Introduction

Perstack runs business processes with AI agents, under rules the agents cannot break.

You describe a process in plain language. Perstack turns that into a definition — a readable document naming the records the work keeps, the operations that can be performed on them, and the rules that govern each one. Deploy the definition and you have a running system: agents that carry out the work, screens for the people involved, and a record of everything that happened.

What makes it different

An agent given real work can do real damage. The usual answers are to keep agents away from anything consequential, or to wrap them in hand-built guardrails that someone has to maintain.

Perstack takes the rules out of the agent's reach. They are part of the definition, enforced by the platform before an operation executes, not instructions the agent is asked to follow. An agent cannot pay above a limit, post the same invoice twice, or call an operation nobody defined, because those are not choices available to it.

That is what makes it reasonable to hand over work that matters. Every run leaves a record showing what was requested, what was checked, what committed, and why — so the delegation stays inspectable after the fact.

What you work with

A Project is the boundary for a workflow. It contains Environments, Agents, Definitions, Releases, and Deployments. An Agent is a versioned coordinator or delegate that carries out operations. A run is one unit of work an Agent performed, and the record is what it left behind.

Some operations complete on their own. Some wait for a person to approve them. Some are refused outright. Which is which is stated in the definition, and the platform enforces it.

Where to go next

To build something, start with Your first Project. To understand the model before building, read Definitions.

To integrate Perstack with systems you already run, see the API reference.