Instructions a person can read
Each agent has a plain description for your team and instructions for the model. Draft one with help, but nothing is created until a person saves it.
An agent is a set of instructions, a set of permissions and a memory. It reads what arrived, decides the next step, and proposes actions. What it may actually do is decided by you, tool by tool.

Each agent has a plain description for your team and instructions for the model. Draft one with help, but nothing is created until a person saves it.
Every tool is set to run automatically, to wait for approval, or to be denied. A tool with no setting is denied.
A memory is a note, never an instruction. Memories are visible, deletable and capped, and they cannot grant a permission.
Thirteen templates, eight of them written for energy supply. Each refuses by design to promise savings, take bank details or invent an offer.

Answers the questions energy customers actually ask: bills, readings, tariffs, payments, moving home.

Explains a bill line by line from the figures your system holds.

Takes, checks and confirms meter readings.

Follows up missed payments kindly, with no threats and no offers.

Responds when a customer says they are leaving.

Gathers what a new contract or change of holder needs.

Compares the tariffs you have written down. Never promises savings.

Summarises what ran, what failed and what it cost.

Answers customers from your Company Brain.

Decides what needs a person today.

Keeps contact records current.

A short daily summary for the team.

Pulls together what you need before a meeting.
The agent reads the message and works out what is being asked: a move, the dates, and what is missing.
Example run. Each run ends as completed, awaiting approval, denied, or failed, and says which.
Specialists with instructions, permissions and a memory. They read, decide and propose the next step.
It stands on the 2 layers beneath it.
Start from a template, set its permissions, and watch every run.