This is a pretty simple node.js app that serves as a fulfilment service for API.AI agents. This example expects an intent that provides a given-name parameter. You can download this intent and import it if you don'thave one.

1. Deploy this webhook to heroku

   

2. Import this intent to your agent.

3. Activate the Webhook in API.AI with the url provided by heroku and add the header Auth-Token: an-example-token

You're set! Tell your agent My name is James and you should receive a reply Hello James! Welcome from the webhook..

Let's check out the code.

const express = require('express')
const bodyParser = require('body-parser')
const app = express()
app.use(bodyParser.json())
app.set('port', (process.env.PORT || 5000))

You're probably too familiar with these lines. So what we have is a node app, created with express, listening to port 5000 (unless PORT is defined).

app.post('/webhook', function (req, res) {
  ...
}

Now, this is the part that captures the data from API.AI. It's not required to use /webhook as the route but we'll just use it for clarities sake. You can use / if you wanted to. The JSON payload will be stored in req.body.

const REQUIRE_AUTH = true
const AUTH_TOKEN = 'an-example-token'

...

  if (REQUIRE_AUTH) {
    if (req.headers['auth-token'] !== AUTH_TOKEN) {
      return res.status(401).send('Unauthorized')
    }
  }

Here lies a very primitive check where the auth-token header is verified against a single token. Customize this to your needs.

  if (!req.body || !req.body.result || !req.body.result.parameters) {
    return res.status(400).send('Bad Request')
  }

Just to be sure that we receive the information we need, we did some validation here. Again, expand it base on what you needed.

  var userName = req.body.result.parameters['given-name']
  var webhookReply = 'Hello ' + userName + '! Welcome from the webhook.'

  // the most basic response
  res.status(200).json({
    source: 'webhook',
    speech: webhookReply,
    displayText: webhookReply
  })

Here's the main event. We took the value of the parameter given-name and used it for the return response. The response needs to be of Content-Type: application/json, so res.json takes care of that for us.

In an example response in the webhook documentation, we should have a format like:

{
    source: 'source-of-the-response',
    speech: 'Response to the request.',
    displayText: 'Text displayed on the user device screen.'
}

This is the simplest response. You can check out Rich Messages and Response design for Actions on Google for other types of responses.

We of course need to develop and test the webhook before being deployed to the host. Below are the utilities I used.

• Postman - Use it to send some test payloads to the webhook. So I don't need to "talk" using api.ai during initial tests.

• ngrok - Creates a public https tunnel mapped to a port on my machine. It gives me a url that I can use as a webhook for my agent. Use this to actually integrate your agent to the webhook.

• heroku-cli - Well, only if you'll be deploying to heroku.