iceflinger / discord-jambot

Jambot modular Discord bot

discord jambot

Discord-jambot is a modular bot platform built for the discord chat service. It's a conceptual port of my modular IRC bot of the same kind, with some enhancements to better suit discord-specific concepts and take advantage of the full Discord API.

Modules

All jambot really is is a system for configuring certain module features to load into certain contexts with certain settings, as well as a shared database for them to use. Simply put, there is a single config file describing what jambot is doing where. Modules can be loaded into a global, server-specific, or channel-specific context, and will only receive API events specific to that context. If a module is loaded into a narrower (channel) scope than a module of the same type, the narrow-scope module will receive the API events instead of the wider-scoped one. This is designed to let you either only run certain features in certain channels, or to run features globally and configure them differently for specific channels.

Config

As mentioned, there's a single config file for a single jambot instance. Other than the following required top-level configs, any other settings in this file represent where a module is loaded context-wise as well as settings for that specific module's instance in that context. A more robust example of config format is in the sample_jambot.yml file.

version: discord-jambot 0.1
token: reallylongdiscordtoken
owner: 123456789
database: jambot.db
command_prefix: '>'

Database

Modules can run SQL queries against the shared SQLite DB Instance using the async client.db_query(query) and client.db_commit() methods on the bot-client object passed to module hooks. Modules should make any required table setup/checks in the on_connect() hook.

Features

Mirrorkov2

Pretty much the main feature this bot was intended for, a community-learning conversation bot using primitive markov chains. It uses pairs of 2 words as nodes and builds a graph based on the observed frequencies of their orderings as it learns, and then traverses that graph to generate a reply. The name of this specific version comes from the most recently added feature of changing graph weights based on the contexts learned from the user triggering the response (mirroring them). A disadvantage of this feature is only being able to learn directly from discord channels: the legacy markov3 module can learn from an arbitrary uploaded text file.

Commands:
*>feed {lines} {channelId} // Learn x lines from the history of channelId, default to current channel
*>words //check how many total words known in current model
*>known {word} //check for how many learned context include a specific word
*>clean //halve the weights of all currently learned contexts; future learned ones will be twice as likely as past onces going forward

Twitch

Dedicate a discord channel to announcing the beginning of streams on twitch accounts linked to any members of the server. No need to keep track of any lists of users; any linked twitch accounts get announced (and can be filtered away through config)

Others

Theres plenty of other modules on the modules folder but many are either obsolete or just one-off things; go ahead and take a look and ask me on discord at Ice#6667 if you have any questions.