hikari SDK

pip install mochi-analytics mochi-analytics-hikari

Requires Python 3.10+.

Quick start

import os
import hikari
from mochi_analytics import MochiClient
from mochi_analytics_hikari import attach_mochi
 
mochi = MochiClient(
    url="https://mochi.example.com",       # your Mochi instance
    api_key=os.environ["MOCHI_API_KEY"],   # from the bot's settings page
)
 
bot = hikari.GatewayBot(token=os.environ["DISCORD_TOKEN"])
attach_mochi(bot, mochi)
 
bot.run()

That’s it. Mochi now records slash/context-menu command usage, guild joins and leaves, and an hourly server-count snapshot.

attach_mochi subscribes to InteractionCreateEvent, GuildJoinEvent, GuildLeaveEvent and StartedEvent, and returns a detach callable that unsubscribes all of them.

Sharding

hikari does not attach a shard to an interaction, so a command event’s shard id is derived from its guild id the same way Discord does — (guild_id >> 22) % shard_count.

Options

attach_mochi(
    bot,
    mochi,
    include_guild_names=True,   # put guild names in join/leave metadata
    ignore_commands=["ping"],   # skip noisy commands
    snapshot_interval=30 * 60,  # seconds; default 1 hour
    auto_track_commands=False,  # see "accurate timings" below
)

Accurate duration & success

hikari ships no command framework, so attach_mochi records commands straight off InteractionCreateEvent — which means it can’t see whether your handler succeeded or how long it took. For that, disable auto-tracking and wrap your handlers:

from mochi_analytics_hikari import wrap_command
 
attach_mochi(bot, mochi, auto_track_commands=False)
 
@wrap_command(mochi)
async def play(interaction: hikari.CommandInteraction):
    ...  # duration and raised exceptions are recorded

If you use a command framework on top of hikari (lightbulb, tanjun, …), call wrap_command on the callbacks it dispatches.

Custom events

from mochi_analytics import MochiEvent
 
mochi.track(MochiEvent(
    type="custom",
    name="premium_purchased",
    user_id=str(interaction.user.id),
    guild_id=str(interaction.guild_id) if interaction.guild_id else None,
    meta={"tier": "gold"},
))

Design guarantees

• Events are batched (flushed every 5 s or 100 events) and sent in the background — track() never blocks or raises.
• Transient failures retry with backoff; the queue is bounded (oldest dropped first), so a dead Mochi instance can never leak memory or crash the bot.
• Raw user ids are hashed server-side with a per-bot salt and never stored.