New The half-day pilot is free

Private AI for your whole team. Client data stays in the office.

Chat, a browser assistant and an admin console, running on a server we install and manage in your building. Everyone gets the AI they already know from ChatGPT, without sending your files to it.

Book a demo See how the pilot works
1 dayFrom delivery to your team signing in
Internet off?It keeps answering. Nothing it needs is in the cloud.
Every questionLogged on your server, for your admins only
From $40,000Upfront, installed. Prices published below.
Works with Google Drive Gmail Google Calendar Attio
The product

Everything your team needs to use AI at work, on hardware you own

No new habits to learn. If someone has used ChatGPT, they already know how to use On Premises.

Solutions

Built for firms that hold other people's private information

Your clients trust you with their files. Private AI lets your team move faster without asking them to trust anyone else.

Accounting & tax

Client returns, statements and workpapers stay on your server.

  • Check a packet for missing forms
  • Pull figures from statements
  • Draft client request emails
Client tax data and Section 7216 →

Law firms

Privileged files never sit on someone else's servers.

  • Compare contract versions
  • Summarize depositions
  • Search past matters
Client files and privilege →

Healthcare

Prompts and files stay in your building, inside your HIPAA program.

  • Draft referral letters
  • Summarize long records
  • Answer policy questions
What HIPAA still requires →

Engineering & architecture

Drawings, bids and client plans stay off the public internet.

  • Draft proposals from past bids
  • Search specs and RFIs
  • Summarize meeting notes
Talk to us about your projects →
How it works

Live in a day. Managed for as long as you have it.

You don't need an AI engineer. You need one person who can approve network access. We do the rest.

The server installed on the office floor, with power and network connected
01

We install the server

About the size of a desktop tower. It plugs into a normal outlet and joins your network the same afternoon.

02

We connect your tools

Your shared drive, email and calendar, with the permissions you approve. Nothing is copied off site.

03

Your team signs in

Each person gets their own login and saved history, in the browser and in Chrome.

04

We keep it running

We watch hardware health around the clock and install updates. We never see your content.

Security

The difference is where your data goes

Cloud AI is careful with your data, but your data still has to leave the building to reach it. Ours doesn't.

ChatGPT, Claude, CopilotOn Premises
Your questionsLeave your networkStay on the server
Documents you uploadCopied to their serversNever copied off
Record of who asked whatTheir logsYour logs
What we can see remotelyYour content, per policyHardware health only
If the internet dropsStops workingKeeps working
If you cancelDeleted on their scheduleYou keep hardware and data

Those vendors are careful. Your data still has to leave to reach them.

For your IT owner

Built to pass an IT review

One machine on your network, with nothing exposed to the internet by default.

  • Reachable only from your office network
  • Personal logins, roles, and one-click suspend
  • A request log of every question, kept on the server
  • Cloud models are off unless an admin turns them on
  • An OpenAI-compatible API for your own apps
Pricing

Simple pricing. You own the hardware.

The upfront price covers the hardware, burn-in, installation and setup. Software, updates and support bill monthly after go-live.

1 GPU · Focused team

1× RTX PRO 6000

A single tower, sized for a couple of people using it every day.

Upfrontfrom $40,000
+ $2,000/month software and upkeep
  • GPU memory96 GB
  • System RAM128 GB
  • Best fit1–2 people
  • Ready inTwo weeks
Start here
Recommended
2 GPUs · Growing team

2× RTX PRO 6000

Twice the power. Fits bigger models and more people at once.

Upfrontfrom $53,000
+ $2,000/month software and upkeep
  • GPU memory192 GB
  • System RAM256 GB
  • Best fitGrowing team
  • Ready inTwo weeks
Choose this system
4 GPUs · Company

4× RTX PRO 6000

Enough for a whole department, still one machine in one room.

Upfrontfrom $84,000
+ $2,500/month software and upkeep
  • GPU memory384 GB
  • System RAM256 GB
  • Best fit5–10 people
  • Ready inTwo weeks
Choose this system
GB300 · Frontier

DGX Station GB300

The largest machine we install, for the biggest models you can run.

Upfrontfrom $126,000
+ $3,200/month software and upkeep
  • Coherent memoryUp to 748 GB
  • Best fitLargest models
  • Ready in8 weeks
Choose this system

We order hardware once the upfront payment clears. Software and upkeep start at go-live on a 12-month initial term. Final configuration and supplier pricing are confirmed in the written quote.

Try before you buy

Not sure yet? Start with a free pilot.

Half a day, one of your real workflows, and a demo machine on the table. You leave with a go or no-go and a fixed quote.

See the pilot
Price$1,500 Free
TimeHalf a day, plus a written report
What we testYour real workflow, on approved samples
HardwareNo purchase required
FAQ

Questions we get before the first call

Nothing gets quoted until we've talked with your IT owner about the workload, who can reach it, and how it behaves on your network.

More answers →
Is this the same as ChatGPT?

It works the same way: you type a question, attach files, and get an answer. The difference is where it runs. Ours is a server in your office, so your questions and documents never go to an outside company.

Does our data ever leave the building?

Answers are produced on the server in your office, and it doesn't need the internet to answer a question. Anything the machine does reach out for, like model downloads, backups or our remote support, is set up to your rules and written into the deployment scope.

What happens if the internet goes down?

Your team keeps working. Everything needed to answer runs on the server in your office. Only plugins that reach online services, like Gmail, wait for the connection to come back.

Do we need an AI engineer?

No. We install the hardware and software, connect the tools you approve, and pick up the phone when something's wrong. We need one person on your side who can grant network access and sign off on the data.

Can we use it with patient records?

Only if your HIPAA program covers it. Prompts and files stay on the machine in your office, but no product makes a practice compliant by itself. What still applies under HIPAA.

What do the upfront and monthly prices include?

Upfront covers hardware, burn-in, on-site setup and initial configuration. Monthly covers the chat software, the model-serving stack, monitoring, routine updates, remote maintenance and support. Taxes, travel, replacement parts and custom integrations are quoted separately before you buy.

Are you affiliated with NVIDIA?

No. On Premises is an independent integrator. Manufacturer warranties and license terms stay between you and them.

See it answer your own questions

Bring a real task. We'll show you how it runs on a private server, and size the right system for your team.

Book a demo Try the free pilot