What is private AI?

How it works

An AI model is a set of very large files. Some companies keep those files to themselves and rent you access over the internet. Others publish them, and anyone can download it and run it on their own machine. Those are called open models. OpenAI’s gpt-oss and Alibaba’s Qwen are two of them.

A private AI setup is made up of:

• A server with powerful graphics chips (GPUs) and enough memory to hold the model. Ours is about the size of a desktop tower.
• An open model, downloaded once and kept on that server.
• A chat app your team types into. It handles logins, saves chat history, and connects to your files and email.

When someone asks a question, the server works out the answer on its own. Nothing goes to an AI company, and it keeps answering if your internet goes down.

Private AI vs. cloud AI

What vendors mean by “private”

Vendors use the word for different setups, so ask which one you’re buying.

• Business plans from cloud AI companies promise not to train on your data. Your files still go to their servers and stay there under their retention rules. What that means for client data.
• Private cloud runs the model in a cloud account set up just for you. Your data is kept apart from other customers’ data, but it still sits in someone else’s data center.
• On-premises runs the model on hardware in your building, and your data never has to leave. That’s what we install.

Is it as good as ChatGPT?

For everyday office work, open models do the job well:

• Summarizing long documents, depositions, and records
• Drafting emails, letters, and proposals
• Pulling figures from statements and spreadsheets
• Comparing two versions of a contract
• Searching your own files by asking in plain English

On the hardest reasoning and coding problems, the top cloud models are often still ahead. A server with more memory can run a larger model, which closes more of that gap. To see how it does on your own work, try the free pilot: we run one of your real workflows on a demo machine.

Who it’s for

Any firm that holds other people’s private information and wants its staff to use AI without sending that information out. For example:

• Accounting and tax firms, whose client returns fall under Section 7216 and the AICPA confidentiality rule.
• Law firms, whose client files carry confidentiality and privilege duties.
• Healthcare practices, whose patient records fall under HIPAA.
• Defense contractors, whose controlled information falls under NIST 800-171 and CMMC.

Private AI doesn’t make a firm compliant by itself. But it answers the biggest question in a compliance review: which outside company now has your data.

What it costs

You pay for the hardware once, then a monthly fee for software, updates, and support. Our systems start at $40,000 upfront plus $2,000 a month for a team of up to 8. Larger systems are sized for a whole department, and starting prices for every system are in our pricing section. You own the hardware, and if you cancel, you keep it and your data.

What you need on your side

You don’t need an AI engineer. We install the server, connect the tools you approve, and keep it running. You need one person who can approve network access and decide which data the AI can see.