Private AI, on our own hardware
We run local AI models on an NVIDIA DGX Spark in our office in Ghent. Your documents, mails and questions never go to a cloud AI provider. They stay with us, and stay private.
No cloud AI provider · Models picked for your task
Why companies hold back on AI
Client files in a chatbot
Your team wants to use AI on real documents, and you don't want those documents leaving the company. With private AI they never reach a cloud AI provider.
A policy that just says no
If AI is simply forbidden, people end up using it on their own phone. A private option gives them something they're allowed to use.
Not knowing where your data ends up
With local models you can say exactly where prompts and files are: on one machine in Ghent, or inside your own network.
Your own AI, without the cloud
Nothing leaves
Prompts, files and answers stay on our machine. No cloud AI provider ever sees them.
Models for your work
We pick and tune open models for your task: documents, mail, internal knowledge.
Inside your own network
Local models can also run inside your own network, so your team can use AI on sensitive material without it leaving.
Connected to your tools
The same agents and automations we build for other work, with the AI part running locally.
Not a rented GPU
An NVIDIA DGX Spark in our office, not capacity in someone else's data centre. It already runs the AI behind our security scans.
One machine, many uses
The same Spark powers our security checks and the agents that watch your projects.
What runs on it today
Sidestream Security
The AI behind our scans runs on this machine, so your findings never leave the building.
Read more →Project monitoring
The agents that watch our clients' projects around the clock run on the same hardware.
From first call to live private AI
Every project runs the same way, and you always know what's next and what it costs.
- 01
Discovery
One call, usually the same week. We ask until the scope is crystal clear for both of us.
- 02
A fixed price
A written plan, a fixed price to build and what maintenance costs. The plan is yours to keep, even if you stop there.
- 03
Build
A working first version in two to four weeks. You see it working on real data while we build, not in a demo at the end.
- 04
Live, and kept running
Hosting, monitoring, updates and small changes on a monthly plan. The people who built it are the people on the phone.
What people ask us
What is private AI?
AI models that run on hardware you or your supplier control, instead of at a cloud AI provider. With us that means open models on an NVIDIA DGX Spark in Ghent, or inside your own network.
Does our data go to OpenAI, Google or anyone else?
No. Prompts, files and answers stay on our machine. No cloud AI provider ever sees them.
What can we use it for?
Mostly work with text you don't want to share: documents, mail and internal knowledge. We pick and tune the open model that fits the task.
Is it as good as the big cloud models?
For a lot of everyday work on documents and mail, a well-chosen open model does the job. For some tasks a cloud model is still stronger, and if that's the case for you, we'll say so.
Which hardware does it run on?
An NVIDIA DGX Spark with the Grace Blackwell GB10 superchip. Units can be linked over ConnectX-7 at 200 Gbps and share one memory pool, which is what decides how large a model you can hold.
What does it cost?
We quote a fixed price after a first call, based on what it has to do and where it runs. Not sure whether you need it? That's a good first call too.
Also from Sidestream
Want AI without sending your data away?
Tell us what you'd want to use it on. We'll tell you honestly whether private AI makes sense, and what it would cost.