Physical AI and Sovereign AI Systems

Build at frontier speed. Operate inside your boundary.

Open Engineering

Built in the open. Designed for the real world.

Models, research, reference architectures, and engineering from the teams building sovereign and physical AI — spanning tool-capable small models, robotics, Vision-Language-Action systems, local inference, and GPU optimization.

Open Model Lab · Research & Engineering

Hugging Face GitHub

Agentic / Tool Use

smol-tools-4b-32k

Compact tool-capable model work designed for local and sovereign agentic applications.

Tool UseSmall ModelLocal AI32K Context

Text generation1.8k downloads

Sovereign Inference

Qwen2.5-72B-Instruct-NVFP4

NVFP4 quantization of an upstream open model, tuned for capable inference on private GPU infrastructure.

NVFP4GPUPrivate AIInference

Text generation793 downloads

Physical AI

edgevla-base-fmb

Vision-Language-Action research for robotics and intelligent systems operating at the edge.

RoboticsVLAPhysical AIEdge

Robotics

Physical AI

edgevla-small-fmb

Compact Vision-Language-Action model work intended for constrained physical-AI environments.

VLARoboticsEdge AI

Robotics

GPU Optimization

Mixtral-8x22B-Instruct-v0.1-NVFP4

Quantized build of an upstream mixture-of-experts model, exploring efficient large-model deployment on modern GPUs.

NVFP4InferenceGPUSovereign AI

Text generation

Edge NLU

distilbert-joint-intent-slot-smarthome

Joint intent and slot understanding sized for on-device voice and control workloads.

Edge AINLUOn-Device

Token classification

Research

ALA — Asynchronous LLM Advisor

enfuse/ala-paper

Research exploring LLM-guided reinforcement learning through bounded logit perturbation channels.

ResearchLLMReinforcement Learning

View on GitHub

Applied AI

Document Vision

enfuse/groq_doc_vision

Multimodal document-processing examples, SDK, and CLI for accelerated AI inference.

Computer VisionMultimodalInference

View on GitHub

Developer Tooling

Backstage LLM Plugin

enfuse/backstage-chatgpt-plugin

LLM assistance embedded directly into internal developer platforms and engineering workflows.

Developer PlatformLLMInternal Tooling

View on GitHub

Frontier speed, sovereign execution

Physical AI and enterprise agentic systems

We engineer AI systems that connect real-world signals, enterprise knowledge, and business workflows — perception, multi-agent orchestration, secure data integration, AI infrastructure, and production operations — deployed across Microsoft Azure, Google Cloud, private infrastructure, on-premises environments, and the edge. From an individual workflow to a customer-controlled AI platform, we build what it takes to move beyond a demonstration.

Plan with the frontier

Frontier models act as planner agents — decomposing work, designing and orchestrating multi-agent systems — and you can swap the model without rebuilding the app.

Govern the boundary

A sovereignty contract over data, identity, tools and agent-to-agent handoffs: every agent action is policy-checked, traced and audited in the runtime, not bolted on.

Execute inside your walls

Executor agents run on private cloud, on-premises, edge and air-gapped hardware — GPU, Jetson and sensors — so multi-agent workflows act locally and no operational data leaves.

What we do — four areas of expertise