I’ve messed with Discord bots in servers quite a bit and I’ve always had fun with the various game bots that I’ve encountered. I’ve played with complex bots that play Cards Against Humanity, and relatively simple bots like one that emulates an 8 ball. All the 8 ball did was respond with ‘Yes’ or ‘No’ at random, and even a bot that simple is very fun when played with friends. I wanted to get into the Discord bot creation game to see how these things tick.

The Idea

I wanted to create an interactive trivia bot that wasn’t frustrating to use. I’ve played with trivia bots in the past and while some of them have great Q/A, sometimes they can be a bit rigid and unforgiving with misspellings. I love the Warcraft universe so I made a Warcraft themed bot and named it after a character named ‘Lorewalker Cho’, which as the name suggests, is a character that knows much about the universe’s lore.


First and foremost the bot needed to be concurrent. Discord bots can be running in thousands of Discord servers at once and the bot needs to be responsive when invoked. With this requirement it would make sense to use a language or runtime that supports concurrency well like node, golang or .NET.

I chose to write the bot in Python 3.5 as this version of python has first-class support for async/await and it’s supported on EL7, my operating system of choice when I provision VPS’s. Having your bot be concurrent is a requirement if you plan on using the bot in more than one Discord server; and if you don’t code with concurrency in mind then your users will experience unusual delays and the bot will get slower as it’s used in more servers.

There are ways to achieve this concurrency in earlier versions of python as well, but the async/await primitives in the newer versions are much more convenient to use. The only barrier to entry for me is the library support for async/await as lots of libraries use blocking calls when communicating over the network. Fortunately my bot doesn’t need to make external calls, but even if it did, it’s not hard to write an API client that uses aiohttp or the new requests 3 library.

Checking Answers

At the heart of the bot is the ability to check user submitted answers. Answers are written by humans and may contain spelling mistakes or other grammatical errors that a human may be able to forgive, but a naively programmed machine may not.

Answers are compared with the Levenshtein distance function to work around this issue. This function is able to determine the ’edit distance’ between two strings. The reason why I used this was so the bot would allow for error detection in answers as some names in the Warcraft universe can be hard to spell properly. All of this is implemented by the wonderful library jellyfish, but if you’re curious about how Levenshtein distance actually works then check out the source code or the math in the wikipedia page.

In addition to using Levenshtein distance, I also calculate a ratio so longer answers allow for more forgiveness when it comes to misspellings. Trying to spell many elven names properly when in a rush can be a bit of a challenge.

import jellyfish


def levenshtein_ratio(source, target, ignore_case=True):
    """Calculates the levenshtein ratio between two strings.

    The ratio is computed as follows:
        (len(source) + len(target) - distance) / (len(source) + len(target))

    This function has been ported from (MIT license):

    :param str source:
    :param str target:
    :rtype: float

    if ignore_case:
        distance = jellyfish.levenshtein_distance(
        distance = jellyfish.levenshtein_distance(source, target)

    source_len = len(source)
    target_len = len(target)

    return (source_len + target_len - distance) / (source_len + target_len)

This method isn’t without its flaws. The allowed ratios may need to be tweaked for certain answers and I plan on adding this option in the future. A good example of how this can be exploited is a question I have whose answer is ‘Grond’, but ‘Gron’ will also be accepted as a correct answer even though it’s 100% incorrect. I chose 80% correctness as a reasonable default, but this isn’t reasonable for all answers.

Game State

Game state is stored in memory and in a SQL database. The reason for storing it in SQL is so that games can be resumed when the bot is restarted. If the bot is shutdown for an update mid-game, it can be confusing to users when the bot stops asking them questions and forgets that a game was even playing. The code to resume games is actually fairly simple as all the active games are stored in a dictionary mapping guild ids to the game state.

async def resume_incomplete_games(self):
    """Resumes all inactive games, usually caused by the bot going down."""

    incomplete_games = sql.active_game.get_incomplete_games(self.engine)

        "Found %d incomplete games that need to be resumed",

    for guild_id, active_game_dict in incomplete_games:
        existing_game_state = GameState(
        self.active_games[guild_id] = existing_game_state

        guild = self.get_guild(guild_id)

        if guild:
            channel = guild.get_channel(existing_game_state.channel_id)
                self._ask_question(channel, existing_game_state))

If you have a watchful eye you might notice that there’s no sharding logic in here. This is something I haven’t gotten to yet.


The bot also uses the SQL backend to store server-specific configurations. It’s able to keep track of custom prefixes and allows administrators to limit the bot to a certain channel so that users in the server can decide to mute game messages from the bot.

Problems with the Bot

This bot isn’t without its flaws though. Currently it doesn’t support running with more than one shard, however discord.py makes this easy to implement so I’m not worried about it right now. The hardest part to change if I decide to add sharding would probably be the incomplete game resumption logic, however it shouldn’t be hard to apply the following sharding formula to the SQL query.

(guild_id >> 22) % num_shards == shard_id

If I want to cheat in regards to sharding I can even use auto sharding if the instance is spec’d out and CPU processing power isn’t a limiting factor when reaching the connection limit.


The experience of creating this bot was a lot of fun and taught me quite a bit. I’ve never worked with modern python async/await in a serious capacity until I decided to make this bot as most of the python work I’ve done up to this point has been working on legacy applications written in Python 2 or earlier versions of Python 3. Check out the source code on GitHub if you’re curious about how this bot works! Just keep in mind that if you want to get it working under Windows you may need to do some additional work as I don’t provide instructions for it, but the bot itself should work in Windows.

[1] https://en.wikipedia.org/wiki/Levenshtein_distance
[2] https://pypi.org/project/jellyfish/
[3] https://discordpy.readthedocs.io/en/latest/migrating.html#sharding
[4] https://github.com/amdrel/lorewalker-cho