Physical and digital research infrastructure powering sovereign defence AI development. Edge compute nodes, additive manufacturing, acoustic sensing, ISR test range. Built in the UK. Sovereign by design.
R&D and testing space with 33 buildings, installations, and sensor test areas
Acoustic/environmental test bed β water flow, ambient audio, environmental monitoring
Additive manufacturing for drone parts, sensor housings, and edge node enclosures
Apple Silicon inference mesh β M2 (right brain perception), M3 (left brain reasoning), M4 (cognition)
The DEFONEOS R&D programme operates from a 19,000 sqft facility in Yorkshire, UK. The site hosts the physical sensor infrastructure, edge compute nodes, additive manufacturing, and ISR testing capabilities that validate the digital DEFONEOS platform in the real world.
Command centre, server room, development workstations. Houses the primary SOV3 substrate node and the GCP VM bridge for live inference.
Acoustic and environmental sensor test bed. Water flow sensors, hydrophones, temperature probes, and ambient audio capture β simulating maritime ISR scenarios at micro-scale.
Additive manufacturing bay. 4Γ Qidi Max4 3D printers for rapid prototyping of drone airframes, sensor housings, and edge compute enclosures. Tab 6 in the DEFONEOS workspace.
Physical installations mapped to the 22 Hebrew letters / Major Arcana β each housing a different sensor or compute node, creating a distributed physical sensing grid across the property.
| Principle | Implementation | Status |
|---|---|---|
| Air-gap capable | All critical compute can operate disconnected from the internet. Local inference via Ollama + SOV3 edge nodes. | β Operational |
| No cloud dependency | No AWS, Azure, or foreign cloud for sensitive data. GCP VM used only for non-sensitive orchestration. All defence data stays on sovereign hardware. | β Enforced |
| PQC ready | ML-DSA-65 (Dilithium) signatures and ML-KEM-768 (Kyber) key exchange. Post-quantum cryptography standard. | β Deployed |
| Ed25519 SIGIL chain | Every action on the infrastructure is hash-chained and Ed25519-signed. Immutable audit trail. | β Live |
| UK data residency | All data processing occurs on UK soil. No data leaves the jurisdiction. ICO-compliant. | β Enforced |
| Renewable power | Solar-assisted power for edge nodes. Battery backup for 72-hour autonomous operation. | π‘ In progress |
DEFONEOS operates a heterogeneous compute fleet spanning Apple Silicon inference mesh, NVIDIA Jetson edge nodes, and Raspberry Pi sensor hubs. All nodes run the SOV3 sovereign substrate and report to the central BFT council.
Three MacBooks form the sovereign inference mesh: M2 handles right-brain perception (vision, audio, sensors via Ollama), M3 handles left-brain reasoning (MoE, BFT council), M4 handles cognition synthesis (SOV3 middle layer).
Edge AI nodes for field deployment. Run YOLOv8 object detection, OpenAthena geospatial processing, and local Mamba-2 state compression. 5-15W power consumption. Air-gap capable.
Low-power sensor aggregation nodes. Collect data from RTSP cameras, MQTT IoT devices, RTL-SDR radio, and environmental sensors. 5W minimum power. Can run for days on battery.
The only cloud component β used strictly for non-sensitive orchestration. Runs the King hive, 33 hives, SOV3 :3101 MCP endpoint, council :3200 BFT endpoint, and the OLM router. All sensitive data stays on sovereign hardware.
| Node Type | Count | Role | Power | Air-Gap |
|---|---|---|---|---|
| Apple Silicon (M2/M3/M4) | 3 | Inference mesh (perception, reasoning, cognition) | 30-60W each | β |
| NVIDIA Jetson | 2+ | Edge AI (YOLOv8, OpenAthena, SSM) | 7-15W | β |
| Raspberry Pi 5 | 3+ | Sensor aggregation, IoT bridge | 5W | β |
| GCP VM | 1 | Orchestration (non-sensitive only) | Cloud | β |
| Total edge TOPS | ~130 TOPS across sovereign hardware (excl. cloud) | |||
The FORGE Lab (Tab 6 in the DEFONEOS workspace) houses 4Γ Qidi Max4 3D printers for rapid prototyping. We manufacture drone airframes, sensor housings, edge compute enclosures, and mounting hardware in-house β reducing supply chain dependency for defence-grade prototypes.
