Skip to main content

EdgeEmbed · Physical AI

Models predict. EdgeEmbed decides what machines are allowed to do.

EdgeEmbed is the AI-to-control bridge for physical AI — the deterministic layer between your AI model's intent and the controller that moves the machine. Every intent is gated against authorable safety policy, executed with real-time discipline, and recorded for bit-exact replay. The model can be uncertain. The gate is deterministic. The recorder remembers everything.

Explore the RuntimeExplore HIL
How it fits togetherDecide on the machine · Prove on the bench
On the machine — in production
AI model intentEdgeEmbed RuntimeThe gate decides. The recorder proves.PLC · controller · actuators
same deterministic behavior — proven before it ships
On the bench — in development
AI agent · CI · engineerEdgeEmbed HILHands and eyes on hardware — open sourceSimulator · real board

The bridge

Two worlds that don't speak the same language

The AI world speaks in probabilities — model intents, confidence scores, per-frame guesses. The control world demands determinism — a machine either stops or it doesn't, and someone must be able to prove why. Today's safety controllers decide deterministically, but only from hard-wired, classical inputs; none of them accept a model's confidence-weighted intent as an input class. That intake is the bridge EdgeEmbed builds: probabilistic AI intents in, policy-governed deterministic decisions out, with a record of why. EdgeEmbed does not replace AI stacks, PLCs, safety controllers, or robot controllers. It connects them.

What EdgeEmbed is

  • The AI-to-control bridge — the deterministic layer between the model and the machine
  • A Model Gate: every AI intent is checked against authorable safety policy before it can act
  • A flight recorder: every decision journaled for bit-exact replay and evidence
  • Driven by declarative config — safety behavior as data, on an ordinary core, no NPU

What EdgeEmbed does not replace — it connects

  • Not a model team or an inference runtime — model outputs enter as events; we never run inference
  • Not a PLC and not a safety PLC — vetted decisions leave toward the controllers you already trust
  • Not a robot controller or a motion planner
  • Not a certified safety system today — evidence-ready by design; the certification is yours, we co-develop the evidence

Products

Two products, one bridge

The runtime decides on the machine; HIL proves the behavior on real hardware. Each stands on its own — together they close the loop from AI intent to validated machine action.

EdgeEmbed Runtime

The gate decides. The recorder proves.

The deterministic engine at the heart of the bridge: AI intents and machine events in, one policy-governed, control-ready action out — every decision recorded and replayable.

Explore the Runtime

EdgeEmbed HIL · Open source

Hands and eyes for AI agents on hardware

The open hardware-in-the-loop protocol: discover a board, drive its pins, watch its logs, run bounded commands — with a local simulator so the first ten minutes need no hardware.

Explore HIL

Solutions

One runtime product. Three deployment solutions.

Every solution is the same engine, deployed differently and meeting each ecosystem at its own interface — a PLC, a ROS graph, a fleet stack. What changes is the bundle and the boundary. What never changes is the gate and the record.

Machine Safety · on the machine

AI perception in. Machine safety out.

The runtime runs on the target itself. Camera and acoustic AI become PLC-ready safe-stop decisions for an industrial cell — gate, slow or stop, alert, and record.

Explore Machine Safety

ROS 2 · beside the robot stack

A node for distribution, never for architecture.

A standard ROS 2 package feeds topics in as events — while the decision itself runs off the ROS executor path, where its timing guarantee still holds.

Explore ROS 2

Fleet Operations · on the edge

Every order a robot receives was approved first.

The runtime runs on a PC or edge server beside the fleet stack you already operate, admitting or rejecting every transport task before a vehicle moves. It speaks VDA 5050 — it does not replace your fleet manager.

Explore Fleet Operations

Get in touch

Talk to us about your machine

Our mission is a trustworthy Decide stage in every machine that runs AI. Putting a model on a real machine, or bridging AI outputs into a PLC or robot controller? We would love to hear from you.

EdgeEmbed teamOnline — typically replies within a day
Hi — this is the EdgeEmbed team. Tell us about your machine, the AI you are putting on it, and what you need the bridge to do.