Botkit - Building Blocks for Building Bots
Botkit is designed to ease the process of designing and running useful, creative bots that live inside Slack, Cisco Spark, Facebook Messenger, Twilio IP Messaging, and other messaging platforms. Support for new platforms is added regularly!
It provides a semantic interface to sending and receiving messages so that developers can focus on creating novel applications and experiences instead of dealing with API endpoints.
Botkit features a comprehensive set of tools to deal with popular messaging platforms, including:
- Slack
- Cisco Spark
- Facebook Messenger
- Twilio IP Messaging
- Microsoft Bot Framework
- Yours? info@howdy.ai
Botkit Studio is a hosted development environment for bots from the same team that built Botkit. Based on feedback from the developer community, as well as experience running our flagship Botkit-powered bot, Howdy, the tools in Botkit Studio allow bot designers and developers to manage many aspects of bot behavior without writing additional code.
Start building your bot with Botkit Studio and you'll start from day one with extra tools and features that help you create and manage a successful bot application. It is also possible to add Studio features to your existing Botkit application. With a few lines of code, you can add access new features and APIs.
Botkit Studio is built on top of Botkit, so everything that works with Botkit continues to just work. All of the available plugins and middleware are compatible!
There are two ways to start a Botkit project:
-
Install the Botkit Studio Starter Kit for Slack or for Cisco Spark or for Facebook and build on top of an already fully functioning bot that comes pre-configured with popular middleware plug-ins and components.
-
Install Botkit directly from NPM or Github and build a new app from scratch, or use one of the included examples as a starting point.
After you've installed Botkit using one of these methods, the first thing you'll need to do is register your bot with a messaging platform, and get a few configuration options set. This will allow your bot to connect, send and receive messages.
If you intend to create a bot that lives in Slack, follow these instructions for attaining a Bot Token.
If you intend to create a bot that lives in Cisco Spark, follow these instructions for configuring your Cisco Spark bot.
If you intend to create a bot that lives in Facebook Messenger, follow these instructions for configuring your Facebook page.
If you intend to create a bot that lives inside a Twilio IP Messaging client, follow these instructions for configuring your app.
If you intend to create a bot that uses Microsoft Bot Framework to send and receive messages, follow these instructions for configuring your app.
Botkit is available via NPM.
npm install --save botkitYou can also check out Botkit directly from Git. If you want to use the example code and included bots, it may be preferable to use Github over NPM.
git clone git@github.com:howdyai/botkit.gitAfter cloning the Git repository, you have to install the node dependencies. Navigate to the root of your cloned repository and use npm to install all necessary dependencies.
npm installUse the --production flag to skip the installation of devDependencies from Botkit. Useful if you just wish to run the example bot.
npm install --productionBots built with Botkit have a few key capabilities, which can be used to create clever, conversational applications. These capabilities map to the way real human people talk to each other.
Bots can hear things, say things and reply to what they hear.
With these two building blocks, almost any type of conversation can be created.
To organize the things a bot says and does into useful units, Botkit bots have a subsystem available for managing multi-message conversations. Conversations add features like the ability to ask a question, queue several messages at once, and track when an interaction has ended. Handy!
After a bot has been told what to listen for and how to respond, it is ready to be connected to a stream of incoming messages. Currently, Botkit supports receiving messages from a variety of sources:
- Slack Real Time Messaging (RTM)
- Slack Incoming Webhooks
- Slack Slash Commands
- Cisco Spark Webhooks
- Facebook Messenger Webhooks
- Twilio IP Messaging
- Microsoft Bot Framework
Read more about connecting your bot to Slack, connecting your bot to Cisco Spark, connecting your bot to Facebook, connecting your bot to Twilio, or connecting your bot to Microsoft Bot Framework
These examples are included in the Botkit Github repo.
slack_bot.js An example bot that can be connected to your team. Useful as a basis for creating your first bot!
spark_bot.js An example bot that can be connected to Cisco Spark. Useful as a basis for creating your first bot!
facebook_bot.js An example bot that can be connected to your Facebook page. Useful as a basis for creating your first bot!
twilio_ipm_bot.js An example bot that can be connected to your Twilio IP Messaging client. Useful as a basis for creating your first bot!
botframework_bot.js An example bot that can be connected to the Microsoft Bot Framework network. Useful as a basis for creating your first bot!
examples/demo_bot.js another example bot that uses different ways to send and receive messages.
examples/team_outgoingwebhook.js an example of a Botkit app that receives and responds to outgoing webhooks from a single team.
examples/team_slashcommand.js an example of a Botkit app that receives slash commands from a single team.
examples/slackbutton_bot.js an example of using the Slack Button to offer a bot integration.
examples/slackbutton_incomingwebhooks.js an example of using the Slack Button to offer an incoming webhook integration. This example also includes a simple form which allows you to broadcast a message to any team who adds the integration.
example/sentiment_analysis.js a simple example of a chatbot using sentiment analysis. Keeps a running score of each user based on positive and negative keywords. Messages and thresholds can be configured.
