Who may act?
A prompt isn’t a permission. Each agent gets a contract that says exactly what it may do.
Saturn is designed to give each AI agent its own identity, its own contract and a record of what it asked to do and what Saturn decided.
Private preview. See what’s working today →
Write down what each agent is allowed to do. Saturn is designed to check each request it handles against that agent’s contract and record the decision. In our local reference tests it does. Review the record, share it with your team, adjust the contract, and repeat.
Saturn doesn’t replace your tools or your judgment. It makes the boundary visible, so the decisions stay with you.
Name the agent, its owner and the job it’s there to do.
Grant only what that job needs. Terminalia shows this in a local reference test.
See what was allowed, what was refused and what Saturn couldn’t observe.
Expand, shrink or revoke access as the job changes. When you replace a contract, the old one stops working. Our reference test checks this.
A prompt isn’t a permission. Each agent gets a contract that says exactly what it may do.
What an agent asked for, what a tool said happened, and what was independently confirmed are kept separate, so you can tell them apart.
Saturn is being designed so that authority and records stay tied to the agent, so a person you approve can pick up the work later. This part is still in development.
| Component | Purpose | Stage |
|---|---|---|
| Terminalia | Delegated authority for your agents | Private preview · Request preview access by email ↗ |
| Saturnalia | What your agents asked to do, over time | In development |
Pick one agent, one job, and one thing it should never be allowed to do. We’ll start there.