openAut — v0.1 — Building Intelligence, Open

The building
knows. Now you
can too.

openAut brings supercomputer-grade AI to every building — without replacing your existing systems, locking you into proprietary platforms, or requiring expensive specialists on retainer. Open source. MIT licensed. Ready for public procurement.

See how it works GitHub →
Scroll to explore
The Problem

Buildings are smart. Their software isn't.

Every modern building is already generating a torrent of data — temperatures, pressures, energy draws, occupancy patterns, fault states. Your heating system, ventilation units, chillers and meters are running and recording around the clock.

But that data goes nowhere useful. Siloed inside proprietary BMS platforms, locked behind consultant contracts, invisible to the people who actually operate the building day-to-day.

The result? Energy waste that nobody sees. Faults that fester for weeks. Decisions made on gut feeling instead of data. And when you do want to change that, you face a bill for a new platform, a migration project, and a dependency on a single vendor for the next 20 years.

Public sector organisations face an extra layer: procurement law means you can't simply buy your way out. The solution has to be open, auditable, and reusable.

01

Proprietary lock-in

Existing BMS platforms trap organisations in vendor ecosystems. Switching costs are enormous. Each system speaks its own dialect.

02

Expert dependency

Advanced diagnostics and energy optimisation require expensive consultants. Institutional knowledge walks out the door.

03

Data trapped in silos

Field data sits in isolated historian systems. Cross-building correlation and pattern recognition are practically impossible.

04

No room for AI

Cloud-based AI means sensitive operational data leaves your network. On-premise alternatives have required hardware investments only large enterprises could justify — until now.

05

Structured procurement requirements

Public sector organisations operate within the Public Procurement Act, requiring solutions that are transparent, auditable, and vendor-neutral.

How it Works

Four layers.
One coherent system.

LAYER 01 — FIELD
Your existing equipment
openAut never replaces what's already there. Sensors, meters, chillers, AHUs, PLCs and controllers continue running exactly as before — fully in control of their own processes.
BACnet IP
Modbus RTU / TCP
M-Bus (energy meters)
KNX / DALI
Read-only access only
LAYER 02 — EDGE
Intelligent edge nodes
Ruggedised industrial edge computers sit quietly at the edge of your OT network. They translate every field protocol into a unified data model, buffer locally, and forward upstream — gracefully degrading if cloud connectivity drops.
Protocol translation
Local 24–72h buffer
OPC UA server
Rule-based alarms
IEC 62443 SL2 capable
LAYER 03 — AI
On-premise intelligence
The NVIDIA DGX Spark sits inside your building — not in the cloud. It runs large language models and specialised fault-detection models entirely locally. Your operational data never leaves the premises.
NVIDIA DGX Spark
Local LLM inference
Fault detection & FDD
Energy optimisation
Anomaly correlation
LAYER 04 — INTERFACE
Operator-first interfaces
Insights reach the people who need them — in tools they already use. Chat interfaces for field technicians. Web dashboards for energy coordinators. APIs for integration teams. All advisory. All human-in-the-loop.
Microsoft Teams / Slack
Web HMI & dashboards
Alarm management
Auto-documentation
REST API
openaut — field technician session
openaut> status --building "City Hall Block A"
▸ 312 data points online · 3 active anomalies · last sync 4s ago
 
openaut> diagnose --system AHU-03
Analysing 72h trend data for AHU-03...
✔ Supply air temperature deviation detected since 06:14
Root cause (87% confidence): heating coil valve hysteresis
Correlated: return temperature spike, increased SFP, Zone B CO₂ rise
Recommended action: inspect actuator feedback signal on CV-103
 
openaut> energy --report weekly --format pdf
✔ Report generated: /reports/2025-W22-city-hall-a.pdf
# All analysis is advisory — no write-back to field equipment
The AI Layer

Supercomputer intelligence.
Inside your building.

At the heart of openAut sits the NVIDIA DGX Spark — a GB10 Grace Blackwell Superchip system capable of running models with up to 200 billion parameters, entirely on-premise. This is the same generation of hardware powering frontier AI research, now deployable in a building's server room.

With 128 GB of unified memory and up to 1 petaFLOP of AI compute at FP4 precision, the DGX Spark can simultaneously run large language models for operator interaction, specialised fault-detection models, energy optimisation algorithms, and anomaly correlation engines — all without a single byte of operational data leaving your network.