Here's an example of using Botkit with Slack's real time API, which is the coolest one because your bot will look and act like a real user inside Slack.
This sample bot listens for the word "hello" to be said to it -- either as a direct mention ("@bot hello") or an indirect mention ("hello @bot") or a direct message (a private message inside Slack between the user and the bot).
The Botkit constructor returns a controller object. By attaching event handlers
to the controller object, developers can specify what their bot should look for and respond to,
including keywords, patterns and various messaging and status events.
These event handlers can be thought of metaphorically as skills or features the robot brain has -- each event handler defines a new "When a human says THIS the bot does THAT."
The controller object is then used to spawn() bot instances that represent
a specific bot identity and connection to Slack. Once spawned and connected to
the API, the bot user will appear online in Slack, and can then be used to
send messages and conduct conversations with users. They are called into action by the controller when firing event handlers.
var Botkit = require('botkit');
var controller = Botkit.slackbot({
debug: false
//include "log: false" to disable logging
//or a "logLevel" integer from 0 to 7 to adjust logging verbosity
});
// connect the bot to a stream of messages
controller.spawn({
token: <my_slack_bot_token>,
}).startRTM()
// give the bot something to listen for.
controller.hears('hello',['direct_message','direct_mention','mention'],function(bot,message) {
bot.reply(message,'Hello yourself.');
});As of version 0.4, Botkit records anonymous usage statistics about Botkit bots in the wild. These statistics are used by the Botkit team at Howdy to measure and analyze the Botkit community, and help to direct resources to the appropriate parts of the project.
We take the privacy of Botkit developers and their users very seriously. Botkit does not collect, or transmit any message content, user data, or personally identifiable information to our statistics system. The information that is collected is anonymized inside Botkit and converted using one-way encryption into a hash before being transmitted.
To opt out of the stats collection, pass in the stats_optout parameter when initializing Botkit,
as seen in the example below:
var controller = Botkit.slackbot({
stats_optout: true
});Table of Contents
Once connected to a messaging platform, bots receive a constant stream of events - everything from the normal messages you would expect to typing notifications and presence change events. The set of events your bot will receive will depend on what messaging platform it is connected to.
All platforms will receive the message_received event. This event is the first event fired for every message of any type received - before any platform specific events are fired.
controller.on('message_received', function(bot, message) {
// carefully examine and
// handle the message here!
// Note: Platforms such as Slack send many kinds of messages, not all of which contain a text field!
});Due to the multi-channel, multi-user nature of Slack, Botkit does additional filtering on the messages (after firing message_received), and will fire more specific events based on the type of message - for example, direct_message events indicate a message has been sent directly to the bot, while direct_mention indicates that the bot has been mentioned in a multi-user channel.
List of Slack-specific Events
Similarly, bots in Cisco Spark will receive direct_message events to indicate a message has been sent directly to the bot, while direct_mention indicates that the bot has been mentioned in a multi-user channel. Several other Spark-specific events will also fire. List of Cisco Spark-specific Events
Twilio IPM bots can also exist in a multi-channel, multi-user environment. As a result, there are many additional events that will fire. In addition, Botkit will filter some messages, so that the bot will not receive it's own messages or messages outside of the channels in which it is present. List of Twilio IPM-specific Events
Facebook messages are fairly straightforward. However, because Facebook supports inline buttons, there is an additional event fired when a user clicks a button. List of Facebook-specific Events
Botkit bots receive messages through a system of specialized event handlers. Handlers can be set up to respond to specific types of messages, or to messages that match a given keyword or pattern.
These message events can be handled by attaching an event handler to the main controller object. These event handlers take two parameters: the name of the event, and a callback function which is invoked whenever the event occurs. The callback function receives a bot object, which can be used to respond to the message, and a message object.
// reply to any incoming message
controller.on('message_received', function(bot, message) {
bot.reply(message, 'I heard... something!');
});
// reply to a direct mention - @bot hello
controller.on('direct_mention',function(bot,message) {
// reply to _message_ by using the _bot_ object
bot.reply(message,'I heard you mention me!');
});
// reply to a direct message
controller.on('direct_message',function(bot,message) {
// reply to _message_ by using the _bot_ object
bot.reply(message,'You are talking directly to me');
});In addition to these traditional event handlers, Botkit also provides the hears() function,
which configures event handlers based on matching specific keywords or phrases in the message text.
