Saved recipes run on a schedule, repeated work turns itself into reusable skills, and quiet background workers surface the one thing worth your attention. All of it on your machine.
Saved tasks run in the background on a cron, interval or one-off schedule. Transient failures retry, runs missed while your machine was asleep are recovered, and anything that keeps failing is escalated to you rather than retried forever.
What that gets you. Overnight work that is actually there in the morning, including after a machine that was closed at midnight.
Do something twice and Permagent offers to remember it. Accept, and it becomes a skill: a plain folder you can search, open, edit or delete, tracked with how often you use it and how far it has been proven.
What that gets you. The work you repeat stops being work you repeat, and the recipe is a file you own rather than a setting in someone's cloud.
It notices when you run the same terminal command over and over. Once it crosses a threshold and you have been quiet a few minutes, it drafts an automation and puts the proposal in front of you.
What that gets you. Suggestions that come from what you actually do, and never run until you approve them.
A background worker that surfaces at most one thing a day: a thread in your memory that has gone quiet while you moved on, or genuine news about a project you are working on right now.
What that gets you. One useful nudge, not a notification stream. It is capped at daily on purpose.
A live roster of every background worker, scheduled job and active session, each showing what it is doing, when it last acted, and a one-click stop.
What that gets you. Autonomous work you can see and interrupt, instead of guessing what your machine is up to.
A day-by-day record of what the agents actually did: goals moved, decisions raised and resolved, memories described, nudges sent, tasks failed. Kept for 90 days and filterable.
What that gets you. An answer to what did it do while I was away, with the evidence attached to each entry.
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 run without sending your data to the cloud?
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?