Frontier Open-weight Models.
On your premises, on your terms.
Today’s best open models run on an appliance that sits on your desk. Your team uses it as much as it likes, for a fixed price, and your know-how stays with you.
Your cloud provider learns your business
Online AI is simple, but the company that provides it receives more than your questions and files. Request after request, it also learns how you work: your business processes, your clients, your suppliers and partners.
Even when the provider commits not to train its models on your conversations, its terms often allow other uses: so-called anonymised or aggregated data derived from your usage, the feedback you give, retention to detect abuse.
A custom local stack keeps all of this in-house, but brings a complex project to maintain. Neural ICE sits in between: the simplicity of a ready appliance, with processing on your premises.
| Solution | Access to AI | Your files stay with you | Dedicated environment | Simple setup | Updates included |
|---|---|---|---|---|---|
| Working without AI | No | Yes | Yes | Yes | n/a |
| Consumer online AI | Yes | No | No | Yes | Yes |
| Enterprise cloud | Yes | No | No | Complex project | Your responsibility |
| Custom local solution | Yes | Yes | Yes | Complex project | Your responsibility |
| Neural ICE | Yes | On the device | Yes | Yes | Yes |
The best open models, on your premises
Frontier Open-weight Models: two words, one simple promise.
A model you can run on your own premises
Its creators publish the “weights”, which is the model itself. So it runs on your machine, without sending your files to anyone.
The newest generation
We test it on the machine before we ship it; new models then arrive as signed updates. You are not stuck with last year’s model.
Your know-how stays yours
Lawyers and legal departments, fiduciaries, notaries, cybersecurity teams, executives and sales teams: their value lies in their files, their methods and their client relationships. Neural ICE brings them modern AI without handing any of it to a cloud provider.
Review a draft against your precedents
Ask which clauses depart from your usual positions. The answer points to the documents it relies on.
Prepare a file, check a deed
Turn a year of client documents into a working note, or compare a draft deed with the prior acts it relies on and see where each statement comes from.
Prepare proposals and meetings from your client files
Your prices, margins and client portfolio stay with you.
Analyse your vulnerabilities in-house
Cross-check scan reports, logs and threat intelligence with your own infrastructure data, which stays with you.
And for everyone else: an AI assistant that sorts your e-mail, prepares replies and appointments and writes up your meetings, on your premises.
What your team does with it
Your documents, your model, work you can check.
Pick the model
Open-weight models from a supported catalogue, compact or larger. Change the model without changing the product.
Ask your own documents
Question contracts, files and archives in plain language. Answers can point to the passages they rely on.
You choose what gets connected
Mail, calendar, Microsoft 365, Google Workspace or MCP connectors, off by default. The assistant sorts and prepares; sending stays with you.
Ready-to-use deliverables
Notes, summaries, tables, data analyses, charts, slides, editable office documents and much more, with sources attached.
Optional professional modules: CaseLaw CH · Cybersec · High Risk (AI Act). See pricing
In the office, in the field, even without the Internet
The Desk fits in a bag: a power socket and an Ethernet cable are all it takes. With the offline option, it takes your models and your documents where the network does not reach.
An AI that travels with you
Plug it in on assignment, on a construction site or at a client’s office.
Without the Internet, as an option
Dead zones, travel, isolated sites: the AI runs locally and updates arrive on signed media.
Where the Internet is scarce, costly or not allowed
Construction sites, mines, platforms, ships, isolated industrial networks: the AI works locally from the site office or control room, without depending on a satellite link or opening the network to the cloud.
Desk: 15 × 15 × 5 cm, 1.2 kg, about 165 W in typical use, operates from 0 to 35 °C, 10 to 90 % humidity (non-condensing), up to 3,000 m altitude.
Plug in. Pair. Start working.
One stack, from the hardware to the application: hardware and software arrive as one product. No cloud account, no integration project, no server to configure.
Plug in
We deliver the appliance in the format you chose, preinstalled and tested. Power and a network cable: that is all.
Pair your computers
Install the Neural ICE app and pair each computer with the appliance’s pairing code, on your local network.
Start working
Add your documents and connect your mailbox if you wish. Signed updates then keep the appliance current without routine administration.
Built for security, compliance and control of your data
Security your IT lead can check: a documented list of every flow, not a promise.
Your documents stay with you
The appliance contacts Neural ICE only for licence checks and signed updates. Your documents, conversations and usage stay on site.
Every flow is documented
Everything that enters or leaves the appliance is described in the flow matrix of the technical details.
Works without the Internet, as an option
For isolated sites, an offline mode is available on quote; updates then arrive on signed offline media.
Secured from boot
Verified boot chain, signed immutable OS with local rollback, hardware-bound disk encryption.
For your IT or security lead: read the technical brief →
Use it as much as you like. The bill stays the same.
Online AI bills by usage: the more your team uses it, the higher the bill. And an assistant working on its own through a long task consumes a lot. Here, the price is fixed.
No meter
No per-token or per-request billing.
Long tasks while you sleep
File analyses and research run overnight, at no extra cost.
Everyone uses it freely
Nobody has to watch consumption.
Buy the machine once. Subscribe to the software.
No meter, no surprises. Prices are revealed soon; the first units go to founding customers.
Three formats, delivered by us
NVIDIA DGX Spark · 1 to 2 userson quote
NVIDIA RTX PRO 6000 · up to 8 or 16on quote
NVIDIA DGX Station · up to 32coming soon
Preinstalled and tested before delivery.
Everything to get started
- 1 to 2 users included; extra users on Workstation and Station, up to 32
- Frontier Open-weight Models and signed updates
- Mail and calendar: Microsoft 365, Google Workspace
- Billed once a year
Swiss case lawsoon
Vulnerability assessment, threat intelligencesoon
AI Act compliance foundation for high-risk usessoon
Per appliance, billed annually. No per-request or per-token billing.
Hardware manufacturers and partners: talk to us.
Built in Switzerland by cybersecurity and governance specialists
Selected for Trust Village and based on SICPA's unlimitrust campus. When you write, you reach the people building the product.
The questions we hear first
Is an open model as good as ChatGPT?
For work on your own documents, the current generation does the job: analysing, comparing and drafting, with its sources cited.
Do we need an IT person?
No. The appliance arrives ready and updates itself. If you have an IT lead, they can check everything in the technical details.
Who can read our data?
You. Documents, questions and answers stay on the machine. They are not used to train a model.
Does it work without the Internet?
Yes, as an option on quote: isolated sites work offline and updates arrive on signed offline media.
What work would you entrust to truly private AI?
Tell us about your team’s documents and workflows. We will propose the Neural ICE appliance that fits your team.
- A no-obligation conversation with the founders
- Clear answers on your use cases and constraints
- Be told first when prices are announced
Request received.
We will review your use case and contact you as soon as possible.