Private AI versus ChatGPT: where your data actually is.
The trouble with public AI
The worry many managers have is a reasonable one: uploading finances, contracts or client records to a public AI means sending them outside the company. For a Spanish small business that means reviewing international data transfers, contractual clauses, and what the provider does with whatever you send.
What a private AI is
A dedicated server — in your own office — with a graphics card capable of running language models locally, using tools like Ollama. It reads and processes invoices, PDFs and contracts at full speed, and not one byte goes to OpenAI or any third party.
The comparison, in cold blood
| Public AI (ChatGPT and similar) | Private AI (on your server) | |
|---|---|---|
| Where the data is processed | The provider's servers, usually in the US | In your office |
| GDPR | Requires international transfer agreements | Your existing processing agreement still stands |
| Cost | Monthly fee per user or per use | Investment in your own hardware, no per-seat fees |
| Control | The provider's | Complete: the system is yours |
Who it makes sense for
Any small business handling sensitive client paperwork: accountancy firms, law practices, logistics, wholesalers with supplier data. The private AI page explains what the installation covers, and the free audit assesses whether your document volume justifies it.
Quick questions
Yes, done properly. With a private AI running on your own server, the data never leaves your infrastructure, so the standard processing agreement you already have with your clients remains fully valid.
For business tasks — reading, classifying, extracting data and answering questions about your own paperwork — open models running locally perform excellently. This is not about writing poetry: it is about processing your documents without them leaving the building.
