Yes. Permagent runs on your machine and calls model providers with your own keys. Turn on sovereign mode and the boundary becomes fail-closed: cloud calls are refused before anything leaves, and only local models run.
A switch in Settings that makes the data boundary fail closed. With it on, every cloud inference call is refused before any data leaves the machine, and only local models are allowed to run: the built-in local provider, or Ollama on localhost.
What that gets you. Privacy you can verify rather than trust. When a request is refused you are told why, and offered the local route instead.
Settings shows what each session and project has cost in tokens and dollars, lets you set ceilings that pause work for your approval, and keeps an append-only log of everything that has left the machine.
What that gets you. Straight answers to what is this costing me and what actually left, on your own terms rather than a vendor's dashboard.
Before any automated work is committed, a scanner inspects every file for credential-shaped content: env files, private keys, connection strings with passwords, API keys and tokens. If it finds one, the work stops.
What that gets you. A secret cannot reach your git history because the agent thought it looked fine. The check is code, not judgement.
Pair a phone or laptop to the Mac that holds your Brain. Devices connect over your own private network, are listed by name with a last-seen time, and any of them can be revoked.
What that gets you. Permagent on your phone without your memory moving anywhere. The hub stays the only copy.
Everything on this page describes behaviour that ships today. The facts come from the runtime’s own capability brief, the same one the agent reads about itself on every turn, so nothing here is a roadmap item wearing a present tense.
How does an AI agent remember across sessions?
Can an AI agent work while you sleep?
What can an AI agent actually do on your computer?
How do you stay in control of an autonomous agent?
What stops an AI agent from going off the rails?
Which AI models can you use, and who pays for them?