On-Prem AI, Built in the EU
AI that never leaves your desk
Pre-installed agents for KYC/AML, claims and clinical documentation — running inside your own infrastructure, air-gapped if you need it. Built for banks, insurers and hospitals on the NVIDIA GB10 Grace Blackwell Superchip.
Engineered & supported in the EU/Air-gap capable/No data ever leaves your premises
250 × 250 × 275 mm
Why On-Prem
Every prompt sent to a third-party model is a compliance question you can't fully answer
Cloud AI was built for speed of adoption, not for the scrutiny regulators apply.
Data residency you can't fully guarantee.
Even "EU region" cloud AI still runs on infrastructure operated by a company you don't control, with terms that can change.
US CLOUD Act exposure.
If your vendor is a US company, US authorities can compel disclosure wherever the data is hosted — a gap most EU residency claims don't address.
Compliance retrofitted after the fact.
Most AI platforms were built for experimentation, then bolted onto GDPR and the EU AI Act afterward. Audits find the gaps.
Per-token cost, unpredictable at scale.
Per-token pricing scales with exactly the workload you're proudest of, right when volume matters most.
Complete Software Stack
Pre-built agents, not a framework to build your own
No separate hardware vendor, model vendor and integrator to coordinate — and no ML team to hire. Compute, runtime and agents ship together.
Runtime
A local inference engine with an OpenAI-compatible API, serving open-weight models at full speed on unified memory.
Agents
KYC/AML review, claims triage, clinical documentation and knowledge search, running from day one. Add your own on the same framework — each one sandboxed.
Knowledge
Index document stores, case files and internal systems in place. Answers cite your own documents, and nothing is sent out.
Governance
Role-based access, immutable audit logs, and policy controls your compliance team can actually verify.
Security
Built to run with zero trust in anything outside your walls
Every layer assumes the network around it, and everyone on it, could be hostile.
Air-Gap by Design
Disconnect the uplink and the system keeps running — fully isolated, permanently or on demand.
Hardware Root of Trust
Secure boot and signed firmware across all three units, so only software your team approved ever runs.
Sandboxed Agents
Every agent runs isolated. A compromised one can't reach another agent's data or tools.
Zero Telemetry
Nothing phones home — not to JetNode, not to NVIDIA. Logs are for your security team, not ours.
Detailed security architecture documentation is available under NDA for your security review — request it here.
Compliance
Built around the frameworks that matter in the EU
Not a module you add later — a consequence of the architecture. Open a framework to see how.
EU AI Act
Covers high-risk uses like credit scoring and risk assessment, from August 2026.
| Risk & data governance | Continuous risk tracking, and training/retrieval data that never leaves your infrastructure. |
| Logging & documentation | Every inference logged immutably; technical files export straight from the console. |
| Human oversight | Sign-off required before an agent takes a consequential action. |
GDPR
Applies to any processing of claims files, patient records, or account data.
| Impact assessments | Impact-assessment template pre-filled with your actual, on-site data flow. |
| Automated decisions | Human review enforced by default for any legally significant outcome. |
| Cross-border transfers | Moot by design — nothing leaves your infrastructure to transfer. |
NIS2
Banks, insurers, and hospitals are essential entities, with board-level liability.
| Risk management | On-prem, encrypted, and access-controlled by default. |
| Incident reporting | Audit logs feed straight into your existing reporting workflow. |
| Supply-chain security | One EU vendor, one documented bill of materials. |
DORA
Financial firms must treat every AI vendor as critical infrastructure.
| Risk framework | Runs on your network, under the risk framework you already have — not a new dependency. |
| Third-party register | One appliance, one vendor, one entry — not a chain of third parties. |
| Concentration risk | Running on site avoids putting a critical workload on one big cloud provider. |
A summary, not legal advice — final obligations depend on your use case. We work with your compliance team to confirm scope.
At A Glance
Cloud AI vs. Node_01
How an on-premise appliance compares on the metrics you actually get audited on.
Scroll the table sideways to compare →
| Cloud / SaaS AI | Node_01 | |
|---|---|---|
| Can run fully air-gapped | No — requires internet by design | Yes — disconnect the uplink, keep running |
| Data residency | Often leaves the EU | Never leaves your premises |
| US CLOUD Act exposure | Yes, if US-hosted | None — fully on-prem |
| Compliance posture | Retrofitted after findings | GDPR & EU AI Act ready by design |
| Cost model | Per-token, scales with usage | One-time hardware cost |
| Deployment time | Weeks of procurement & DPAs | Under a day |
| Data used for third-party training | Often unclear | Never — it's your infrastructure |
How It Works
From purchase order to production in under a week
No integration project, no consultants, no ML team to hire. The appliance model collapses deployment to three steps.
Scope the pilot.
A 30-minute call with your team: choose the first agent and use case, confirm the compliance scope, and receive a pilot plan your DPO and security team can review before anything ships.
Week 0 · RemoteUnbox to first inference in 15 minutes.