The hears function works just like the other event handlers, but takes a third parameter which
specifies the keywords to match.
| Argument | Description |
|---|---|
| patterns | An array or a comma separated string containing a list of regular expressions to match |
| types | An array or a comma separated string of the message events in which to look for the patterns |
| middleware function | optional function to redefine how patterns are matched. see Botkit Middleware |
| callback | callback function that receives a message object |
controller.hears(['keyword','^pattern$'],['message_received'],function(bot,message) {
// do something to respond to message
bot.reply(message,'You used a keyword!');
});When using the built in regular expression matching, the results of the expression will be stored in the message.match field and will match the expected output of normal Javascript string.match(/pattern/i). For example:
controller.hears('open the (.*) doors',['message_received'],function(bot,message) {
var doorType = message.match[1]; //match[1] is the (.*) group. match[0] is the entire group (open the (.*) doors).
if (doorType === 'pod bay') {
return bot.reply(message, 'I\'m sorry, Dave. I\'m afraid I can\'t do that.');
}
return bot.reply(message, 'Okay');
});Bots have to send messages to deliver information and present an interface for their functionality. Botkit bots can send messages in several different ways, depending on the type and number of messages that will be sent.
Single message replies to incoming commands can be sent using the bot.reply() function.
Multi-message replies, particularly those that present questions for the end user to respond to,
can be sent using the bot.startConversation() function and the related conversation sub-functions.
Bots can originate messages - that is, send a message based on some internal logic or external stimulus -
using bot.say() method.
All message objects must contain a text property, even if it's only an empty string.
Once a bot has received a message using a on() or hears() event handler, a response
can be sent using bot.reply().
Messages sent using bot.reply() are sent immediately. If multiple messages are sent via
bot.reply() in a single event handler, they will arrive in the client very quickly
and may be difficult for the user to process. We recommend using bot.startConversation()
if more than one message needs to be sent.
You may pass either a string, or a message object to the function.
Message objects may also contain any additional fields supported by the messaging platform in use:
Slack's chat.postMessage API accepts several additional fields. These fields can be used to adjust the message appearance, add attachments, or even change the displayed user name.
This is also true of Facebook. Calls to Facebook's Send API can include attachments which result in interactive "structured messages" which can include images, links and action buttons.
| Argument | Description |
|---|---|
| message | Incoming message object |
| reply | String or Object Outgoing response |
| callback | Optional Callback in the form function(err,response) { ... } |
Simple reply example:
controller.hears(['keyword','^pattern$'],['message_received'],function(bot,message) {
// do something to respond to message
// ...
bot.reply(message,"Tell me more!");
});Slack-specific fields and attachments:
controller.on('ambient',function(bot,message) {
// do something...
// then respond with a message object
//
bot.reply(message,{
text: "A more complex response",
username: "ReplyBot",
icon_emoji: ":dash:",
});
})
//Using attachments
controller.hears('another_keyword','direct_message,direct_mention',function(bot,message) {
var reply_with_attachments = {
'username': 'My bot' ,
'text': 'This is a pre-text',
'attachments': [
{
'fallback': 'To be useful, I need you to invite me in a channel.',
'title': 'How can I help you?',
'text': 'To be useful, I need you to invite me in a channel ',
'color': '#7CD197'
}
],
'icon_url': 'http://lorempixel.com/48/48'
}
bot.reply(message, reply_with_attachments);
});Facebook-specific fields and attachments:
// listen for the phrase `shirt` and reply back with structured messages
// containing images, links and action buttons
controller.hears(['shirt'],'message_received',function(bot, message) {
bot.reply(message, {
attachment: {
'type':'template',
'payload':{
'template_type':'generic',
'elements':[
{
'title':'Classic White T-Shirt',
'image_url':'http://petersapparel.parseapp.com/img/item100-thumb.png',
'subtitle':'Soft white cotton t-shirt is back in style',
'buttons':[
{
'type':'web_url',
'url':'https://petersapparel.parseapp.com/view_item?item_id=100',
'title':'View Item'
},
{
'type':'web_url',
'url':'https://petersapparel.parseapp.com/buy_item?item_id=100',
'title':'Buy Item'
},
{
'type':'postback',
'title':'Bookmark Item',
'payload':'USER_DEFINED_PAYLOAD_FOR_ITEM100'
}
]
},
{
'title':'Classic Grey T-Shirt',
'image_url':'http://petersapparel.parseapp.com/img/item101-thumb.png',
'subtitle':'Soft gray cotton t-shirt is back in style',
'buttons':[
{
'type':'web_url',
'url':'https://petersapparel.parseapp.com/view_item?item_id=101',
'title':'View Item'
},
{
'type':'web_url',
'url':'https://petersapparel.parseapp.com/buy_item?item_id=101',
'title':'Buy Item'
},
{
'type':'postback',
'title':'Bookmark Item',
'payload':'USER_DEFINED_PAYLOAD_FOR_ITEM101'
}
]
}
]
}
}
});
});For more complex commands, multiple messages may be necessary to send a response, particularly if the bot needs to collect additional information from the user.
