The Appliance

Introducing

airon

Sovereign AI appliance — currently in development.

A purpose-built, air-gapped private AI appliance designed to anchor the sovereign end of your operation's inference topology. Runs locally, on your hardware, for the work that must stay in your environment. Approved cloud services, where you use them, run under the same controls.

airon is in active development. Specifications, form factors, and deployment models may evolve as the product matures.

Scroll

In Development

Hardware for the rooms where your data stays.

A sovereign AI appliance, engineered in Australia for the work that stays inside the building. Full reveal to come.

A single warm amber light seam suspended in pure darkness — Sovereign.

A dim server-hall corridor with one rack bay glowing warmly through its vented door — Private.

Macro detail of a matte white surface lit by a single warm grazing light — Australian.

airon — in development. Imagery is indicative of mood and intent; specifications, form factors, and finishes are subject to change.

Context

Where airon fits.

Airon Systems is an Australian private AI consultancy. We design and deploy private AI architectures for Australian businesses, combining appliance pilots, model runtimes, retrieval, governance, and cloud-to-edge routing into one operating model.

airon is the Airon Systems appliance at the centre of that topology — a self-contained, on-premise private AI appliance concept built specifically for air-gapped deployment, in environments where control of the data is the starting point — where policy, not convenience, decides what a model can reach.

The base offering, as a product

Air-gapped, on-premise AI is what Airon builds and runs. The appliance is that base offering in product form — one machine, installed on your site.

See where AI can run →

Beside the AI you already run

The appliance replaces nothing your teams rely on. Policy decides which work stays local and which uses approved services, and the same controls apply to both.

How placement is decided →

A service you can buy today

AI infrastructure design, configuration, sourcing and integration are delivered now as a service; airon is its productised form. An appliance pilot is a Governed Pilot with hardware included — same steps, same keep-or-stop gate.

See the infrastructure service →

Planned Hardware

Hardware form factors under development.

airon is being designed around specialised, low-power Neural Processing Units and deployable compute profiles — targeting high-throughput on-premise inference at desk-side power budgets.

Executive Node

Boardroom Class

A compact, low-power appliance designed for boardroom or executive office deployment. Intended to provide sovereign retrieval for leadership teams without requiring server room infrastructure.

Core Hub

Rack-Mounted

Industrial rack-mounted configuration designed for high-throughput site control centres. Targeted at sustained NPU inference handling concurrent workloads at operational scale.

Edge Mesh

Low-Power Nodes

Scalable edge SoC devices designed for embedding in heavy machinery or rugged field tablets. Intended for offline voice transcription, SOP caching, and local inference where connectivity is limited.

Reference Configuration

Pilot hardware target.

A compact reference profile for on-premise inference pilots: enough local compute, memory, storage, and I/O for document intelligence, video analysis, and private retrieval without relying on cloud processing.

Reference configuration shown for pilot planning only. Final production specifications, availability, and deployment packaging may vary.

AI compute

80 TOPS edge NPU

System power

70 W thermal design envelope

Memory

32 GB DDR5

Storage

512 GB NVMe

Networking

3 x 2.5GbE

I/O

HDMI, 2 x USB 3.2, 4 x USB 2.0

Power input

19 V DC

Dimensions

178 x 116 x 67 mm

Weight

1.3 kg

Operating range

-20 C to 60 C

Planned Capabilities

Designed as appliance plus control plane.

Four systems are being designed in: model orchestration, policy controls, retrieval, and an audit ledger. That is the difference between a model in a box and something you can let near your safety files. Answers keep their sources. Policies hold. Every query lands in the log.

Multi-Model Architecture

Designed to run multiple sovereign-friendly, open models simultaneously and cross-examine their outputs on complex queries — reducing single-model failure modes.

Programmatic Safety Layer

A rules-based safety layer intended to intercept outputs that deviate from configured policies, enforcing approved fallbacks. Designed to limit prompt-injection risk and maintain output compliance.

Structured Document Retrieval

Organises ingested documents into logical retrieval spaces, serving approved content verbatim rather than relying on generative summarisation. Designed to reduce hallucination risk for policy and procedure queries.

