Botkit - Building Blocks for Building Bots
Botkit is designed to ease the process of designing and running useful, creative bots that live inside Slack, Facebook Messenger and other messaging platforms.
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
- Facebook Messenger
- Yours? info@howdy.ai
Botkit is available via NPM.
bash
npm install --save botkit
You 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 --productionAfter you've installed Botkit, 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.
The fastest way to get a bot online and get to work is to start from one of the examples included in the repo.
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 Facebook Messenger, follow these instructions for configuring your Facebook page.
Bots 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. Bots can 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
- Facebook Messenger Webhooks
Read more about connecting your bot to Slack or connecting your bot to Facebook
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!
facebook_bot.js An example bot that can be connected to your Facebook page. 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/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 say 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.');
});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_recieved), 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
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.
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 | 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();
}
}
]);
})
});
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_recieved', function(bot,message) {
askFlavor = function(response, convo) {
convo.ask('What flavor of pizza do you want?', function(response, convo) {
convo.say('Awesome.');
askSize(response, convo);
convo.next();
});
}
askSize = function(response, convo) {
convo.ask('What size do you want?', function(response, convo) {
convo.say('Ok.')
askWhereDeliver(response, convo);
convo.next();
});
}
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.
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.
Botkit currently supports middleware insertion in three 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
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
});
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 nine 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.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.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.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({
transports: [
new (winston.transports.Console)(),
new (winston.transports.File)({ filename: './bot.log' })
]
})
});##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/.