Botkit provides a Conversation object type that is used to string together several
messages, including questions for the user, into a cohesive unit. Botkit conversations
provide useful methods that enable developers to craft complex conversational
user interfaces that may span a several minutes of dialog with a user, without having to manage
the complexity of connecting multiple incoming and outgoing messages across
multiple API calls into a single function.
Messages sent as part of a conversation are sent no faster than one message per second, which roughly simulates the time it would take for the bot to "type" the message.
| Argument | Description |
|---|---|
| message | incoming message to which the conversation is in response |
| callback | a callback function in the form of function(err,conversation) { ... } |
startConversation() is a function that creates conversation in response to an incoming message.
The conversation will occur in the same channel in which the incoming message was received.
Only the user who sent the original incoming message will be able to respond to messages in the conversation.
| Argument | Description |
|---|---|
| message | message object containing {user: userId} of the user you would like to start a conversation with |
| callback | a callback function in the form of function(err,conversation) { ... } |
startPrivateConversation() is a function that initiates a conversation with a specific user. Note function is currently Slack-only!
| Argument | Description |
|---|---|
| message | incoming message to which the conversation is in response |
| callback | a callback function in the form of function(err,conversation) { ... } |
This works just like startConversation(), with one main difference - the conversation
object passed into the callback will be in a dormant state. No messages will be sent,
and the conversation will not collect responses until it is activated using convo.activate().
Use createConversation() instead of startConversation() when you plan on creating more complex conversation structures using threads or variables and templates in your messages.
This function will cause a dormant conversation created with bot.createConversation() to be activated, which will cause it to start sending messages and receiving replies from end users.
A conversation can be kept dormant in order to preload it with variables, particularly data that requires asynchronous actions to take place such as loading data from a database or remote source. You may also keep a conversation inactive while you build threads, setting it in motion only when all of the user paths have been defined.
| Argument | Description |
|---|---|
| message | String or message object |
Call convo.say() several times in a row to queue messages inside the conversation. Only one message will be sent at a time, in the order they are queued.
controller.hears(['hello world'], 'message_received', function(bot,message) {
// start a conversation to handle this response.
bot.startConversation(message,function(err,convo) {
convo.say('Hello!');
convo.say('Have a nice day!');
});
});| Argument | Description |
|---|---|
| message | String or message object containing the question |
| callback or array of callbacks | callback function in the form function(response_message,conversation), or array of objects in the form { pattern: regular_expression, callback: function(response_message,conversation) { ... } } |
| capture_options | Optional Object defining options for capturing the response |
When passed a callback function, conversation.ask will execute the callback function for any response. This allows the bot to respond to open ended questions, collect the responses, and handle them in whatever manner it needs to.
When passed an array, the bot will look first for a matching pattern, and execute only the callback whose pattern is matched. This allows the bot to present multiple choice options, or to proceed only when a valid response has been received. At least one of the patterns in the array must be marked as the default option, which will be called should no other option match. Botkit comes pre-built with several useful patterns which can be used with this function. See included utterances
Callback functions passed to ask() receive two parameters - the first is a standard message object containing
the user's response to the question. The second is a reference to the conversation itself.
Note that in order to continue the conversation, convo.next() must be called by the callback function. This
function tells Botkit to continue processing the conversation. If it is not called, the conversation will hang
and never complete causing memory leaks and instability of your bot application!
The optional third parameter capture_options can be used to define different behaviors for collecting the user's response.
