/ cloud API decisions
Local AI: when it's required, when it's just wiser
This page is for deciding where your AI work can run. The first list is what the rules stop; the second is what good judgement keeps off a cloud API. A machine you own is how both stop being a problem.
private
Local AI for your own affairs
For the household and the individual. A diagnosis, a divorce, a family dispute, tax and investment papers, a will — the documents with no company behind them and no compliance department between them and a provider.
- Privacy. The file never leaves the room. No copy at a provider to be subpoenaed, breached or used to train someone else's model.
- Autonomy. No account to be closed, no terms to be changed, no service to be discontinued under you. It works when the internet does not.
- Capability. The whole model, not a rate-limited version of it — running on the documents you would never upload.
- Being early, deliberately. Few households own this yet. That is part of what is being bought.
business
Local AI for the company
For the business that has data it is responsible for, work it would rather not hand over, and a bill it has to justify. Three reasons, in the order they usually arrive.
- Regulation. Data you cannot lawfully send to a third country, professional secrecy, client confidentiality, contracts that forbid it.
- Your own IP. The drawings, the code, the pricing model, the method. Nothing that defines how you compete should be a training example.
- Cost. Per-token pricing scales with use; a machine is bought once. Above a certain volume the arithmetic turns, and it stays turned.