didymOS

// why physics-first

Most “digital twins” are either renders or black boxes.

A 3D model with no physics can't predict. A pure-ML model can't extrapolate and can't explain itself. didymOS runs the asset's real equations and lets a strictly bounded ML residual absorb what the physics misses — so every prediction is explainable, and every divergence is evidence.

CAD render “twin”

No governing equations
No live telemetry
Can't forecast or warn
Demo-ware, not a runtime

Pure-ML twin

Blind outside training data
Unexplainable predictions
Silently drifts with sensors
Needs months of history

Physics kernel per asset class
ML residual, hard-bounded (±3 K)
Residual saturation ⇒ anomaly alarm
Runs from day one, no history needed

// features

A twin runtime, not a picture.

⚛️

Physics kernel

Pluggable per asset class — IEC 60076-7 thermal model for transformers, pump & motor models next. Difference equations, unit-tested.

🎯

Bounded ML residual

A small learner corrects the physics from live data — clamped to a hard envelope. When it saturates, that's not noise. That's a fault signature.

📡

MQTT-native · `dimorph` agent

The edge agent ingests telemetry over MQTT and republishes twin state — from a €35 Raspberry Pi to industrial gateways. No cloud dependency, ever.

🔮

What-if API

GET /twin/{id}/whatif?load=130 — raise the load 30% and didymOS answers with the hot-spot trajectory before you touch the real asset.

🥽

VR & AR native

The twin isn't trapped in a dashboard: walk around it on Meta Quest, or pin it onto the real machine through your phone's camera. Step inside ↓

🔗

Audit ledger

Every prediction and recommendation hash-chained to an append-only log — an AI-Act-shaped audit trail, out of the box.

// in the field

Not slideware. Mirrors of real machines.

didymOS earned its scars in the field before it had a landing page — the physics-plus-residual recipe, the MQTT edge agents and the failure-prediction line were all run against real machines first, then productised.

PILOT · GR

⚡

Energy

Distribution transformer fleet

Oil-immersed transformers mirrored by dimorph edge agents on Raspberry Pi. IEC 60076-7 kernel, live hot-spot tracking, 8-hour forecasts on the operator's morning screen.

4 assets · 8 h forecast horizon · divergence alarms on residual saturation

PILOT · FACTORY FLOOR

🏭

Manufacturing

Production-line twin

Cycle-time physics with an ML residual absorbing drift — wear, operator variance, seasonal air pressure. The what-if API answers takt-time questions before anyone touches the real line.

Line-level twin · what-if before changeover · drift caught as residual, not surprise

PILOT · OFFSHORE

🌊

Maritime

Offshore substation monitoring

Fault anticipation for equipment you can only reach by boat: didymOS's identification layer was proven on offshore-substation telemetry first, where a missed warning costs a vessel charter.

Failure prediction · days of warning, not minutes · visits planned, not provoked

// live demo — real equations, in your browser

A transformer twin, running right now.

This is not an animation. The panel below integrates the IEC 60076-7 thermal difference equations (top-oil and hot-spot dynamics) at ~16 simulated minutes per second, with a bounded ML residual learning the plant/model mismatch live. Inject a cooling fault and watch the residual saturate — divergence is detection.

twin://transformer-TX1 · IEC 60076-7 kernel + residual ● NOMINAL

LOAD 80%

AMBIENT 25°C

ML residual (±3 K bound)

Hot-spot (meas.)

— °C

Twin prediction

— °C

Steady @ load

— °C

Residual

— K

Ageing rate

— ×

measured physics only physics + ML residual 8 h forecast 98°C design 118°C alarm

// step inside the twin

The twin isn't on a screen. You're standing in it.

Dashboards flatten machines into charts. didymOS renders the same live twin state at 1:1 scale — in your room through a VR headset, or pinned onto the physical machine through a phone camera. Same API, three realities.

META QUEST 3 · PASSTHROUGH

HOT-SPOT 84.2°CLOAD ▸ 130%+8h ▸ 106°C

🥽 Virtual Reality — walk around the future

The full asset at 1:1 scale in your room. Grab the load slider with your bare hands, push it to 130%, and watch tomorrow's hot-spot bloom across the windings — before it ever happens on the real machine. What-if analysis stops being a query and becomes a place you visit.

hand tracking1:1 scalewhat-if in 3Dmulti-user review

VUZIX SHIELD · BINOCULAR WAVEGUIDES · ANDROID

ANCHORED · TX1RESIDUAL +2.1K● NOMINAL

📱 Augmented Reality — the machine wears its data

Through binocular AR glasses — or any ARCore phone — live gauges, the 8-hour forecast ribbon and divergence alerts anchor onto the physical asset itself. The technician sees the machine's future standing right in front of it, hands free — no control room, no laptop.

anchors on-assetlive overlaysdivergence alertsoffline on-site

// through the waveguides — what you see wearing the glasses · live twin state

twin://TX1 · LIVE—

● NOMINAL

LOAD —AMB —

RESIDUAL —AGEING —×

COMPONENT · —

click a component of the twin
to inspect it · drag to rotate

BINOCULAR FUSED VIEW · SAME LIVE STATE AS THE SIMULATOR ABOVE

↑ click a component — tank, radiators, bushings, conservator — to inspect its live attributes · drag to rotate · change the LOAD in the live demo above and this view follows, live

// get started

Three commands to your first twin.

Install the runtime

Python ≥3.10. One package, no cloud account, no license key.

$ pip install didymos

Point a twin at your asset

Pick an asset class, point at your MQTT broker — the dimorph agent starts mirroring immediately.

$ didymos init --asset transformer --broker mqtt://192.168.1.40 # or: --replay data.csv

Open the viewer — browser, Android or Quest

Live state, 8-hour forecast and the what-if API at :8642/twin/TX1.

$ didymos view # → http://localhost:8642

Most “digital twins” are either renders or black boxes.

CAD render “twin”

Pure-ML twin

didymOS

A twin runtime, not a picture.

Physics kernel

Bounded ML residual

MQTT-native · dimorph agent

What-if API

VR & AR native

Audit ledger

Not slideware. Mirrors of real machines.

Distribution transformer fleet

Production-line twin

Offshore substation monitoring

A transformer twin, running right now.

The twin isn't on a screen. You're standing in it.

🥽 Virtual Reality — walk around the future

📱 Augmented Reality — the machine wears its data

Three commands to your first twin.

Install the runtime

Point a twin at your asset

Open the viewer — browser, Android or Quest

MQTT-native · `dimorph` agent