Plain-English definitions.

An AI worker with a set of tools and instructions that can take actions on your behalf.

A group of agents where a coordinator hands work to specialists and pulls the answers together — like a real team.

A capability an agent can use — query a database, call an API, send a message, take an action.

A permanent, tamper-proof log of every consequential action: what the agent saw, what it did, and what it cost.

An automatic safety filter that runs on every input and output — catching prompt attacks, personal data, secrets and more.

Replaying a candidate policy change against your real history to see what it would have caught — before you turn it on.

A check that blocks a policy change if it would stop catching problems your current policy catches.

An agent acting for a specific person. It's allowed only while that permission is switched on — revoke it and the agent stops.

Requiring two different people to approve a risky action; the person who started it can never approve their own.

A fleet-wide limit on how much automated agents can spend or do — plus a one-click emergency stop (kill-switch).

How much an agent is trusted to do without a human. It rises as the agent proves reliable and tightens automatically if quality drops.

Bring Your Own Key — use your own AI-model account. Your keys stay in your infrastructure, encrypted, never shared.

Running fully disconnected from the internet, for teams that need maximum security or data sovereignty.

Role-based and relationship-based access control — deciding who can see and use each agent, team and data source.

Defining your agents in a text file (ceez.yaml) and shipping changes like software, with a preview of every change.

Something that starts an agent automatically — a schedule (cron), an incoming webhook, or a platform event.