Is private AI as good as ChatGPT?

For everyday office work over your own documents, yes. For the hardest reasoning tasks, the largest cloud models are still ahead. We map your tasks to both sides and, where you want it, build a setup that uses the local model by default and asks you before anything goes to a cloud model.

What does the machine cost?

Between a good workstation and a small server, depending on volume. You get a specific recommendation with a specification after the call, and you buy the hardware yourself, from any vendor you like.

Do we need our own IT department?

No. The setup runs like an appliance. Updates and monitoring run on a retainer with us or through your existing IT partner, whom we brief.

The data stays on hardware you control, which removes the AI vendor from your list of processors for this work. Your own duties under the GDPR remain, and we hand over the system description your data protection officer needs for the records.

Fully private AI · open models on your own hardware

Fully private AI, on your server.

Free 30-minute call

Pick a time.

Frederik Theissen, founderTakes every first call himself.

Booking runs through Cal.com. Your name and email go to the calendar invite and nowhere else. Details in our privacy policy.

01Thirty minutes with the founder. We look at what your team already pastes into ChatGPT, and what must never go there.
02A straight answer. Which jobs a local model does well today, which still need a bigger one, and what the machine costs. If it is not worth it, we say so.
03A written next step. One fixed price. The hardware is yours, the models are open, the logs stay on your disk.

What it looks like

One machine in your office.
Everything on it, nothing outside.

In plain terms

Private AI, in plain terms

The contract your paralegal pasted into a chatbot last Tuesday went to a server you have never seen, under terms you have never read. For many firms that is the whole reason AI is still banned in the office. Private AI removes the trip. An open model runs on a machine in your building, or on a server rented under your own name, and your documents never leave it.

People also call this on-premise AI, a local LLM, self-hosted AI or an on-premise ChatGPT. The idea is the same in every case: the model is a program you run, like your accounting software, and the data stays on your disk.

Current open models handle the work a small firm needs most: reading and sorting mail, drafting replies from your documents, summarising long files, answering questions from the knowledge vault, pulling fields out of invoices and forms. The very largest cloud models are still stronger at hard reasoning. We tell you which of your jobs fall on which side before you buy a machine.

The hardware is smaller than most people expect. A workstation with a strong graphics card serves a team of ten to twenty for everyday tasks. A firm with heavier volumes gets a server in its rack. Either way you buy it once, you own it, and there is no per-user fee for the model.

For firms in Germany, Austria and Switzerland this is often the cleanest answer to the data question. Client data stays in the country, the list of processors gets shorter, and your data protection officer can read the request log on the machine itself.

Limits we name up front. Someone has to keep the machine running and the models updated; we set that up and can hold the retainer for it, or hand it to your IT partner. Open models improve every few months, so the setup is built to swap the model without rebuilding the rest. And a private setup only pays off when the volume or the sensitivity of the work justifies it. If it does not, we say so on the call.

Questions we get asked

Is private AI as good as ChatGPT?
For everyday office work over your own documents, yes. For the hardest reasoning tasks, the largest cloud models are still ahead. We map your tasks to both sides and, where you want it, build a setup that uses the local model by default and asks you before anything goes to a cloud model.
What does the machine cost?
Between a good workstation and a small server, depending on volume. You get a specific recommendation with a specification after the call, and you buy the hardware yourself, from any vendor you like.
Do we need our own IT department?
No. The setup runs like an appliance. Updates and monitoring run on a retainer with us or through your existing IT partner, whom we brief.
What about the GDPR?
The data stays on hardware you control, which removes the AI vendor from your list of processors for this work. Your own duties under the GDPR remain, and we hand over the system description your data protection officer needs for the records.

Who you talk to

Frederik Theissen

A former academic researcher who has spent a decade building data and AI systems that run in production: analysis systems in neuroscience research, quantitative risk at a crypto hedge fund, and AI workflow automation for the analyst team of a Swiss market-intelligence firm. Building with large language models daily since December 2022.

Joinery builds the vault, the harness around it and the first agents that work from it, then hands it over as yours. I run my own company this way: one vault, a team of agents on top of it.

Founding-client terms available for the first cohort. Mention it on the call.

Start with the call.

Thirty minutes on what your team can do with AI without sending a single document out of the house. A clear answer, a concrete next step, and a working system within weeks if it fits.

Pick a time or write to hello@joinery-ai.com