This object can contain the following fields:
| Field | Description |
|---|---|
| key | String If set, the response will be stored and can be referenced using this key |
| multiple | Boolean if true, support multi-line responses from the user (allow the user to respond several times and aggregate the response into a single multi-line value) |
controller.hears(['question me'], 'message_received', function(bot,message) {
// start a conversation to handle this response.
bot.startConversation(message,function(err,convo) {
convo.ask('How are you?',function(response,convo) {
convo.say('Cool, you said: ' + response.text);
convo.next();
});
})
});controller.hears(['question me'], 'message_received', function(bot,message) {
// start a conversation to handle this response.
bot.startConversation(message,function(err,convo) {
convo.ask('Shall we proceed Say YES, NO or DONE to quit.',[
{
pattern: 'done',
callback: function(response,convo) {
convo.say('OK you are done!');
convo.next();
}
},
{
pattern: bot.utterances.yes,
callback: function(response,convo) {
convo.say('Great! I will continue...');
// do something else...
convo.next();
}
},
{
pattern: bot.utterances.no,
callback: function(response,convo) {
convo.say('Perhaps later.');
// do something else...
convo.next();
}
},
{
default: true,
callback: function(response,convo) {
// just repeat the question
convo.repeat();
convo.next();
}
}
]);
})
});While conversations with only a few questions can be managed by writing callback functions,
more complex conversations that require branching, repeating or looping sections of dialog,
or data validation can be handled using feature of the conversations we call threads.
Threads are pre-built chains of dialog between the bot and end user that are built before the conversation begins. Once threads are built, Botkit can be instructed to navigate through the threads automatically, allowing many common programming scenarios such as yes/no/quit prompts to be handled without additional code.
You can build conversation threads in code, or you can use Botkit Studio's script management tool to build them in a friendly web environment. Conversations you build yourself and conversations managed in Botkit Studio work the same way -- they run inside your bot and use your code to manage the outcome.
If you've used the conversation system at all, you've used threads - you just didn't know it. When calling convo.say() and convo.ask(), you were actually adding messages to the default conversation thread that is activated when the conversation object is created.
| Argument | Description |
|---|---|
| message | String or message object |
| thread_name | String defining the name of a thread |
This function works identically to convo.say() except that it takes a second parameter which defines the thread to which the message will be added rather than being queued to send immediately, as is the case when using convo.say().
| Argument | Description |
|---|---|
| message | String or message object containing the question |
| callback or array of callbacks | callback function in the form function(response_message,conversation), or array of objects in the form { pattern: regular_expression, callback: function(response_message,conversation) { ... } } |
| capture_options | Object defining options for capturing the response. Pass an empty object if capture options are not needed |
| thread_name | String defining the name of a thread |
This function works identically to convo.ask() except that it takes second parameter which defines the thread to which the message will be added rather than being queued to send immediately, as is the case when using convo.ask().
| Argument | Description |
|---|---|
| thread_name | String defining the name of a thread |
Cause the bot to immediately jump to the named thread.
All conversations start in a thread called default, but you may switch to another existing thread before the conversation has been activated, or in a question callback.
Threads are created by adding messages to them using addMessage() and addQuestion()
// create the validation_error thread
convo.addMessage('This is a validation error.', 'validation_error');
convo.addMessage('I am sorry, your data is wrong!', 'validation_error');
// switch to the validation thread immediately
convo.gotoThread('validation_error');| Argument | Description |
|---|---|
| thread_name | String defining the name of a thread |
| message | String or message object |
Like gotoThread(), jumps to the named thread. However, before doing so,
Botkit will first send message to the user as a transition. This allows
developers to specify dynamic transition messages to improve the flow of the
conversation.
// create an end state thread
convo.addMessage('This is the end!', 'the_end');
// now transition there with a nice message
convo.transitionTo('the_end','Well I think I am all done.');You can direct a conversation to switch from one thread to another automatically
by including the action field on a message object. Botkit will switch threads immediately after sending the message.
// first, define a thread called `next_step` that we'll route to...
convo.addMessage({
text: 'This is the next step...',
},'next_step');
// send a message, and tell botkit to immediately go to the next_step thread
convo.addMessage({
text: 'Anyways, moving on...',
action: 'next_step'
});Developers can create fairly complex conversational systems by combining these message actions with conditionals in ask() and addQuestion(). Actions can be used to specify
default or next step actions, while conditionals can be used to route between threads.
