Features AI Chat

Included everywhere

AI Chat

Answers in chat when someone tags it, using a model we host ourselves.

The bot writes in chat like a regular member: short, casual, English. By default it replies only when someone mentions it, and you can limit it to specific channels. Replies come from a small language model running on our own hardware next to the bot, not from a third-party API. It also comments on images people post and keeps a few facts users state about themselves.

What it does

  • Replies only when mentioned, or only in the channels you list, and never to @everyone, @here or other bots
  • Filters the incoming message for prompt injection and screens its own output, so it does not post links, ping roles, or discuss being a bot
  • Applies per-server rate limits you can change: 3 seconds per user, 10 replies per minute per channel, 5 replies at a time
  • Stores up to 15 self-stated facts per user (preferred name, game, country, rank, age) and drops them after 90 days
  • Reads the last couple of messages, or the message being replied to, when a question refers to something said earlier

Worth knowing before you turn it on

The base personality is written into the bot and mentions the AceV community; the 'persona' setting is added on top of it rather than replacing it, so on your server the bot can still refer to AceV. It answers in English even when written to in another language. The model is a 1B model, so replies are one or two sentences and it gets things wrong. If the model host is down the bot stays silent instead of showing an error. Per-user memory is keyed on the Discord user ID only, so a fact learned on one server is visible on every server the bot is in. The per-plan reply quotas listed in the plan matrix (5/20/60 per minute) are not connected to this feature yet: it uses its own limit, 10 by default.

Add the bot See the plans