Controlling AI agents

AI agents (the AvairAI CLI, Claude, ChatGPT, Cursor and other apps connected through MCP) act as one person in your company, with that person's role, plan limits and campaigns. As an Admin you decide whether agents may work at all, whether they may take live actions and how many Tools credits each one may spend. You can also see and stop everything they do. All of it is in Account → AI Agents.

The two switches

Switch	When it is off
Allow AI agents	No one in the company can use an agent. Every agent request is refused, and connected apps cannot list their tools.
Allow launch from agents	Agents cannot launch or resume campaigns, stop active or paused ones, send drafted replies, build Playbook campaigns, turn on Playbook Automation or start paid Tools runs. They can still create, edit, build and test campaigns.

When the launch switch is on, each of those actions still needs a confirmation from the person using the agent.

The launch switch does not cover edits to a campaign that is already running. An agent may change a running campaign's settings and content the same way the person could in the app, for example turning on auto-replies or AI calls. Give agent access to people you trust with those changes.

Tokens and connected apps

• Personal access tokens belong to the person who made them. A token is shown once. Its owner can revoke it; it stops at its next request.
• Connected apps and CLI sessions are the apps people signed in to with their AvairAI login. One entry covers every computer where the person uses the CLI, so revoking it ends it everywhere.
• All tokens and apps (Admins only) lists every token and app in the company. You can revoke any of them, and restore one you revoked. After a restore the owner signs in again.

Daily credit cap

Paid Tools spend the company's monthly Tools credits. You can give each token or app a daily cap in All tokens and apps; a run that would go over it is refused until the next UTC day. Only an Admin sets or changes a cap.

Agent activity

Agent activity lists what agents did: the action, the campaign, the app or token and whether AvairAI allowed or refused it. An Admin can switch between the whole company and their own requests; everyone else sees the requests made by their own tokens and apps.

What agents can and cannot do

Agents can read and build campaigns, review and edit content, find and verify contacts, work the task queue, run Tools after a quote, use Company Knowledge and Playbooks and assess a step before launch. Assessing is free; applying a suggested fix edits the campaign the same way a person would. Content fixes apply while a campaign runs, as content edits do; the other steps take fixes only before launch.

Agents cannot buy credits or change a subscription, manage people, roles or API keys, reach staff administration, skip launch readiness or act on LinkedIn for anyone. Every agent request stays within the role of the person it acts for.

Limits

Limit	Value
Requests per token or app	120 a minute and 20,000 a day
Requests that generate content, targets or contacts	30 an hour and 150 a day per person
Tools credits	The company's monthly credits, plus each token's daily cap