Docs

AI

The whole AI section: assistants, prompts, training, moderation, providers, usage and logs.

How the AI section is laid out

AI is not one module. It is ten pages in the Modules menu under AI, and they share one configuration per server.

  • Overview: the state of the AI in this server in one place.
  • Assistants: which AI modules run. Assistant, support helper, moderator, autoresponder, knowledge base and translator. Auto Responder has no module of its own any more; it lives here.
  • Prompts: the prompt library and templates.
  • Automations: AI-driven rules you add, switch on and off, and delete.
  • Training: what the AI knows. Add text yourself, or point it at a URL and let it read a site.
  • Moderation: the AI moderation rules, each one switchable on its own.
  • Providers: models, routing and the default model.
  • Usage: what has been spent, in requests and tokens.
  • Logs: the interaction history.
  • Settings: rate limits, cost alerts and feature flags.

How the assistant behaves

Personality is friendly, professional, casual, technical, or your own. A custom system prompt can be up to 65,535 characters.

Trigger mode decides when it answers: only when mentioned, only through a command, or on every message. You can set this per channel as well as for the whole server.

The context window is 0 to 50 earlier messages. A reply is capped at 256 to 8192 tokens.

Generation settings: temperature 0 to 2, top-p 0 to 1, frequency and presence penalty each -2 to 2.

You choose which channels the AI is active in, and separately which channels it may take moderation action in, plus the roles it must never act on.

Limits and alerts

Rate limits run per member: 1 to 1000 requests an hour (default 30) and 1 to 10,000 a day (default 200). Roles can be exempted, and you can write the message someone gets when they hit the limit, up to 500 characters.

Cost alerts fire at percentages of your budget. The defaults are 80, 95 and 100. Each alert can go to a channel, to the server owner by DM, or both.

Training

Add knowledge as text with a title and category, or as a URL. A site import walks up to 100 pages and defaults to 50.

Memory has its own settings: whether it is on, whether it captures from Discord, whether it is per member, and a privacy mode that is either normal or strict. Staff review and memory manager roles decide who can see and remove what was remembered.

Provider and key

The API key and the provider are set centrally by Modora staff, not per server. A server picks a model and how it behaves, not who supplies it.

What is stored

Warning

AI chat stores every message and the reply to it, and there is no scheduled deletion for that data. This is worth knowing before you switch the AI on in a busy channel.

Members can act on this themselves: /ai opt-out stops the AI learning from their messages, /ai forget requests deletion of their AI conversation data, and /ai memories shows what has been remembered about them.

Commands

Everything sits under /ai. The admin subcommands are marked.

/ai ask

Ask a question. Optionally as a private reply.

/ai tokens

The server's token balance.

/ai mode

Your own preferred mode, standard or coding.

/ai remember

Save an instruction or preference for the AI to keep.

/ai memories

See what has been saved about you.

/ai forget

Request deletion of your AI conversation data.

/ai opt-out

Stop the AI learning from your messages. /ai opt-in undoes it.

/ai cross-guild

Decide whether your preferences follow you to other servers.

/ai summary

Summarise your recent conversation. /ai tldr summarises recent channel messages instead.

/ai setup

Set the AI up in a channel. Admin only.

/ai train

Add knowledge by hand. /ai learn-url reads a website instead. Admin only.

/ai config

Model, temperature and maximum tokens from Discord. Admin only.

/ai stats

Usage statistics. Admin only. /ai disable switches the chatbot off for the server.

Was this page helpful?