Local. Default.
Yours.
MingLLM is a personal-intelligence runtime. It runs where you are, with the things you already use. These four sentences are the whole design brief — every product decision is downstream of them.
Models, memory, and reasoning run on your machine.
The base model is on your disk. Your memory is in your sandbox. The reasoning happens in your RAM. There is no cloud round-trip in the path between your intent and the answer — because there is no cloud in the path. The constraint is the product. We pick a base model small enough to live with you, and we tune it until it earns its seat.
Nothing calls out without a line in the log.
Every outgoing request is named, timestamped, and reviewable before it fires. Every tool MingLLM uses on your behalf is listed, with the inputs, the response, and the result. The receipts log is not compliance theatre — it's what makes the agent trustworthy enough to grant real authority to. If we can't show you what happened, we shouldn't have done it.
Apps get only the permissions you grant.
MingLLM does not ask for blanket access. Each capability — read mail, write a file, send a message, run a shortcut — is a discrete grant you can revoke from one screen. The default state is locked down. The opening posture is suspicious of itself. We would rather you grant a permission once, knowingly, than have us assume it on your behalf and learn later that you didn't want us to.
Every action leaves a line. Reviewable. Reversible. Yours.
If you cannot see it, it didn't happen here. If you cannot undo it, we shouldn't have done it autonomously. The whole product is shaped around the idea that an agent worth trusting is an agent that lets you trust it — by surfacing what it does, by undoing what it can, by leaving the record with you and not on a server somewhere.