Reference Architectures
Hardware and software designs for sovereign embodied AI — from edge terminals to full-rack deployments.
These are reference architectures — working designs validated on our own infrastructure. They represent what we're building, not what's sitting in a warehouse. Each specification reflects real hardware running in our lab today.
Praxis Envoy
A self-contained presence terminal designed to bring embodied AI into a single room. Connect to a display, and a photorealistic MetaHuman appears — capable of seeing, hearing, and conversing with persistent memory.
| Specification | Value |
|---|---|
| Processor | NVIDIA Jetson Orin Nano Super |
| AI Performance | 67 TOPS |
| Memory | 8GB LPDDR5 |
| Vision | Intel RealSense D435 (RGB + Depth) |
| Audio | Dual MEMS microphone array |
| Connectivity | WiFi 6, Bluetooth 5.4, HDMI 2.0 |
| Power | 15W typical |
In development. Architecture specification available on request.
Praxis Rack
Full-stack AI inference for embodied presence. Runs large language models, neural voice synthesis, vector memory with temporal decay scoring, and MetaHuman rendering — entirely on local hardware. Our lab prototype runs on 10+ enterprise servers with 200GB+ total VRAM.
| Specification | Value |
|---|---|
| Form factor | 20–24U rack |
| LLM capacity | Up to 70B parameters (quantized) |
| TTS engine | Orpheus 3B neural TTS |
| STT | faster-whisper (real-time, CUDA) |
| Vector memory | Milvus with TDR scoring |
| Networking | 40Gb RoCE internal fabric |
| Sovereign operation | No internet connection required |
Running in our lab. Architecture documentation available on request.
Praxis Sense
An integrated sensory layer for embodied AI — combining depth vision, spatial audio, voice activity detection, and emotion recognition into a unified perception pipeline. Designed to give the MetaHuman awareness of its physical environment without cloud processing.
| Component | Description |
|---|---|
| Vision | Intel RealSense depth + RGB, person detection, gaze tracking |
| Audio | Spatial microphone array, Silero VAD, voice isolation |
| Emotion | wav2vec2-based affect detection from speech |
| Fusion | Multi-modal event stream, real-time to switchboard |
Sensory subsystems operational. Integration ongoing.
Enterprise Presence Routing
For multi-terminal installations: centralized identity, roaming presence across rooms, and fleet management. Built around the Praxis Exchange Gateway (PEG) and Real-Time Arbiter (RTA).
| Function | Description |
|---|---|
| Terminal orchestration | Coordinate multiple Envoy units from a single rack |
| Presence routing | MetaHuman follows the user between rooms |
| Centralized identity | Single user profile across all terminals |
| Fleet management | Remote updates, configuration, diagnostics |
Architecture specification. Not yet deployed beyond the lab.
86” Holographic Light Display
Exploring large-format holographic display integration for life-size MetaHuman presence. Target: a floor-standing 86” panel that renders the MetaHuman at human scale without VR headsets or projection mapping.
Under evaluation. Display hardware assessment expected June 2026.
We explore.
Interested in the architecture?
We're happy to share technical details with researchers, labs, and serious builders.
Contact Us