From inside a callback function, use convo.gotoThread() to instantly switch to a different pre-defined part of the conversation. Botkit can be set to automatically navigate between threads based on user input, such as in the example below.
bot.createConversation(message, function(err, convo) {
// create a path for when a user says YES
convo.addMessage({
text: 'You said yes! How wonderful.',
},'yes_thread');
// create a path for when a user says NO
convo.addMessage({
text: 'You said no, that is too bad.',
},'no_thread');
// create a path where neither option was matched
// this message has an action field, which directs botkit to go back to the `default` thread after sending this message.
convo.addMessage({
text: 'Sorry I did not understand.',
action: 'default',
},'bad_response');
// Create a yes/no question in the default thread...
convo.ask('Do you like cheese?', [
{
pattern: 'yes',
callback: function(response, convo) {
convo.gotoThread('yes_thread');
},
},
{
pattern: 'no',
callback: function(response, convo) {
convo.gotoThread('no_thread');
},
},
{
default: true,
callback: function(response, convo) {
convo.gotoThread('bad_response');
},
}
]);
convo.activate();
});In addition to routing from one thread to another using actions, you can also use one of a few reserved words to control the conversation flow.
Set the action field of a message to completed to end the conversation immediately and mark as success.
Set the action field of a message to stop end immediately, but mark as failed.
Set the action field of a message to timeout to end immediately and indicate that the conversation has timed out.
After the conversation ends, these values will be available in the convo.status field. This field can then be used to check the final outcome of a conversation. See handling the end of conversations.
Pre-defined conversation threads are great, but many times developers will need to inject dynamic content into a conversation. Botkit achieves this by processing the text of every message using the Mustache template language. Mustache offers token replacement, as well as access to basic iterators and conditionals.
Variables can be added to a conversation at any point after the conversation object has been created using the function convo.setVar(). See the example below.
convo.createConversation(message, function(err, convo) {
// .. define threads which will use variables...
// .. and then, set variable values:
convo.setVar('foo','bar');
convo.setVar('list',[{value:'option 1'},{value:'option 2'}]);
convo.setVar('object',{'name': 'Chester', 'type': 'imaginary'});
// now set the conversation in motion...
convo.activate();
});Given the variables defined in this code sample, foo, a simple string, list, an array, and object, a JSON-style object,
the following Mustache tokens and patterns would be available:
The value of foo is {{vars.foo}}
The items in this list include {{#vars.list}}{{value}}{{/vars.list}}
The object's name is {{vars.object.name}}.
{{#foo}}If foo is set, I will say this{{/foo}}{{^foo}}If foo is not set, I will say this other thing.{{/foo}}
Botkit ensures that your template is a valid Mustache template, and passes the variables you specify directly to the Mustache template rendering system. Our philosophy is that it is OK to stuff whatever type of information your conversation needs into these variables and use them as you please!
| Argument | Description |
|---|---|
| variable_name | The name of a variable to be made available to message text templates. |
| value | The value of the variable, which can be any type of normal Javascript variable |
Create or update a variable that is available as a Mustache template token to all the messages in all the threads contained in the conversation.
The variable will be available in the template as {{vars.variable_name}}
Botkit provides several built in variables that are automatically available to all messages:
{{origin}} - a message object that represents the initial triggering message that caused the conversation.
{{responses}} - an object that contains all of the responses a user has given during the course of the conversation. This can be used to make references to previous responses. This requires that convo.ask() questions include a keyname, making responses available at {{responses.keyname}}
One way to have multi-stage conversations is with multiple functions which call each other. Each function asks just one question. Example:
controller.hears(['pizzatime'], 'message_received', function(bot,message) {
var askFlavor = function(err, convo) {
convo.ask('What flavor of pizza do you want?', function(response, convo) {
convo.say('Awesome.');
askSize(response, convo);
convo.next();
});
};
var askSize = function(response, convo) {
convo.ask('What size do you want?', function(response, convo) {
convo.say('Ok.')
askWhereDeliver(response, convo);
convo.next();
});
};
var askWhereDeliver = function(response, convo) {
convo.ask('So where do you want it delivered?', function(response, convo) {
convo.say('Ok! Good bye.');
convo.next();
});
};
bot.startConversation(message, askFlavor);
});The full code for this example can be found in examples/convo_bot.js.
| Pattern Name | Description |
|---|---|
| bot.utterances.yes | Matches phrases like yes, yeah, yup, ok and sure. |
| bot.utterances.no | Matches phrases like no, nah, nope |
In order to direct the flow of the conversation, several helper functions are provided. These functions should only be called from within a convo.ask handler function!
convo.sayFirst(message) Works just like convo.say, but injects a message into the first spot in the queue
so that it is sent immediately, before any other queued messages.
convo.stop() end the conversation immediately, and set convo.status to stopped
convo.repeat() repeat the last question sent and continue to wait for a response.
convo.silentRepeat() simply wait for another response without saying anything.
convo.next() proceed to the next message in the conversation. This must be called at the end of each handler.
convo.setTimeout(timeout) times out conversation if no response from user after specified time period (in milliseconds).
