octroi.ai
Open the app

Nothing passes without a green light.

And every green light leaves a trace.

To grant, in French, is octroyer. In Paris, the octroi was also the ring of stone tollhouses built by Ledoux, where nothing entered the city without being declared, checked and authorised. octroi.ai does the same for the agents and the people who touch your infrastructure.

0
infrastructure credentials held by an agent
1
lifecycle for human and agent work
4
autonomy levels, each set by a human
100%
of transitions recorded as Events
A Job approved in a messaging Channel.
The catalog of Actions of a Platform.

The symbol

Our symbol is the diagram of a changeover switch. A request comes in on the left; a decision closes the arm on one output. The chosen contact lights up, the other stays open.

That is the product: a request arrives, a decision routes it, and the contact closes for one precise Job, then opens again.

We built octroi.ai on one conviction: safety belongs in the domain model and the network topology, never in a prompt. An agent is a Principal with its own identity, a scope, a closed catalog and an autonomy level that a human sets.

For the infrastructure engineer, it is an on-call colleague that triages and proposes. For the lead, delegation becomes a versioned object. For security and audit, the question of what the AIs did in production, and who authorised it, becomes a query.