Private AI for a small business, on a Mac in the office.
Some records should not be sent to an outside model provider. A Mac in your office running open-weight models keeps the work and the documents on hardware…
The usual first question about AI in an accounting practice, a law office, or a business mid-acquisition is not about capability. It is where the documents go. With most cloud tools, the answer involves an outside provider, a retention policy, and a contract someone has to read closely. Private AI changes the answer. The model runs on a computer in your office. The documents it reads stay on that computer. Apple Silicon Macs have enough memory and processing to run open-weight language models locally. On that machine sit the model, the agent's working environment, and its log. The agent reads documents from your systems, prepares the work, and puts it in front of a person for approval, the same as a cloud deployment. What leaves the office is what the job requires: the approved email to the client or the entry in your accounting system. Inference, the step where the model reads your documents, happens on hardware you own. Local models are capable, and they are not the largest models available. Jobs that rely on structured documents and written rules suit them well. Jobs that need broad open-ended reasoning may be better in the cloud, with the data redacted first. The hardware is also a line item, and the machine needs an owner for power, updates, and backups. I set that up and document it, and the appliance is priced openly: hardware at cost plus $5,000. Name the records that cannot leave. Pick one recurring job that touches them, with a person who owns it and a cost you can count. The free AI Value Scan maps that job and records whether it should run in the cloud or on a private Mac. From there, the Pilot puts one agent into production on your real data within 30 days, with a full refund if it is not running.