If you have explicitly opted in to simulations (they are disabled by default!), the following process happens behind the scenes to facilitate making simulations of your messages:

Mini splits every message you send into individual words and takes note of how often certain words come after other words. For example, for the words I'm doing, the next word might be homework, well, dishes... Mini assigns each word to a number and stores that frequency in a database. The actual words are individually encrypted and stored separately from their frequencies. When you request a simulation, Mini goes down a random path of connected nodes until it reaches either the length limit or the end of a message. Only at this point will Mini decrypt the words in that path so you get a readable result back. Messages are never stored in full, they are always broken up into words, making it impossible to trace back what you said, when you said it or what the context was.

Mini does not collect any message data on you by default. To opt in to simulations, open the Settings panel with /mini settings and select the Simulations tab. You will then have the option Opt In to enable the creation of your language model (as outlined in the "How it works" section), at which point all future messages you send in channels with Mini in them will be analyzed.

If you want to exclude a channel from simulations explicitly, while in that text channel, you can head to /mini settings again and open the Simulations tab. Select the option Exclude Channel and messages sent in that channel will not be used for your language model in the future. Note that this does not retroactively remove any data from the language model as the data is no longer tied to its origin text channel after it's processed.

If you want to delete the data collected about your messages, simply open the /mini settings page once more to the Simulations tab. Click Opt Out and confirm your choice, and the data is erased immediately. There is no way for anybody to recover it after that point.