node package manager


Build a markov model of everything Hubot hears.

Hubot Markov Model

Generates a markov model based on everything that your Hubot sees in your chat.


  1. Add hubot-markov to your package.json with npm install --save hubot-markov:
    "hubot-markov": "~1.6.0"
  1. Require the module in external-scripts.json:
  1. Run npm update and restart your Hubot.


Saying anything at all in chat appends to the model. The robot is always watching!

Hubot: markov will randomly generate text based on the current contents of its model.

Hubot: markov your mother is a will generate a random phrase seeded with the phrase you give it. This command might output "your mother is a classy lady", for example. Remember: Hubot is an innocent soul, and what he says only acts as a mirror for everything in your hearts.

Hubot: remarkov and Hubot: mmarkov are similar, but traverse node transitions in different directions: remarkov chains backwards from a given ending state, and mmarkov chains both forward and backward.


The Hubot markov model can optionally be configured by four environment variables:

HUBOT_MARKOV_PLY controls the order of the model that's built; effectively, how many previous states (words) are considered to choose the next state. You can bump this up if you'd like, but the default of 1 is both economical with storage and maximally hilarious.

HUBOT_MARKOV_LEARN_MIN controls the minimum length of a phrase that will be used to train the model, default 1. Set this higher to avoid training your model with a bunch of immediate terminal transitions like "lol".

HUBOT_MARKOV_GENERATE_MAX controls the maximum size of a markov chain that will be generated by the "markov" command.

HUBOT_MARKOV_RESPOND_CHANCE controls the chance that Hubot will respond un-prompted to a message it sees by using the last word in the message as the seed. Set this to a value between 0 and 1.0 to enable the feature. Leaving this variable unset or setting it to 0 will disable the feature.

HUBOT_MARKOV_NOREVERSE, if set, will prevent the reverse model from being generated. This saves some space in your database, but doesn't let you use remarkov or mmarkov.

HUBOT_MARKOV_IGNORELIST is interpreted as a comma-separated list of usernames to ignore for purposes of markov indexing. You can use this to prevent the output of other bots or integrations from clogging up your model. By default, Hubot sees all messages.