As Botkit has added support for more and more new platforms, it has become necessary to formalize the process by which incoming messages are received and prepared for use within the bot's brain, and also how outgoing messages are formatted and delivered to the appropriate messaging platform APIs.

The new message pipeline introduced in Botkit 0.6 defines this formal path by exposing a series of middleware endpoints that occur throughout the lifetime of a message as it is is transformed from a "raw" message received from the platform into its normalized form. This normalization process allows Botkit and its array of plugins to handle a message without necessarily caring about its origin.

Message normalization like this has always been part of Botkit - what is being introduced in 0.6 is a formalized process through which all messages must now flow.

This document should be useful to any developer interested in using Botkit's middleware system. However, the focus of this document is on the application of the available middleware endpoints to the process of message normalization in platform connectors. Developers interested in using middleware for building plugins or for feature development should check out the main middleware documentation.

Each messaging service that works with Botkit sends messages in a different format - some even come in multiple formats! Some of the services send messages via incoming webhook, while others send them via web sockets. Some require decryption or other processing before being used by the bot. This means that each different type of bot has to treat the messages differently, depending on all these factors.

The message pipeline is a series of steps that Botkit takes for every single message that tries to manage all this chaos. It is a universal process that is applied to all messages from all platforms that results in a simple normalized message format.

We hope that this will simplify the process of learning to use Botkit, increase the ability of developers to port their bots between platforms or build multi-platform bot apps, and make it easier for plugin developers to create extensions to Botkit that work seamlessly with all messaging services.

Messages that arrive to a Botkit-powered app are now passed into the message processing pipeline via a function called ingest() that receives the raw, unprocessed payload of the incoming message - a big blob of information in some platform specific format.

Somewhere in that blob of information are a few key pieces of information that Botkit really cares about and needs to know about to work: the text, the user, the channel, and the type of the message.

Once ingested, Botkit passes the message through a series of transformations, each one with its own middleware plugin endpoint. The result of these transformations is a message object that is guaranteed to have all of those bits of information in the right places. This transformed message is then passed into Botkit to be handled by your application.

As a result, developers writing normal Botkit code can always expect a message to be in the form:

{
    user: <unique_id_of_sender>,
    channel: <unique_id_of_channel>,
    text: <text, payload or other value, if any>,
    type: <type of message or event>,
    raw_message: <original unmodified version of the message>
}

The same process is reversed when sending a message. Developers can author messages in a generic form, in many cases only specifying the outgoing text. These messages are then transformed by the outgoing message pipeline into the platform-specific API payloads necessary.

The incoming message pipeline consists of these steps:

After passing through the receive step, one of three things will happen:

The outgoing message pipeline consists of these steps:

The pipeline middlewares are useful for two primary reasons:

Middleware functions can be developed to do all sorts of useful things, and can now be tied the pipeline at any phase, depending on the purpose of the middleware.

As of version 0.6, all Botkit platform connectors include middleware functions responsible for their platform-specific transformations. These platform middlewares will always fire first, before any additional middlewares defined by the developer or included in the application code.

Developers can specify as many middlewares as desired, including multiple middleware functions registered to the same endpoints. These functions will fire in the order they are defined.

Plugin middlewares can do things like:

Ingestion into Botkit is the first step in the message pipeline.

Message objects that pass through the ingest phase will have 2 additional fields:

message.raw_message contains the unmodified content of the incoming message payload as received from the messaging service

message._pipeline is an object that tracks a message's progress through the pipeline. The subfield message._pipeline.stage will contain the name of the current pipeline step.

Functions added to the ingest middleware endpoint need to receive 4 parameters, as below.

Field Description
bot an instance of the bot
message the incoming message object
response channel the http response object
next function to call to proceed with pipeline
controller.middleware.ingest.use(function(bot, message, res, next) {

    // define action
    // perhaps set an http status header
    // res.status(200);
    // you can even send an http response
    // res.send('OK');

    // you can access message.raw_message here

    // call next to proceed
    next();

});

The ingest phase is useful for actions like:

Note that in the ingest phase, the message object has not yet been normalized and may not contain the fields you expect. Developers should treat these messages as raw, platform specific messages, and as a result should check the bot.type field, which contains the name of the specific messaging platform, before taking any actions on the message object!

