Custom back end integrations

This document describes how to integrate Chatlayer with your back end or third party services in order to share data gathered in the conversation with the bot, or enrich the bot with data captured before.

How custom integrations work

Chatlayer provides a solution to integrate your conversational agent with your own backend business logic, APIs, and databases to create contextual, personalized and actionable conversational experiences for your users.

A Chatlayer API plugin is available in the Chatlayer’s content management user interface plugin list to enable your conversational agent to create agent messages based on user-specific information and other external data and to redirect your users to different conversational flows based on your own business logic. You can use this solution on any platform that supports receiving and responding to HTTP requests.

Sequence diagram

  1. A user types and sends a text message from a conversational agent interface channel like Facebook Messenger, Web chat, … to Chatlayer.

  2. The received message is going through Chatlayer’s NLP engine to detect the user intention and the returned intent combined with user context will be used to retrieve the next dialog state in the conversation.

  3. Chatlayer’s API plugin can be added in a dialog state to send an API request to your server with different types of static data and/or user session data.

  4. Your server can respond with an object with three fields:

    1. session: A session object for saving retrieved user session data

    2. messages: An array of messages to send back to the interface channel

    3. nextDialogstate: A dialog state identifier to redirect the user to a next dialog state in a conversation flow

Session data will first be stored in the user session so that you can use this data in messages defined in the array of messages or in messages defined in a dialogstate defined as nextDialogstate in the API response. Second, messages will be send, an afterwards the user is redirected to the nextDialogstate. All fields are optional.

API plugin

Chatlayer provides an API plugin in the list of action plugins which you can configure in dialog state in one of your conversational flows.

Sending the API request

The API plugin sends a request to your back end server. It supports different configuration settings:

  1. HTTPS method

  2. API endpoint url

  3. Query parameters

  4. Body payload (JSON)

HTTPS method and API endpoint url

The plugin supports three HTTPS methods

  • GET

  • POST

  • DELETE

Query and body payload parameters

Add query parameters and/or a body payload by defining key value combinations. Each key can have three possible value types:

  • text: static text

  • variable: a user session variable. The value of the variable will be stored as value for the key. Dot and array notation are supported, for example: users[0].firstname

  • dialogstate: select a dialog state from the dropdown. The dialog state id which can be used to redirect the user to a certain dialog state based on your business logic when sending back the API response will be stored as value for the key.

Body payloads are only supported for POST and DELETE requests.

In this example, representing a money transfer, we send five keys in the body payload of an HTTPS POST request to our API endpoint https://arg-demo.glitch.me/transaction.

  • The amount key will have the value of user session variable transfer_amount (ex: 500).

  • The destination key will have the value of user session variable transfer_destination (ex: Elon Musk).

  • The accountType key will have the value of user session variable card_type (ex: savings_account).

  • The transactionSuccess key will have the dialog state identifier for the ‘successful transaction’ dialog state. This identifier can be used in the response of this API request to redirect the user to a new dialog state.

  • The transactionNoMoney key will have the dialog state identifier for the ‘unsuccessful transaction’ dialog state. This identifier can be used in the response of this API request to redirect the user to a new dialog state.

  • The test key will have a value of ‘5’.

This will result in the following body payload

{
"amount": 500,
"destination": "Elon Musk",
"accountType": "savings_account",
"transactionSuccess": "successful transaction",
"transactionNoMoney": "unsuccessful transaction",
"test": 5
}

Multi-language

If the API response will send agent messages back to the user and the agent supports multiple languages, don’t forget to send the user language in the request. The user language is available in the user session variable ‘locale'. Your back end service can use this language setting to send back the response in the user his preferred language.

Sending the API response

Your response to an API plugin request can have three optional keys.

  • session: A session object for saving data in to the user session. The session has two mandatory fields:

    • namespace: a key namespace. The data object will be stored in this namespace key in the user session. You can access this object in Chatlayer by using interpolation: {namespace.dataKey}

    • data: an object which will be saved in the user session data in the namespace key.

  • messages: An array of messages to send back to the user interface channel. The structure of different message types such as text, buttons, quick replies, carousels, lists, media, … is available in the message contract appendix.

  • action: an object defining an action such as redirecting the user to a next dialog state in the conversation.

    • nextDialogstate: A dialog state identifier to redirect the user to a next dialog state in the conversation flow

const result = {
session: {
namespace: 'myNamespace',
data: { variable: 'test123'},
},
messages: [{ type: 'text', textmessages: [{ text: 'random message 1' }] }],
action: {
nextDialogstate: 'dialogstate-123-abc',
},
};

Example

This example demonstrates one API endpoint for transferring an amount of money from an account type (regular or savings) to someone. We will redirected the user to a certain dialog state based on the transaction result.

app.post('/transaction', function (req, res) {
let nextDialogstate;
const { amount, destination, accountType, transactionSuccess, transactionNoMoney = req.body;
// get account type (regular - savings account)
const account = account_synonyms[accountType];
if( accounts[account].amount + accounts[accounthlimit — amount < ) {
nextDialogstate = transactionNoMoney
} else {
// transfer amount
accounts[account].amount —= amount;
nextDialogstate = transactionSuccess;
}
res.json({
action: {
nextDialogstate,
},
session: {
namespace: 'account',
data: {
limit: accounts[account].limit,
amount: accounts[account].amount
}
},
})
});

We receive the body payload object as defined in the Chatlayer API plugin. If the user doesn’t have a sufficient amount of money on his account we set the next dialog state to ‘transactionNoMoney’. Else we subtract the desired amount and set the next dialog state to ‘transactionSuccess'.

As a response for the request we send the next dialogs state to redirect the user to that state and we save the amount of money and the limit of his account in his session data under the namespace account. This data can be used in that next dialog state.

As an alternative solution you could also send that chat message as a response of the API plugin requests by using the messages key.

You can have a look at the code of our mock banking backend here. Feel free to reuse parts of this project to create your own custom Chatlayer integrations.