Conversations trigger events during the course of their life. Currently, only two events are fired, and only one is very useful: end.
Conversations end naturally when the last message has been sent and no messages remain in the queue.
In this case, the value of convo.status will be completed. Other values for this field include active, stopped, and timeout.
convo.on('end',function(convo) {
if (convo.status=='completed') {
// do something useful with the users responses
var res = convo.extractResponses();
// reference a specific response by key
var value = convo.extractResponse('key');
// ... do more stuff...
} else {
// something happened that caused the conversation to stop prematurely
}
});Returns an object containing all of the responses a user sent during the course of a conversation.
var values = convo.extractResponses();
var value = values.key;Return one specific user response, identified by its key.
var value = convo.extractResponse('key');| Argument | Description |
|---|---|
| message | A message object |
| callback | Optional Callback in the form function(err,response) { ... } |
Slack-specific Example:
bot.say(
{
text: 'my message text',
channel: 'C0H338YH4' // a valid slack channel, group, mpim, or im ID
}
);Note: If your primary need is to spontaneously send messages rather than respond to incoming messages, you may want to use Slack's incoming webhooks feature rather than the real time API.
Facebook-specific Example:
bot.say(
{
text: 'my message_text',
channel: '+1(###)###-####' // a valid facebook user id or phone number
}
);The functionality of Botkit can be extended using middleware functions. These functions can plugin to the core bot running processes at several useful places and make changes to both a bot's configuration and the incoming or outgoing message.
For information about existing middleware plugins, see here
Botkit currently supports middleware insertion in the following places:
- When receiving a message, before triggering any events
- When sending a message, before the message is sent to the API
- When hearing a message
- When matching patterns with
hears(), after the pattern has been matched but before the handler function is called - When capturing a users response to a
convo.ask()question, after the user has answered, but before the value is stored or passed to the handler function
Send and Receive middleware functions are added to Botkit using an Express-style "use" syntax. Each function receives a bot parameter, a message parameter, and a next function which must be called to continue processing the middleware stack.
Hear middleware functions are passed in to the controller.hears function,
and override the built in regular expression matching.
Receive middleware can be used to do things like preprocess the message content using external natural language processing services like Wit.ai. Additional information can be added to the message object for use down the chain.
controller.middleware.receive.use(function(bot, message, next) {
// do something...
// message.extrainfo = 'foo';
next();
});Send middleware can be used to do things like preprocess the message content before it gets sent out to the messaging client.
controller.middleware.send.use(function(bot, message, next) {
// do something useful...
if (message.intent == 'hi') {
message.text = 'Hello!!!';
}
next();
});Hear middleware can be used to change the way Botkit bots "hear" triggers. It can be used to look for values in fields other than message.text, or use comparison methods other than regular expression matching. For example, a middleware function could enable Botkit to "hear" intents added by an NLP classifier instead of string patterns.
Hear middleware is enabled by passing a function into the hears() method on the Botkit controller.
When specified, the middleware function will be used instead of the built in regular expression match.
These functions receive 2 parameters - patterns an array of patterns, and message the incoming
message. This function will be called after any receive middlewares, so may use any additional
information that may have been added. A return value of true indicates the pattern has been
matched and the bot should respond.
// this example does a simple string match instead of using regular expressions
function custom_hear_middleware(patterns, message) {
for (var p = 0; p < patterns.length; p++) {
if (patterns[p] == message.text) {
return true;
}
}
return false;
}
controller.hears(['hello'],'direct_message',custom_hear_middleware,function(bot, message) {
bot.reply(message, 'I heard the EXACT string match for "hello"');
});It is possible to completely replace the built in regular expression match with
a middleware function by calling controller.changeEars(). This will replace the matching function used in hears()
as well as inside convo.ask(). This would, for example, enable your bot to
hear only intents instead of strings.
controller.changeEars(function(patterns, message) {
// ... do something
// return true or false
});Heard middleware can be used to modify or enrich a message with additional information before it is handled by the callback function. This can be useful for developers who want to use NLP tools, but want to limit the type and number of messages sent to be classified. It is also useful for developers who want to mix internal application data (for example, user account information) into messages.
Whereas the receive middleware will fire for every single incoming message of any type, the heard middleware only fires when a pattern has already been matched.
Heard middleware functions fire anytime Botkit attempts to match a pre-defined pattern: when using the hears() feature, and also when using convo.ask() to capture user responses.
controller.middleware.heard.use(function(bot, message, next) {
// load internal user data and add it to the message
mydb.users.find({id: message.user}, function(err, user_record) {
// amend the message with a new field.