CoreXY FDM printers with enclosed heated chambers
High-velocity prototyping for iterative design
Large enough for drone airframes and sensor arrays
Prints PETG, ABS, ASA, TPU, Nylon-CF
| Material | Use Case | Properties |
|---|---|---|
| PETG | Sensor housings, mounting brackets | Impact-resistant, chemical-resistant, UV-stable |
| ASA | Outdoor drone components, weather stations | UV-resistant, outdoor-rated, temperature-stable |
| Nylon-CF (Carbon Fibre) | Drone airframes, structural components | High strength-to-weight, stiff, lightweight |
| TPU | Vibration dampeners, gaskets, impact protection | Flexible, durable, impact-absorbing |
| ABS | Enclosures, prototype cases | Impact-resistant, machinable, paintable |
Custom quadcopter frames for PX4 flight controllers. Designed for ISR payload mounting (cameras, RTL-SDR, acoustic sensors). Carbon-fibre reinforced nylon for strength-to-weight ratio.
Weatherproof enclosures for Raspberry Pi + Jetson edge nodes. IP65-rated outdoor deployment. Integrated thermal management for passive cooling in field conditions.
Custom mounts for RTL-SDR antennas, AIS receivers, and ADS-B antennas. Designed for mast deployment and vehicle mounting.
The 19,000 sqft facility serves as a living sensor test range. Every DEFONEOS MCP server category has a physical counterpart being tested on-site.
| Sensor Category | MCP Server | Physical Test Setup | Status |
|---|---|---|---|
| Satellite Imagery | sentinel-hub-mcp | Sentinel-2 imagery pipeline β Cesium globe visualisation | β Live |
| Maritime AIS | aisstream-maritime-mcp | AIS receiver β vessel tracking from Humber/port approaches | β Live |
| Aviation ADS-B | (planned) | RTL-SDR ADS-B receiver β aircraft tracking within 200nm radius | π‘ Hardware ready |
| OSINT News | gdelt-news-mcp | GDELT real-time global news monitoring pipeline | β Live |
| Air Quality | openaq-air-mcp | OpenAQ + local PM2.5/COβ sensors in environmental monitoring grid | β Live |
| Government Data | data-gov-uk-mcp, ons-statistics-mcp, companies-house-mcp | Live UK government open data feeds | β Live |
| IoT | mqtt-bridge-mcp | MQTT broker bridging local sensor network to DEFONEOS pipeline | β Live |
| IP Cameras | rtsp-camera-mcp | 4Γ RTSP cameras (flock-cam-1..4) β wildlife/perimeter monitoring with PII redaction | β Live |
| Ordnance Survey | os-opendata-mcp | OS OpenData integration β UK mapping layers for Cesium | β Live |
| Radio Frequency | (RTL-SDR) | RTL-SDR dongles β FM/ADS-B/NOAA weather satellite reception | π‘ Hardware ready |
| Acoustic | (batear C-UAS) | 13m koi pond hydrophone + ambient audio capture β drone acoustic detection R&D | π‘ R&D |
22 physical installations across the property, each mapped to a Hebrew letter and housing a different sensor or compute node. This creates a distributed sensing grid that validates the DEFONEOS MCP federation in the physical world.
Reducing the kill-chain latency from sensor detection to council-authorized action. Current target: sub-40 seconds using Mamba-2 SSM state compression for real-time signal correlation across 198+ sources.
Multi-agent reinforcement learning for drone swarms. Using Mava MARL framework with PX4 SITL for simulation. Goal: demonstrate coordinated ISR swarm behaviour β distributed search patterns, adaptive formation flying, and autonomous threat response.
$10 acoustic drone detection system. Using MEMS microphone arrays + ML classification to detect and classify drone acoustic signatures. Tested against the koi pond environmental audio baseline. Low-cost, deployable at scale.
Building a Cesium 3D digital twin of Yorkshire with real-time data layers: satellite imagery, AIS vessel tracking, ADS-B aircraft, OSINT events, air quality, and government data. The first DEMO video target.
Building the UK's first automated JSP 936 (AI in Defence) compliance generator. Maps DEFONEOS architecture to JSP 936 controls and produces audit-ready compliance documentation. The compliance differentiator vs Palantir.
Migrating all SIGIL chain signatures from Ed25519 to NIST PQC standards: ML-DSA-65 (Dilithium) for signatures, ML-KEM-768 (Kyber) for key exchange. Preparing for the quantum computing threat to classical cryptography.