How Modu works

Modu is an agent, not a chatbot. You describe what your Discord server should look like in plain language. It inspects the current server, plans the concrete operations needed, asks for approval before anything destructive, and then executes.

Every run follows the same loop: inspect → plan → approve → execute → undo. Reads and additive changes run immediately; renames and permission changes follow your settings; destructive operations always require explicit approval that names each operation.

Sign in with Discord and add the bot to a server you manage. The bot asks for Administrator, so moderation, voice and other features work without re-authorising. Then drag its role to the top of the role list in Server Settings.

Once connected, open a chat for that server and describe the change. For example, "create a #releases channel visible only to the Maintainers role" or "audit the server for risky permissions".

The bot is added with Administrator. If you untick it on Discord's screen, it still receives the specific permissions it needs (such as Manage Channels, Manage Roles, Send Messages and Manage Messages), and features that need more will ask.

Administrator does not override Discord's role hierarchy: the bot can only manage roles and members below its own role, so place its role at the top. Its tools refuse to create or grant Administrator roles.

Destructive operations are classified DANGEROUS and cannot execute without approval that names each operation. Everything returned from Discord (channel names, messages, topics) is treated as untrusted data, never as instructions, so a prompt hidden in a channel name cannot redirect the agent.

Reversible operations record their previous state, so created channels and roles, renames, permission overwrites and posted messages can be undone. Deletions are permanent in Discord, and the product says so before you approve them.

Each plan has a monthly allowance of AI credits. A request to the agent uses credits for the AI work it takes (about 18 for a typical request); the bot's own features never use credits. The agent stops when the credits run out and they reset each billing period.

Each plan maps to a reasoning-model tier and a monthly action allowance. See the pricing page for the current plans.