Documentation
How a local, private agent actually works.
How Permagent works: permanent memory on your own disk, a fail-closed data boundary, background automation, and thirty-plus model providers with your own keys. Open source, local-first, macOS.
Start here
What is Permagent?
Permagent is an open-source AI agent that runs on your Mac. Everything it learns is stored in a knowledge graph on your own disk rather than a vendor's cloud, so its usefulness compounds instead of resetting each session. It is free, you bring your own model API keys, and it keeps working whether or not the company behind it does.
How is it different from ChatGPT or Claude?
Those run in someone else's cloud and start most conversations without durable memory of your work. Permagent runs locally, remembers permanently on your disk, and can act on your machine through a real terminal and browser. You connect directly to model providers with your own keys, so nobody meters or marks up your usage.
Is my data private?
Your memory, documents and keys stay on your machine. Model calls go directly from your computer to the provider you chose. In sovereign mode the data boundary is fail-closed: anything not explicitly allowed does not leave, and an in-app egress view records what did.
What does 'it compounds' mean?
Every conversation, file and decision is written to permanent local memory with typed entities and relationships. A background local model consolidates recurring facts during idle time. The agent you use in December knows what it learned in March, so the value accrues to you rather than being discarded at the end of a session.
By topic
Permanent memory
How does an AI agent remember across sessions?
6 capabilitiesPrivacy and the data boundary
Can an AI agent run without sending your data to the cloud?
4 capabilitiesWork that happens without you
Can an AI agent work while you sleep?
6 capabilitiesThe workbench
What can an AI agent actually do on your computer?
6 capabilitiesStaying in control
How do you stay in control of an autonomous agent?
6 capabilitiesGuardrails
What stops an AI agent from going off the rails?
4 capabilitiesModels, keys and cost
Which AI models can you use, and who pays for them?
2 capabilities