Audit Ledger

Query and output logging is planned as a first-class, on-device operating requirement — designed for regulatory inspection and compliance workflows, supporting chain-of-custody needs.

Deployment Model

Local first. Governed by design.

The pattern airon is being shaped around is simple. Sensitive data stays inside your walls. The model reads only from sources you have approved. Policy sits between the model and the user, and every output can be traced back to what produced it.

Approved sources

→

airon appliance

→

Model, retrieval, policy, audit

→

Approved workflows

Security Model

Control surfaces that matter.

The appliance is not just a box. A production-ready deployment needs a governance envelope around source data, users, prompts, outputs, and records.

Local Processing Boundary

Pilot workloads are designed to run inside the customer environment, with sensitive prompts, source documents, and outputs kept away from public AI services.

Approved Source Retrieval

Document answers are intended to come from approved retrieval spaces, with source-aware responses and controlled fallback behaviour for unsupported questions.

Policy & Safety Layer

Configured rules can steer outputs, block disallowed patterns, and route uncertain responses to human review instead of improvising.

Audit Trail

Query, source, and output logging are planned as a first-class operating requirement so decisions can be reviewed after the fact.

Integrations

Designed to meet existing systems.

Pilot planning starts with the systems that already hold the work: SharePoint and Microsoft 365, the file shares, the document register, the camera feeds, your directory. The appliance has to meet them where they are — otherwise it is one more system to feed.

Microsoft 365 / SharePoint

Network file shares

Document management systems

CCTV / VMS feeds

SSO / directory access

Ticketing, CRM, and ERP touchpoints

Pilot Timeline

Week 1

Scope

Confirm workload, data sensitivity, users, site environment, success criteria, and deployment constraints.

Week 2

Load

Prepare approved source material, access model, pilot prompts, and evaluation cases.

Weeks 3-4

Pilot

Run controlled users through representative workflows and capture accuracy, latency, and adoption notes.

Review

Decide

Report findings, open risks, production requirements, and whether the airon appliance path is the right fit.

Target Workflows

What airon is being built to do.

Tender Analysis & Estimation

Secure ingestion of complex commercial tenders. Designed to extract scope, obligations, and risk clauses locally — without exposing pricing data to external services.

On-Premise Video Analytics

Real-time CCTV and drone feed analysis processed on-device. Targeted at PPE compliance, exclusion zone monitoring, and incident reconstruction — without streaming video externally.

Document Intelligence

SWMS generation, policy extraction, and contract risk analysis from approved templates — policy-compliant output with audit traceability designed in.

Ambient Site Intelligence

Offline voice notes from site, transcribed and classified locally on edge nodes. Intended to auto-populate project logs, safety observations, and daily reports without connectivity.

Knowledge Base Retrieval

Operational FAQ and institutional knowledge served from approved sources. Designed to give field crews and new starters instant, accurate answers to operational questions.

Compliance Automation

Cross-referencing of project documentation against current legislation and internal policies — designed to flag potential non-compliance for human review, with remediation reports and audit trails.

Pilot Readiness

What makes a pilot useful.

A named executive or operations sponsor
A high-value workflow with measurable success criteria
Approved documents or feeds for pilot ingestion
A site, server room, or office environment suitable for local hardware
A small review group that can test outputs against real work

Not Final Yet

Open decisions before production.

Production enclosure, finish, and mounting options
Final pricing and support tiers
Exact production configuration by workload
Certified deployment partners and procurement path
Availability windows for broader rollout

Pilot Pack

Bring the concept into review.

The pilot brief summarises the reference configuration, security model, deployment topology, and open production decisions for internal evaluation.

PDF Brief

airon Pilot Brief

A concise overview for stakeholders assessing whether a sovereign, on-premise AI appliance belongs in the pilot roadmap.

Download Brief→

Early Access

Interested in airon?

airon is in active development. If the concept aligns with your sovereign inference requirements, we'd welcome a conversation about pilot fit and deployment timing.

Specifications, pricing, and availability are not yet finalised.

Start a Pilot Conversation→Inference Audit→