Home / Solutions

Deployments

Three machines. One principle:
the model stays with you.

Every AIOD appliance is a sealed, self-contained inference system — Nemotron or Gemma weights resident on local silicon, a knowledge corpus curated for the mission, and no external dependency of any kind. Choose the chassis; the principle never changes.

Unit F-1 // Portable

AIOD Field

Ruggedised inference in a sealed transit case, built for the places where help can't call out. A trained responder opens the lid, asks the question, and gets protocol-grade guidance — at altitude, at sea, or beyond the last cell tower.

SPECIFICATION

The machine

computeNVIDIA Jetson Orin
modelNemotron Nano / Gemma (quantised)
throughputreal-time conversational
powerbattery / solar / 12V vehicle
runtime~18 hr autonomous
chassissealed IP-rated transit case
interfacebuilt-in display + local Wi-Fi
CORPUS

What it knows

Mission-curated retrieval: wilderness and remote medicine protocols, casualty triage frameworks, mechanical and marine repair manuals, navigation and survival references, regulatory documents — in the languages your team works in.

Important: Field units are decision-support tools for trained personnel. They augment protocols and training; they do not replace emergency services or clinical judgement.

MISSION PROFILES

Where it goes

Mountain rescue stations and alpine refuges. Vessels on bluewater passages. Expedition base camps. Remote clinics and research stations. Anywhere the question "what do we do now?" can't wait for a signal.

OFFLINE AI FOR EMERGENCY RESPONSE →

Unit C-1 // Resilience node

AIOD Civic

One node, one town. A community knowledge server installed in a library, town hall or emergency operations centre — quietly useful in peacetime, indispensable when the grid and the internet are gone. If it has power, the whole community has answers.

SPECIFICATION

The machine

computeDGX-Spark-class node
modelNemotron Super · Gemma 27B class
concurrencydozens of simultaneous users
accesslocal Wi-Fi mesh, any browser
powermains / generator / battery bank
fallbackruns degraded on UPS
CORPUS

What it knows

A civilisation-in-miniature reference library: emergency medicine, water purification and sanitation, power and generator repair, food preservation, structural safety assessment, communications procedures, local emergency plans — plus an offline encyclopaedia.

Local knowledge: we ingest your council's own documents — flood plans, evacuation routes, asset registers — so the node speaks for your town.

MISSION PROFILES

Where it goes

Public libraries and town halls. Council emergency operations centres. Civil-defence and community-resilience programmes. Island and rural communities where connectivity was never guaranteed in the first place.

COUNCILS & RESILIENCE →

Unit P-1 // Air-gapped

AIOD Private

Full LLM capability with zero exfiltration risk. Your documents, your model, your building — confidentiality enforced by physics, not policy. Built for organisations whose data must never, under any circumstances, transit a third party.

SPECIFICATION

The machine

computeRTX workstation → DGX rack
modelNemotron or Gemma, fine-tuned
RAGretrieval over your private corpus
usersteam to department scale
networknone. ever.
auditfull local logging, your custody
SECURITY MODEL

How it's locked down

No network interfaces enabled at handover. Signed, encrypted update media. Full-disk encryption at rest. Verified air gap with documented attestation. Zero telemetry — we couldn't see your usage even if we wanted to, and we don't want to.

FULL SECURITY MODEL →

MISSION PROFILES

Where it goes

Law firms running privileged review. Hospitals analysing records under GDPR. Defence contractors and classified environments. Pharma and deep-tech R&D where the prompt itself is the trade secret. Hedge funds, family offices and digital-asset firms — where the prompt is the position.

SEE ALL INDUSTRIES →

Sustainment

Knowledge Packs: updates without a network.

An offline machine should still improve over time. Quarterly, AIOD ships a signed, encrypted drive containing model upgrades, software patches and refreshed corpus material. Your team plugs it in; the appliance verifies the signature and applies the update. The air gap is never broken.

Q1–Q4 / CADENCE

Quarterly by default

Faster cadences available for clinical and regulatory corpora that move quickly.

SIG / INTEGRITY

Cryptographically signed

Every pack is hashed and signed. The appliance refuses anything it can't verify.

ENC / TRANSIT

Encrypted in transit

Hardware-encrypted media, tracked courier, tamper-evident packaging.

LOG / CUSTODY

Audit trail, your custody

Update history lives on the device, in your records — not on our servers. We don't have servers for it.

Questions

Asked before every deployment.

Do AIOD appliances ever connect to the internet?

No. They are designed, installed and verified as air-gapped systems. Model weights, retrieval index and knowledge corpus live on the device. Updates arrive on signed, encrypted drives — never over a network.

What happens if the hardware fails in five years?

You own the machine and the weights outright, and our maintenance contracts cover diagnostics, spares and like-for-like replacement. There is no service that can be discontinued out from under you — that's the point.

Can you fine-tune the model on our documents?

Yes. For most missions, retrieval (RAG) over your corpus is the right tool and keeps your documents as documents. Where genuine fine-tuning is justified, it's performed on isolated hardware under NDA, and every artefact is handed over or destroyed — your choice, documented.

How long does a deployment take?

A standard Private or Field deployment typically runs 6–10 weeks from threat-model workshop to verified on-site handover. Civic deployments vary with corpus and council integration work.

Is an offline model as capable as a cloud model?

Frontier cloud models are stronger in absolute terms. But a well-sized Nemotron or Gemma appliance with retrieval over a mission-specific corpus is remarkably capable at its job — and it answers when the cloud can't, on data the cloud must never see. See our cloud-vs-local comparison.

Begin deployment

Tell us the mission. We'll spec the machine.

The mountain, the courtroom, the town, the lab — every deployment starts with a one-hour threat-model conversation.

DEPLOY@AIOD.APP →

response < 24h · NDAs welcome · on-site demos available