# What the agents do

Place AI Ops in five minutes: what the agents take on, and what stays with your team.

## In one sentence

Naoshift AI Ops hands the operations that repeat to specialist agents: hypervisor version upgrades, on-call, deployment and reports. Your team keeps the decision.

## Three rules that never change

1. **The agent proposes, a person decides.** By default, every write becomes a proposal that a person who holds the right themselves approves, with a second pair of eyes.
2. **The action follows a frozen runbook.** The operations engine runs it, not the model. Model down or budget spent: a campaign stops before the next node, never in the middle.
3. **Never more rights than you.** An agent works for a person: its scope is cut by that person's rights, and every action goes through the same locks as in the inventory.

## Your company's agents

When AI Ops opens, your company receives specialist agents: **Updates**, **Incidents**, **Reporting**, **Capacity**, **Security**, **Backup and resilience**, **VMware migration** and **Analyst**. They sit at the top of the console, next to the Assistant: their initials, and a dot that tells their state.

## SaaS or on-premise

In SaaS, the agents work read-only: they answer, investigate, write the reports and watch the patches. Write actions, update campaigns included, happen in an on-premise deployment.

## What the model sees

The AI model only receives pseudonymised identifiers: never the real names of your systems, nor your secrets. Real names are applied on display, in your browser.


