don't use that feature if you didn't set a metadata tag in your response.

message = {
  "text": "Hi!",
  "metadata":{
    "tooltip": true
   }
 }
emit('bot_uttered', message, room=socket_id)

tooltipDelay

This prop is as number, it lets you set a delay in milliseconds before calling the tooltip payload. It defaults to 500ms.

customMessageDelay

This prop is a function, the function take a message string as an argument. The defined function will be called everytime a message is received and the returned value will be used as a milliseconds delay before displaying a new message. This is the default value

(message) => {
    let delay = message.length * 30;
    if (delay > 2 * 1000) delay = 3 * 1000;
    if (delay < 400) delay = 1000;
    return delay;
}

API

Method	Description
WebChat.toggle()	Toggle the open/close state of the chat window, send initPayload if webchat is not initialized and is toggled open
WebChat.open()	Open the chat window, send initPayload if webchat is not initialized
WebChat.close()	Close the chat window
WebChat.isOpen()	Get the open/closed state of the widget
WebChat.show()	Show the chat widget, send initPayload if the chat is in open state and not initialized
WebChat.hide()	Hide the chat widget
WebChat.isVisible()	Get the shown/hidden state of the widget

Styles

hierarchy:

.conversation-container
  |-- .header
        |-- .title
        |-- .close-function
        |-- .loading
  |-- .messages-container
        |-- .message
              |-- .client
              |-- .response
        |-- .replies
              |-- .reply
              |-- .response
        |-- .snippet
              |-- .snippet-title
              |-- .snippet-details
              |-- .link
        |-- .imageFrame
        |-- .videoFrame
  |-- .sender
        |-- .new-message
        |-- .send

Class	Description
.widget-container	The div containing the chatbox of the default version
.widget-embedded	div of the embedded chatbox (using embedded prop)
.full-screen	div of the fullscreen chatbox (using fullScreenMode prop)
.conversation-container	the parent div containing the header, message-container and sender
.messages-container	the central area where the messages appear
.sender	div of the bottom area which prompts user input
.new-message	the text input element of sender
.send	the send icon element of sender
.header	div of the top area with the chatbox header
.title	the title element of the header
.close-button	the close icon of the header
.loading	the loading status element of the header
.message	the boxes holding the messages of client and response
.replies	the area that gives quick reply options
.snippet	a component for describing links
.imageFrame	a container for sending images
.videoFrame	a container for sending video

Usage with Docker

Since you have to install the package from GitHub, npm will clone the repo to the global .npm directory before building the module in your node_modules directory. For this reason docker will have trouble installing the package, of course the global .npm directory doesn't exist in the container. To solve this simply add the following line in your Dockerfile before the RUN npm install command

RUN mkdir -p /root/.npm

Contributors

@PHLF @znat @TheoTomalty @Hub4IT @dliuproduction @MatthieuJnon @mofortin