Guides
How to keep business data on your computer (November 2026)
For owners who handle customer data and want AI help without sending it away. After reading, you can choose where each task runs and check what leaves the machine.
In short
- Decide per project whether text may go to a cloud model or must stay on your computer.
- A local model keeps the work on your machine; a cloud model receives what you send it.
- Do not trust a promise alone; look at the network activity yourself.
What it is
Keeping data local means that the files, messages and records you work with are processed on your own computer instead of on a vendor's server. For AI, it mostly comes down to where the model runs and what text is sent to it.
How to do it
- Classify your data. Public, internal and personal is a workable split. Personal data about customers is the class to protect first.
- Match the class to a route. Personal data goes to a local model. Internal data may go to a cloud model you pay for under terms you have read. Public data can go anywhere.
- Run a local model for the sensitive routes. It needs enough memory; check the model size against your computer before you commit.
- Mask before sending to the cloud. Names, emails and numbers can be replaced with placeholders before text leaves, and restored afterwards.
- Store keys in the system keychain. Never paste an API key into a document or chat.
- Verify with a network log. List the outbound connections of the tool you use. Empty means nothing left; unexpected hosts mean you should ask why.
How Growth OX helps
With a local model in Growth OX, your work stays on the computer and there is no Growth OX server in between. With your own provider, the text goes straight to that provider, never to us. There is no account and no tracking. The local-first approach is described in the privacy documentation and in what leaves your computer.
A project can be set to Only on this computer, and its text is then never sent to a cloud model. Personal details are masked before text goes to a cloud model; this is on unless you turn it off. Three privacy modes exist: Only this computer, Local AI + connected tools, and Cloud-assisted. A network log lists every outbound connection of the app's own processes, and in Only this computer it stays empty. An API key is stored only in the Mac's Keychain.
Limits: local models run through Ollama, which must be installed and running on the same Mac. A local model may be less capable than a large cloud model. Update downloads are the app's own network calls.
Other ways to do it
| Option | Good for | Limit |
|---|---|---|
| Cloud assistant with a business agreement | Strongest models | Data goes to the vendor under their terms |
| Manual redaction before pasting | Occasional use | Slow; easy to miss a detail |
| Self-hosted model on a server | Teams with an administrator | Needs hardware and upkeep |
| Local model on your own computer | Personal data, small teams | Smaller models; needs memory |
| Mixed routing per project (Growth OX) | Different data classes | You still need to classify the data |
Where to go next
Pick one project with personal data, set it to local only, and open the network log after a day of work. If you also need to connect outside tools, read how to connect your tools to an AI assistant.
FAQ
Is a local model as good as a cloud model?
Not always. Smaller models handle drafting and sorting well but may struggle with long, complex reasoning. Test on your own tasks.
Does masking guarantee privacy?
No. It reduces what a cloud model sees, but free text can still reveal things. Use local only for the most sensitive work.
How do I know nothing is leaving the machine?
Check the network log, and compare it with the documentation. Verification beats assurance.