Node_01 arrives pre-configured with the Gateway and Switch. Connect the three units, power on, and run the first prompt on your own network — air-gapped from day one if you choose.
Delivery day · On-sitePoint agents at your documents.
Index your document stores, set roles and audit policy in the admin console, and roll out to the first team. We stay on hand until your compliance team has signed off.
Week 1 · Your team + oursIndustry Solutions
Built for banks, insurers, and anyone protecting proprietary data
Every claim review and credit decision can be audited — and every trade secret sent to an outside model is one you no longer control.
KYC, AML & Credit Review
Screen transactions, draft credit memos and answer compliance questions without account data leaving your network.
Claims & Underwriting
Summarise claims files and support underwriting without a policyholder's file leaving your network.
Clinical Documentation
Draft clinical notes and retrieve patient history from infrastructure inside the hospital's own walls.
Protect What You Can't Replace
R&D notes, source code and trade secrets stay off every third-party training set — they never leave your network.
Citizen & Case Services
Internal and citizen-facing tools that meet EU data-sovereignty rules without exception.
Spec Sheet
Full technical specifications
| Compute (Node_01) | NVIDIA GB10 Grace Blackwell SuperchipUp to 1 PFLOP FP4 AI compute (sparse) |
| CPU | 20-core NVIDIA Grace (Arm) CPU |
| Memory | 128 GB unified LPDDR5x (configurable)Coherent CPU + GPU memory architecture |
| Storage | 4 TB NVMe SSD, self-encrypting |
| Clustering | Link two Node_01 units to serve models up to 405B parameters |
| Networking (JetNode Switch) | 8-port managed 10 GbE switch (configurable)Isolated VLAN segment by default; remove the uplink for a full air-gap |
| Management (JetNode Gateway) | Fanless mini host PCRuns the admin console & API gateway, off the compute node's network path |
| Security | Secure boot & signed firmwareSandboxed agents, zero default telemetry, self-encrypting storage |
| Power | 240 W typical (Node), standard power outlet for all three units |
| Enclosure | 250 × 250 × 275 mm (W × D × H), 5.2 kgMachined anodised aluminium, front intake |
| Software | JetNode OS (Ubuntu-based, full CUDA-X stack)Runtime, Agents, Knowledge & Governance layers included |
Pricing
A cost model your CFO can audit in one line
Cloud AI bills scale with exactly the usage you want to encourage. Node_01 is capital equipment: you buy it once, and every prompt after that is free.
- One-time hardware purchaseNode_01, Gateway and Switch are yours — on your books as capex, inside your walls, under your control.
- No per-token or per-seat feesRoll it out to every employee. Your heaviest month costs the same as your quietest.
- Software and agents includedRuntime, agents, knowledge and governance layers ship on the unit — not licensed separately.
- Optional annual support planSigned software updates, new agent releases, and EU-based support — installable offline on air-gapped systems.
€7,999
One-time price. Volume and public-sector pricing available.
- Node_01 compute unitNVIDIA GB10, 128 GB unified memory (configurable), 4 TB encrypted NVMe
- JetNode Gateway & managed switchAdmin console, API gateway, isolated VLAN out of the box
- Full software stackAll four layers and every included industry agent
- First year of updates & supportSigned, offline-installable updates; EU-based engineers
- 3-year hardware warranty
Typical response time: 2 business days
FAQ
Before you book a demo
Does any data ever leave our premises?
No. Inference, retrieval and storage all happen on the appliance, inside your network. Nothing is sent to JetNode, NVIDIA or any third party.
Do we need to build our own agents?
No. Point a pre-installed agent at your document store and it works on day one. When a team needs something custom, add it on the same framework in the admin console — no ML engineering, no extra product to buy.
How does this satisfy GDPR and the EU AI Act?
Because processing never leaves your infrastructure, most cross-border transfer questions don't apply. Governance supplies the access controls, audit logs and documentation your compliance team needs — see the full breakdown above.
Which models does it run?
A curated set of open-weight models tuned for the GB10 Superchip, plus any compatible open model your team wants to standardise on.
What happens when we need more capacity?
Link a second Node_01 over 200Gb/s ConnectX-7 to serve models up to 405B, or add nodes across departments as demand grows.
How is it priced?
€7,999 as a one-time hardware purchase, with the full software stack included — no per-token or per-seat fees, ever. An optional annual plan covers signed updates, new agent releases and EU-based support. Volume and public-sector pricing is available on request.
What's in the box?
Three units that arrive pre-configured to work together: the Node_01 compute unit, the JetNode Gateway running the admin console and API gateway, and the managed 10 GbE switch — plus all cabling. From unboxing to first inference is about 15 minutes.
What about warranty and support?
Node_01 ships with a 3-year hardware warranty and a first year of support and updates included. Support engineers are EU-based, and software updates ship as signed bundles you can install offline — air-gapped systems never need to connect to receive them.
Get Started
Deploy AI your compliance team signs off on
See Node_01 answer questions over your own documents, on your own network, before you commit to anything.
EU-based team. The demo can run entirely on your own premises, too. Prefer to write first? Contact sales.