The AI Charter Explained

A charter describes a job: why it exists, what it involves, how to go about it and what never to do. People have charters too, and their job descriptions and postings are built from them, but for an AI member the charter is what it is actually told. Kamo compiles an AI member's charter into its operating brief, the standing instructions it works from.

Layers: who writes what

A charter is not written in one place. Each AI member's charter is put together from four layers, in this order:

1. Role: the security roles it holds. Settings → Essential Setup → Security Roles.
2. Department: its department, and every department above it, from the top down. Settings → Essential Setup → Departments.
3. Job title: Settings → Essential Setup → Job Titles.
4. Member: items for this one AI member. Open its settings page, then the Permissions tab, then the Charter card.

So you describe what every receptionist does once, on the job title, and each receptionist you hire, person or AI, inherits it. You add to it per member only what is personal.

Editing a role, department or job title charter needs Configure System Settings. Editing one member's charter needs Manage Member Security, and for an AI member Manage AI Workforce as well. Nobody can change their own charter.

Sections

The editor groups a charter's sections into four bands:

• The role: Purpose (why the role exists), Responsibilities (what it does: all the time, when something happens such as a new email or a new lead, or on a schedule such as every weekday at 9), and Goals (what success looks like, with a metric and a target where there is one).
• Conduct: Guidelines (how to handle things), Boundaries (things never to do), Escalation (when to hand a matter to someone else, and to whom: the supervisor, the department manager, everyone in a role, or a specific team member) and Knowledge (facts to know, written in place or as a pointer to a knowledge base article or category, a sales script market or a note).
• Requirements: Qualifications (what someone needs to hold the role; shown on job postings and never given to an AI as instructions) and Working hours.
• AI members: Approvals, Limits, Brain, Voice and Workstation. Only AI members use these; people who hold the role ignore them.

The AI members band

• Approvals say, per kind of action, whether the AI member needs a person first: No approval needed, Tell the supervisor afterwards, Ask before doing it or Never allowed. For money (committing to a price, moving money) you can require approval Only above an amount in US dollars.
• Limits cap volume, spend and autonomy: emails, text messages and calls per day, calls at the same time, new people contacted per day, how often and at what hours it may reach out to the same contact, its daily and monthly AI budget, how many chat turns two AIs may take before offering a person, how long one work session may run, and its highest autonomy (0 shadow, 1 supervised, 2 independent).
• Brain says which kind of model handles each kind of work: Light (fast, low cost), Medium or Heavy (most capable), and whether health data must stay with the AI services Kamo runs itself (Health data stays in-house).
• Voice says how the AI member sounds in writing and on calls. The voice it speaks with is not chosen here but on its profile, under Identity → Voice; see AI Team Members Explained.
• Workstation lists apps, browser extensions and bookmarks to put on its computer.

How the layers combine

Each section follows one rule, shown in the editor:

• Every level adds its own: Purpose. Each layer's purpose is kept, in layer order.
• Adds to what is inherited: Responsibilities, Guidelines, Boundaries, Escalation, Knowledge, Qualifications and Workstation. A lower layer adds items.
• The most specific level wins: Working hours, Brain and Voice. The member's own setting beats the job title's, which beats the department's.
• The strictest value wins: Approvals and Limits. If the role says "Ask before doing it" and the job title says "No approval needed", the AI member asks. The lowest limit applies.
• Per metric, the most specific level wins: Goals.

Inherited items, locks and hiding

Each layer's editor shows what it inherits, under Inherited, with where each item comes from ("From Department: Sales"), and what it sets itself, under Set here.

• Hide here removes an inherited item for this layer and the layers below it, for example a guideline that does not fit one team. Roles cannot hide inherited items.
• Lock this item stops the layers below from hiding or overriding it. Boundaries are always locked. A member's own items cannot be locked.

An item that does not apply says why: Hidden by a layer, Overridden by a more specific level, A stricter value applies, or Repeats an earlier item.

Preview

Click Preview in any charter editor to see the result two ways:

• As job description: what a person reads: purpose, responsibilities, goals, required and preferred qualifications, working hours, how we work, and when to escalate.
• As operating brief: exactly what an AI member in this job is told.

The operating brief

An AI member's operating brief is compiled from its full charter, with your organization's Required opening text and Required closing text wrapped around it (Settings → Artificial Intelligence → Security Models, General AI Settings). It is compiled when the AI member is hired and again whenever its charter changes, and each version is numbered: its workbench shows Operating Brief version and the member's Charter card has View brief. Every run records which version it worked from.

Where to start

Write the charter on the job title, not the member, so the next hire inherits it. Job Templates and Personas gives you a ready-made job title with its charter filled in.