Member of NVIDIA InceptionNVIDIA Inception
Neural ICE
EN · FR · DE · IT Talk to us
From open weights to real work · Built on neutral ground in Switzerland

Your private AI.
Open models, in a box on your desk.

AC:1 turns open-weight models into a ready-to-use individual AI appliance. You choose a local model from the supported catalogue; AC:1 augments it with a bounded agentic harness, grounded in your sources and equipped with deterministic professional tools, to produce verifiable work. It also secures the device across its lifecycle. Your documents, prompts and responses stay on your premises.

// keep your neurons cool
Private data · Independent models · Sovereign operation Your choice of model · Local by design Open intelligence · One professional · One device NVIDIA DGX Spark · first platform validated by Neural ICE
the cloud your walls NI your AI device encrypted link your computer your content stays here
The missing last mile

Open weights are available. Professional use still is not.

Downloading a model does not provide a supported choice, secure it, evaluate it, update it or integrate it into real work. Neural ICE builds that productisation and distribution layer.

Weights are a foundation

A model is not a professional product

Weights alone do not provide inference, document integration, agentic tools, evaluation, traceability or a maintainable user experience. Someone still has to assemble the complete system.

Right-sized intelligence

The biggest model is not always the right model

Frequent, bounded tasks often need a specialised small model; complex work may require a larger one. The supported catalogue makes those trade-offs explicit, so you choose the model without having to assemble and maintain its runtime.

The lifecycle

Every model becomes a security and maintenance problem

Models, runtimes and tools evolve independently. A professional system needs verified components, signed updates, rollback, auditability and controls that remain understandable over time.

Strategic autonomy

No single cloud or model should define your future

A provider can change a model, its features or its terms. Neural ICE keeps models interchangeable and execution local so an essential capability does not depend entirely on one outside decision.

The productisation layer

What is Neural ICE, exactly?

The open AI stack, assembled into an individual professional device.

The physical product

A secure individual AI appliance, ready to work

AC:1 turns a validated hardware platform into professional equipment. It arrives preloaded, connects to your local network and works on your desk without a complex infrastructure project. You buy the machine and keep the tool beside you.

The software product

Your chosen model, equipped for real work

Neural ICE provides a catalogue of supported models; you choose which one to use. AC:1 places it inside a bounded agentic harness with multimodal knowledge, specialised agents, deterministic tools, source grounding and quality gates — then produces usable professional deliverables.

The durable outcomes

Private data. Independent models. Sovereign operation.

AC:1 processes your content directly on the AI device installed on your premises. Its local analysis, retrieval and generation capabilities remain available without an Internet connection. Client files, HR data, contracts and strategic information do not leave the device for processing: they are sent neither to our infrastructure nor to a cloud AI service. Only the technical flows required to operate the product and those triggered by options you enable — connectors or Nomad Access — may leave the device; they are documented in the traffic matrix →

From model to production system

You choose the model. Neural ICE gives it the means to work.

Every supported model runs within the same bounded agentic harness: multimodal RAG, structured understanding, specialised agents, deterministic tools, source lineage, quality gates and human oversight.

Supported model catalogue

Your model, your choice

The catalogue makes compact specialised models and larger general models available with clear capabilities. You remain in control of which supported model runs your work.

Multimodal knowledge

Understands text, images and page structure

Searches both the text and the visual content of PDFs, scans and image files. It can find a relevant passage, page, table or visual element, then return it with its source context.

Structured understanding

Turns complex files into usable data

Identifies entities and repeated fields, extracts them into typed datasets and preserves a link to the supporting source span. Quality controls make the result inspectable before it is reused.

Agentic production

Completes multi-step business tasks

Coordinates specialised agents and deterministic tools to compare sources, calculate KPIs, join datasets, analyse results and consolidate the work into one deliverable under human oversight.

Professional outputs

Produces files, not only answers

Creates reports in PDF or editable Word format, spreadsheets, charts, analyses and presentations in the formats your team already uses, with source lineage and an audit history.

Mail and calendar

Works with Google Workspace and Microsoft 365

Optional local connectors let the agent work with Gmail or Outlook mailboxes, prepare drafts and create calendar events. Each connector is enabled explicitly and operates under your credentials.

Nomad Access · add-on

Reach your AI device while you are away

The Nomad Access option connects authorised devices to your AI device through an end-to-end encrypted private network. No service is exposed directly on the Internet, and every device retains its own mTLS authentication.

Swiss legal knowledge · add-on

Swiss case law, available locally and offline

The Legal Assistant can search and cite a local corpus of about one million Swiss decisions in French, German and Italian. It packages OpenCaseLaw's open corpus and software ↗. Neural ICE mirrors the source and corpus and operates its own ingestion and update chain, so customer devices do not depend on OpenCaseLaw's public online service.

One appliance · the complete stack

Power it on. Connect your computer. Start working.

Your chosen model, inference runtime, documents, bounded agentic harness, professional tools, security and signed updates arrive as one product. No cloud account or specialist integration project is required.

01

Power it on

It arrives preloaded and configured on a platform suited to your needs. Unbox it and plug it in: everything is ready.

02

Connect your computer

Install the Neural ICE app on your computer and pair it with your AI device during initial setup. You can then easily add other compatible devices. By default, every connection remains on your local network; the Nomad Access option also lets you reach your AI device remotely through an end-to-end encrypted private network.

03

Start working

After a few minutes of first-start setup, Neural ICE is ready for daily work without routine system administration. Managed, signed updates are installed for you and retain the previous release for recovery.

the machine · validated platforms validated platforms dgx spark class or rtx pro series the stack · preloaded before shipping ice-coreos · immutable os multimodal rag · your documents ac:1 · private agentic ai in use · sealed, local, yours like a console or nas secure thin-client app encrypted link local processing · no cloud AI service
The open-model last mile

The simplicity of an online service. The control of a product on your premises.