Normalization is the second phase of the message pipeline.

After passing through the normalize phase, the message object is expected to have the following fields:

Note that the entire original unmodified message object will still be available as message.raw_message. Though Botkit's normalization process does not remove any fields from the message as it is normalized, those fields are not guaranteed to be present. It is our recommendation that developers who want to access to platform-specific fields always use the message.raw_message location.

For example, if the originally ingested payload included a field called message.my_platform_value, you should refer to it in your code as message.raw_message.my_platform_value.

Functions added to the normalize middleware endpoint need to receive these parameters:

Field Description
bot an instance of the bot
message the incoming message object
next function to call to proceed with pipeline
controller.middleware.normalize.use(function(bot, message, next) {

    // here's an example of what a message normalizer might do!
    // this is a make believe example not specific to any real platform
    // the idea is to copy/rename or tranform fields from raw_message
    // into an object with {user, text, channel, type} while leaving everything else alone

    // translate a "from" field into message.user
    message.user = message.raw_message.from.id;

    // find the text value and set it in message.text
    message.text = message.raw_message.user_text;

    // make sure a channel value is set
    message.channel = message.raw_message.source_channel

    // call next to proceed
    next();

});

Categorization is the third phase of the message pipeline.

After passing through the catgorize phase, the message object's type field should represent a the final event type that will be handled by Botkit.

The most obvious example of a categorization action is identifying and transforming a message from a generic message_received event into more narrowly defined direct_message, direct_mention, mention or ambient message event.

In addition to changing the message type field, the categorize middleware may also change the value of the message.text field. For example, it should remove direct mentions from the text so that developers do not have to compensate for its possible presence in the input text.

Categorize middlewares can also catch and delegate complex message types to simpler, easier to handle events, taking some of the burden off of developers for handling these subtleties. For example, a categorize middleware might identify different types of button click events that all share the same type value and create new event names for each.

Functions added to the categorize middleware endpoint need to receive these parameters:

Field Description
bot an instance of the bot
message the incoming message object
next function to call to proceed with pipeline
controller.middleware.categorize.use(function(bot, message, next) {

    // messages in Slack that are sent in a 1:1 channel
    // can be identified by the first letter of the channel ID
    // if it is "D", this is a direct_message!
    if (message.type == 'message_received') {
        if (message.channel[0] == 'D') {
            message.type = 'direct_message';
        }
    }

    // call next to proceed
    next();

});

Receive is the final step in the incoming message pipeline before the message actually reaches the bot's internal logic.

By the time a message hits the receive stage, it is in its final form, and is ready to be processed by a Botkit event handler. This middleware endpoint occurs just before a message is evaluated for trigger matches, and before any user-defined handler runs. It will fire for every incoming message, regardless of whether or not it matches a trigger or if any event handlers are registered to receive it.

As noted above, after passing through the receive step, one of three things will happen:

Developers seeking to enrich their messages with data from external sources, such as external NLP services, databases or other third party APIs, may wish to tie this functionality to the receive middleware endpoint. This will cause the enrichment to occur for every single message that is received. This may or may not be desirable, depending on the number and type of messages the platforms send, and the types of messages your bot is supposed to handle.

Before calling any external service in a receive middleware, developers should evaluate the message's type and text field to make sure enrichment is appropriate. For example, you don't want to call an expensive NLP process on messages without text, or messages that represent button clicks.

Alternately, developers may wish to attach their enrichment functionality to the more narrowly defined heard and capture middlewares, which occur after pattern matching has occured, and as a result will only fire for messages that are definitely going to be handled.

Functions added to the categorize middleware endpoint need to receive these parameters:

Field Description
bot an instance of the bot
message the incoming message object
next function to call to proceed with pipeline
controller.middleware.receive.use(function(bot, message, next) {

    // lets call a pretend external NLP system
    // and enrich this message with intent and entity data

    // make sure we have some text, and that this is
    // not a message from the bot itself...
    if (message.text && message.type != 'self_message') {
        callExternalNLP(message.text).then(function(api_results) {
            message.intent = api_results.intent;
            message.entities = api_results.entities;
            next();
        });
    } else {
        next();
    }
});

The heard middleware endpoint occurs after a message has matched a trigger pattern, and is about to be handled. It works just like the receive endpoint, but instead of firing for every incoming message, it will only fire for messages that the is explicitly listening for.