// this will now be available inside the normal handler function
message.internal_user = user_record;
// call next or else execution will stall
next();
});
});As users respond to questions posed using convo.ask(), their answers will first be passed through any capture middleware endpoints.
The capture middleware can modify the message in any way, including changing the value that will be used to test pre-defined patterns
and that will ultimately be stored as the final user answer.
This can be particularly useful when used in conjunction with a natural language processing API. NLP plugins like IBM Watson and Microsoft LUIS typically provide 2 features: translation of raw user text into a pre-defined intent, and extraction of structured data from the raw string into entities.
Another instance in which this is useful is when used in conjunction with buttons and quick replies that, in addition to displayed text may also carry a hidden payload value. Developers can use this middleware endpoint to capture the payload instead of the displayed text.
The capture middleware endpoint allows developers to harness these values and capture them instead of or in addition to the raw user text.
Please note that the signature of the capture middleware is slightly different than the other endpoints, as it includes a parameter for the conversation object:
controller.middleware.capture.use(function(bot, message, convo, next) {
// user's raw response is in message.text
// instead of capturing the raw response, let's capture the intent
if (message.intent) {
message.text = message.intent;
}
// what if there is a hidden payload? let's use that instead
if (message.payload) {
message.text = message.payload;
}
// what if there are entities too? we can use them as part of the conversation...
if (message.entities) {
for (var e = 0; e < message.entities.length; e++) {
convo.setVar(message.entities[e].name, message.entities[e].value);
}
}
// always call next!
next();
});Botkit has a built in storage system used to keep data on behalf of users and teams between sessions. Botkit uses this system automatically when storing information for Slack Button applications (see below).
By default, Botkit will use json-file-store to keep data in JSON files in the filesystem of the computer where the bot is executed. (Note this will not work on Heroku or other hosting systems that do not let node applications write to the file system.) Initialize this system when you create the bot:
var controller = Botkit.slackbot({
json_file_store: 'path_to_json_database'
});This system supports freeform storage on a team-by-team, user-by-user, and channel-by-channel basis. Basically controller.storage is a key value store. All access to this system is through the following twelve functions. Example usage:
controller.storage.users.save({id: message.user, foo:'bar'}, function(err) { ... });
controller.storage.users.get(id, function(err, user_data) {...});
controller.storage.users.delete(id, function(err) {...});
controller.storage.users.all(function(err, all_user_data) {...});
controller.storage.channels.save({id: message.channel, foo:'bar'}, function(err) { ... });
controller.storage.channels.get(id, function(err, channel_data) {...});
controller.storage.channels.delete(id, function(err) {...});
controller.storage.channels.all(function(err, all_channel_data) {...});
controller.storage.teams.save({id: message.team, foo:'bar'}, function(err) { ... });
controller.storage.teams.get(id, function(err, team_data) {...});
controller.storage.teams.delete(id, function(err) {...});
controller.storage.teams.all(function(err, all_team_data) {...});Note that save must be passed an object with an id. It is recommended to use the team/user/channel id for this purpose.
[user/channel/team]_data will always be an object while all_[user/channel/team]_data will always be a list of objects.
If you want to use a database or do something else with your data, you can write your own storage module and pass it in.
Make sure your module returns an object with all the methods. See simple_storage.js for an example of how it is done! Make sure your module passes the test in storage_test.js.
Then, use it when you create your bot:
var controller = Botkit.slackbot({
storage: my_storage_provider
})By default, your bot will log to the standard JavaScript console object
available in Node.js. This will synchronously print logging messages to stdout
of the running process.
There may be some cases, such as remote debugging or rotating of large logs,
where you may want a more sophisticated logging solution. You can write your
own logging module that uses a third-party tool, like
winston or
Bristol. Just create an object with a
log method. That method should take a severity level (such as 'error' or
'debug') as its first argument, and then any number of other arguments that
will be logged as messages. (Both Winston and Bristol create objects of this
description; it's a common interface.)
Then, use it when you create your bot:
var controller = Botkit.slackbot({
logger: new winston.Logger({
levels: winston.config.syslog.levels
transports: [
new (winston.transports.Console)(),
new (winston.transports.File)({ filename: './bot.log' })
]
})
});Note: with Winston, we must use the syslog.levels over the default or else some botkit log messages (like 'notice') will not be logged properly.
##Use Botkit with an Express web server Instead of controller.setupWebserver(), it is possible to use a different web server to manage authentication flows, as well as serving web pages.
Here is an example of using an Express web server alongside Botkit.
You can get an invite here: http://dev4slack.xoxco.com/.