SolutionAccess to AIYour files stay with youDedicated environmentSimple setupUpdates included
Working without AINoYesYesYes
Consumer online AIYesNoNoYesYes
Enterprise cloudYesNoNoComplex projectYour responsibility
Custom local solutionYesYesYesComplex projectYour responsibility
Neural ICE AC:1YesYesYesYesYes
Secured across the device lifecycle

Security is part of the productisation layer.

From verified boot to signed distribution and local data boundaries, security is built into the device and its lifecycle. The result is confidentiality, strategic autonomy and sovereign operation backed by named controls.

A software foundation protected against modification

The system is signed and verified through the boot chain. Atomic operating-system updates retain the previous deployment so it can be restored if needed.

Encrypted at rest and in transit

Data stored on your AI device is encrypted, and internal communications are authenticated and protected. Keys are created and retained inside the machine.

A genuinely dedicated environment

Your AI device, models and memory are not shared with other customers. Isolation rests on dedicated hardware, not only on a software setting.

Concrete evidence for your audits

Detailed traceability, clearly identified models and documented architecture provide concrete evidence for auditors and regulators. Neural ICE is designed for environments subject to GDPR, the Swiss FADP, the AI Act and FINMA requirements.

// key platform controls · detailed status in the technical brief

Verified boot chain Modification-protected system Encrypted data Controlled access No service intended for public Internet exposure Tamper-evident audit trail Offline operation available Export to a SIEM if you use one
The ownership model

The device is yours. The lifecycle is maintained.

You buy the AI device hardware. The subscription covers access to the private software, updates, security fixes and new capabilities so the service remains dependable without becoming another system to administer.

Your equipment

The hardware belongs to you

You buy and keep the physical AI device. The licence service validates your software entitlement; it is not a remote administration channel into your system.

We maintain

Reduced routine administration

The operating-system update mechanism uses signed, atomic images and retains the previous deployment for recovery. We monitor vulnerabilities and prepare fixes. A software bill of materials (SBOM) can be provided for your audits.

It evolves

New capabilities without replacing the machine

The subscription includes compatible model releases, additional features and security improvements for your AI device.

Open intelligence becomes useful only when it remains secure, maintainable and straightforward to use every day.

Your hardware. Clear service terms.

You buy the AI device hardware. The subscription covers access to the private software, updates, maintenance and new features.

Nomad Access is offered as an add-on. It lets authorised devices reach the AI device remotely through an encrypted private network without exposing its services directly on the Internet.

Your documents, knowledge-base archives, configurations and adapted models remain stored locally and can be exported in documented, portable formats. The final offer will state continuity, renewal and exit terms explicitly before you commit.

Pricing and licence terms will be announced at launch for each platform. Founding customers will receive their launch conditions in writing before ordering.

The first units will be offered to founding customers first.

Hardware manufacturers and partners: let's talk.
Consulting · AI governance

Clear, documented AI governance, ready for audit.

Alongside the AI device, the founders help organisations establish an AI management system aligned with ISO/IEC 42001, from the first assessment to certification-audit preparation.

The team behind Neural ICE

Built by cybersecurity and data-protection specialists.

Neural ICE is developed in Switzerland by a team with backgrounds in cybersecurity, governance and data protection. Selected for the Trust Village incubation programme in October 2025, the team works on SICPA's unlimitrust campus in the Swiss digital-trust ecosystem. When you contact us, you speak directly with the people building the product.

TK

Thomas Kristner

Co-founder · CEO & CTO

Thomas has more than fifteen years of experience securing critical systems. His work includes serving as deputy CISO at Audemars Piguet, architecting a next-generation SOC at Kudelski Security and working on encrypted communications for the defence industry. He holds CISSP and CEH certifications.

Before focusing on cybersecurity, Thomas built composite parts for aerospace and Formula 1. That culture of precision and zero defects guides Neural ICE today.

LinkedIn ↗
NU

Nuria Unterlohner Salvat

Co-founder · Cybersecurity and Governance Advisor

Nuria has experience in security governance across regulated Swiss environments. She works in banking cybersecurity and previously held roles at Deloitte and Nestlé. She brings financial-sector scrutiny to Neural ICE and holds the CISM certification.

She challenges the product's risks, controls and governance. Every measure must withstand detailed review, not only a sales presentation.

LinkedIn ↗
MM

Marie G. Mansour

Co-founder · DPO and Data Protection Lead

Marie specialises in data protection and works as a Data Protection Officer. She previously worked as a privacy consultant at Deloitte and as a researcher at IDHEAP. She holds CIPP/E and ISO 27001 Lead Auditor certifications.

She turns GDPR, the Swiss FADP and AI Act requirements into concrete product and governance decisions.

LinkedIn ↗
Let's talk about your work

What work would you entrust to truly private AI?

Tell us about the documents, tools and tasks you need to keep under control. We will show you how AC:1 can combine your knowledge, an open model you choose and a bounded agentic harness — directly on your premises.

  • A no-obligation scoping conversation directly with the founders
  • A demonstration focused on the workflows that matter in your profession
  • A clear proposal for integrating a pre-installed AC:1 appliance
We use the information you provide only to contact you. We do not sell it or share it for commercial purposes, and this site uses no advertising trackers. Your AC:1 AI device runs on your premises. Full privacy notice →

Request received.

We will review your use case and contact you as soon as possible.

Built on neutral ground in Switzerland Independent models · local execution Prepared for AI Act governance Data protection by design Incubated at Trust Village · unlimitrust campus 🇨🇭 Member of NVIDIA Inception