This makes the heard endpoint useful for firing expensive operations, such as database lookups or calls to external APIs tat take a long time, require a lot of processing, or actually cost money to use. However, it makes it less useful for use with NLP tools, since the pattern matching has already occured.

Note that the heard middleware fires only for messages that match handlers set up with controller.hears(), and does not fire with handlers configured with the controller.on() method.

Functions added to the categorize heard endpoint need to receive these parameters:

Field Description
bot an instance of the bot
message the incoming message object
next function to call to proceed with pipeline
controller.middleware.heard.use(function(bot, message, next) {

    // load up any user info associated with this sender
    // using Botkit's internal storage system
    // and enrich the message with a new `user_profile` field
    // now, every message will have the user_profile field automatically
    // and you don't need to load the info in each individual handler function
    controller.storage.users.get(message.user, function(err, user_profile) {
        if (!err && user_profile) {
            message.user_profile = user_profile;
        }

        // call next to proceed, now with additional info!
        next();
    });
});

The capture middleware once again works like the receive or heard endpoints, but fires only on the condition that the incoming message is part of an existing conversation. Generally, this means that the message will actually be handled a callback function passed into convo.ask()

This endpoint is useful for transforming the value used by the conversation to something other than the user's raw input text. For example, in a bot that presents a numbered list of options to a user as part of a multiple choice selection, a capture middleware could be created that transforms the user's numeric input into the full text of the selected item.

When you send a message with bot.send() or bot.reply(), the outgoing message is first sent through the send middleware.

The send middleware receives the raw message, as created in your bot's code. It has not yet been formatted for delivery to the messaging service API. This can be used to modify or track the outgoing messages BEFORE they are formatted for delivery to the platform API.

In particular, this middleware is useful for recording stats about outgoing messages. At this point in the pipeline, the message object will contain the outgoing message text and any attachments, as well as a special message.to field, which represents the unique user id of the message's recipient. Depending on the platform, this value is not always present in the final outgoing message payload.

Any modifications to the content of the outgoing message should happen in a send middleware function. For example, developers can use a send middleware to translate the message text into different languages. Or, developers might pass the message text through a template engine to replace tokens or expand custom shortcodes.

Send middlewares should not make changes to the actual structure or layout of the outgoing message object. Final formatting for delivery to the platform is done by the format endpoint.

Functions added to the categorize send endpoint need to receive these parameters:

Field Description
bot an instance of the bot
message the outgoing message object
next function to call to proceed with pipeline
controller.middleware.send.use(function(bot, message, next) {

    // log the outgoing message for debugging purposes
    console.log('SENDING ', message.text,'TO USER', message.text);

    next();

});

This middleware happens immediately before a message is delivered to the platform API.

Each platform as its own special format for incoming message objects. This middleware should exclusively be used for constructing the final API parameters required for delivering the message. The message object that emerges from this function is intended only for use with the messaging service API.

After being formatted, the resulting platform_message is passed into the platform-specific bot.send() function, which is responsible for the final delivery of the message the appropriate external API endpoint. This allows the bot.send() function to be designed to accept only pre-formatted messages.

Unlike all the other pipeline endpoints, this function does NOT modify the original message object. In fact, the final object is constructed by the middleware in the platform_message parameter, allowing the original message to pass through unmolested.

Functions added to the categorize format endpoint need to receive these parameters:

Field Description
bot an instance of the bot
message the outgoing message object
platform_message the formatted message, ready for delivery
next function to call to proceed with pipeline
controller.middleware.format.use(function(bot, message, platform_message, next) {

    // let's construct an outgoign message payload
    // to an imaginary platform that uses some different fieldnames.

    platform_message.message = {
        text: message.text,
        recipient: message.to,
        room: message.channel,
    }

    platform_message.type = 'message';

    next();

});

Is something missing or out of date?

This file is managed on Github. click here to view the source, and send us a pull request with your improvements!