This page explains the mechanisms behind the system: the processes running in the background, and how they improve your work.
A plugin for the Claude desktop app. One file, installed once, into the Claude subscription you already have.
A skill file turns a chat into an agent
It tells the AI who it is.
It points it to Notion, where all the information is stored.
The skills built over time to help it do its job better.
What it may do without you, and when it wakes up to do it.
These do not run on a clock. They run whenever an agent needs them, and most of them exist so the system is better next month than it was this month.
In your own accounts, and nowhere else. Numerus adds no server and no database of its own. Nothing is copied to us and we cannot see it. Remove the plugin and everything stays where it is.
Anything that leaves the building waits for your approval. Your own rules, written in your words at install, are read by every agent at every boot. Agents cannot read each other's chats, only the log. What an agent can reach is what you connected, and nothing more. The nightly audit writes down anything that failed.
AI models change fast and can act in ways nobody predicted. Numerus adds structure and checkpoints; it does not make the model predictable. Treat it as you would a new hire with access to your accounts: do your own review of the tools you connect, and read what you approve. Numerus AI takes no responsibility for what a model does inside your accounts.
I use Google Analytics cookies to see how people find and use this site. Is that OK?