Two linked units extend the capability to 405-billion parameter models, covering the most demanding building portfolios. Air-gapped deployments are fully supported — no internet connection required, ever.

Local inference
No cloud dependency
ARM64 architecture
DGX OS (Ubuntu 24.04 LTS)
Air-gap ready
Data sovereignty
128GB
Unified Memory
1 PFLOP
FP4 AI Compute
200B+
Parameter Models
0
External API Calls
DGX
Spark
GB10
Open Source

MIT licensed.
No strings.
No exceptions.

Ownership

You own your deployment

MIT is the most permissive licence in open source. Use openAut commercially, modify it freely, integrate it into existing systems, and redistribute it — with or without changes — with no royalties, no notifications, and no permission required.

🔬
Transparency

Every line is auditable

All configuration, all application profiles, all AI prompt logic — published openly. Security auditors, technical managers and your own team can inspect exactly what the system does with your data. No black boxes, no hidden logic.

🌱
Community

Built on contribution

Each organisation that deploys openAut can contribute its application profiles, protocol drivers, and FDD models back to the project. The ecosystem improves with every installation — compounding value for every participant.

🔌
Architecture

No proprietary core

Every dependency in the openAut stack is open source: pymodbus, BAC0, opcua-asyncio, open62541, Mosquitto, EMQX. If you already hold licences for proprietary drivers like Kepware, those plug in — but they are never required.

🔄
Repeatability

Any technician can reproduce it

Each application profile is fully self-contained: hardware selection, protocol configuration, OPC UA namespaces, AI model parameters, FAT/SAT checklists. A technician who has never seen your building can commission openAut from the documentation alone.

🛡
Security

Secure by design

IEC 62443 and NIS2 are designed in from day one — not bolted on later. VLAN segmentation, OPC UA Basic256Sha256, WireGuard VPN, RBAC and audit logging are baseline requirements, not optional extras.

Public Procurement

Built for public sector.
compatible
by design.

For municipalities, regions and public property owners, the Public Procurement Act creates real constraints on technology adoption. openAut was designed from the outset to work within those constraints, not around them.

Because openAut is MIT-licensed and open source, there is no single vendor to procure. The platform itself is freely available. What your organisation procures are the professional services to implement, configure and maintain it — services that can be put out to open tender and awarded to your preferred regional consultants, existing framework agreements, or your own in-house team.

This means openAut is compatible with the full lifecycle of public procurement: open specification, competitive tendering, independent delivery, and ongoing contribution back to the shared platform.

Adopt the open platform. Procure the expertise. Own the result. Contribute back to the commons.

01

Adopt the open standard

Your organisation decides to base its building intelligence strategy on openAut — freely, without signing any vendor agreement.

02

Specify in open tender

Implementation requirements are specified against the openly published openAut architecture. Any qualified integrator can respond.

03

Deliver with chosen consultants

Your procured integrators implement, configure and commission the system using the full openAut documentation. All deliverables are reproducible by any other party.

04

Own the configuration

Application profiles, I/O lists, AI configurations and commissioning records belong to your organisation — not the consultant. Transition between integrators at any time.

05

Contribute back

Improvements, new protocol drivers, or application profiles developed during your project can be contributed back to the shared openAut repository. Your investment strengthens the ecosystem for every other public sector organisation.

Core Principles

What openAut
will never do.

P.01

Write back to field equipment

openAut is read-only against all field devices. Regulation and control remain entirely with field PLCs and DDC controllers. The AI advises. Humans and field controllers decide.

P.02

Send your data to the cloud

All inference runs on the on-premise DGX Spark. No operational data — temperatures, occupancy, energy readings — is transmitted to external APIs or cloud services unless you explicitly configure it.

P.03

Require proprietary dependencies

The core stack runs entirely on open source libraries. No proprietary drivers, licences or SDKs are required to deploy or operate openAut.

P.04

Replace your existing BMS

openAut coexists alongside Siemens Desigo, Schneider EcoStruxure, Trend, Regin and every other platform. It reads their data. It learns their patterns. It never competes with or disrupts them.

P.05

Create a new vendor dependency

Every design decision is evaluated against the question: does this create lock-in? If the answer is yes, openAut finds another path. The architecture is modular, documented, and walkable by any competent integrator.

Get started

Your building is
already talking.
Start listening.

openAut is open source, actively developed, and ready for pilot deployments. Explore the documentation, contribute a protocol driver, or adapt an application profile for your building portfolio.

View on GitHub Explore the architecture