The operator's manual.

looprig is an agent harness for building agents you own and can change. You choose the models, instructions, tools, permissions, storage, interface, and where the system runs.

The core mental model is small:

• A Loop defines one agent.
• A Rig assembles one or more Loops with storage and lifecycle policy.
• A Session is one running instance of that Rig.

The shared glossary defines these and the other terms used across the project.

Choose your path

Ownership is part of the design

looprig does not give you one fixed agent. It gives you composable parts. Your application remains the composition root, so you decide:

• which model providers to trust;
• what each agent knows and can do;
• which actions happen automatically and which require approval;
• where session history and workspace snapshots are stored;
• whether the system runs on a laptop, server, or distributed backend;
• whether people use it through a terminal, an HTTP API, or an interface you build yourself.

Start with one useful task. Add machinery only when the task needs it.