Permit
Every agent has an identity.
Each agent acts as itself — agent:accounts-receivable — rather than borrowing a shared administrative key. That one decision is what makes everything else on this site possible: you cannot permission, audit, rate-limit or revoke something that has no name.
Who acted?
The same action, recorded two ways.
Sharing a key
finance-admin-api-key
- Indistinguishable from the finance administrator
- Revoking it breaks everything else using it
- The audit log records the key, not the decision
Its own identity
agent:accounts-receivable
Acting as
Itself, version v1.4
Under authority of
Finance Manager
Through
Finance ledger · Mail
Revocable
On its own, alone
The alternative is what goes wrong
An agent using a finance administrator's API key is indistinguishable from the finance administrator. The audit log records the key. Revoking it breaks everything else using the same key, so nobody revokes it.
Recorded on every action, with the version
Which agent, which version of it, acting under whose authority, through which tool, with what result. An action whose actor is a shared credential cannot be explained after the fact.
Revocable on its own
Stopping one agent stops that agent. It does not disturb the eight others reading the same system, because they are not sharing anything.
Agents and people are both first-class actors
Every consequential action records whether a human or an agent did it, and which. As more work is done by agents this stops being a nicety and becomes the only way to read your own history.
Questions this answers
The things you currently have to ask somebody.
Each of these is a question an owner asks regularly and currently answers from memory, a spreadsheet, or by interrupting whoever is nearest.
Which agent sent that email, and under what version of its rules?
If we revoke this agent's access today, what else stops working?
Was this change made by a person or by something acting on our behalf?
Which agents can currently read our customer records?
Where this stops
What it does not do.
An identity is not an authorisation. Having a name does not grant an agent anything — every capability is still a separate grant, and a fresh agent with a perfectly good identity can do nothing at all. Identity also cannot tell you whether an action was a good idea, only who took it and under what rules.
Next
Where this connects.
Give AI a job.
Build an agent, choose what it can access, define what it may do, and deploy it to work safely across your business.
No code required · Full SDKs and API · Stop any agent instantly · [email protected]