DOMECOMMAND · DVL · SEPTEMBER 2026

An AI-native operating system
for edge defence

Fuse. Reason. Gate.

Command and control that fuses sensor data, supports decisions and coordinates assets at the site, under central command.

Counter-UAS and drone control
Deployed at the edge
Super OODA loops
Dual use, military and civil

VIVEK GUDAPURI

DOMECOMMAND.AI

ACQUIRED BY INTURAI · CSE: URAI
THE PROBLEM
DOMECOMMAND

RUSSIA · 20 SEPTEMBER 2026

1,000+ drones
in one night

A week ago, Ukraine launched over a thousand drones at Russia, including the largest attack ever on Moscow. Every recent war is now a drone war.

AsiaOne, 20 September 2026: Ukraine fires over 1,000 drones at Russia including largest attack on Moscow
ASIAONE, 20 SEPTEMBER 2026 · RUSSIAN DEFENCE MINISTRY: 1,110 DRONES SHOT DOWN, 450 ON THE WAY TO MOSCOW
AI PROGRESS
DOMECOMMAND

AI IS ONLY GETTING STRONGER

AI is getting smarter, and the curve is exponential

PilotedOne operator flies one drone
UnmannedFlies itself between waypoints
CoordinatedMany drones share one plan
Fully autonomousReasons about what it sees, and chooses

THE SAME SHIFT ON THE ROAD

AUTOPILOT
Tesla took 10+ years
→
NVIDIA ALPAMAYO
It just thinks

“By thinking, by reasoning, you don't need as much data.”

Jensen Huang, NVIDIA, on reasoning self-driving models, 2026

TIME AI CAPABILITY
ILLUSTRATIVE CURVE
AI-enabled FPV kamikaze drone
GA-10FPV-AI KAMIKAZE DRONE, 3 KG CHARGE
PHOTO: LITHUANIAN ARMED FORCES, GFDL

WHY NOW

cheap drones,
now with a brain

$200
a drone that thinks for itself
$1M+
for each missile sent to stop it

Even with a layered defence, answering every drone with a missile is prohibitively expensive.

“We cannot use a missile to take down every drone that comes at us.”

Chan Chun Sing, Minister for Defence, Committee of Supply, 27 February 2026

MISSILE COST: INTERCEPTOR MISSILES RUN FROM ABOUT $1M TO $4M EACH
PRODUCT
DOMECOMMAND

WHAT IT IS

An AI-native operating system for edge air defence

SIMULATION · SINGAPORE HARBOUR · live console, sped up 1.5×
AI native
Fully AI compatible, for AI running inside and outside DomeCommand
Sensor agnostic
One way to connect any sensor
Configurable and modular
Every part over APIs
Small footprint
Runs on a $200 board
Simulation first
Plan, train and improve

“Drone defense is looking less and less like operating a weapon and more like running a real-time distributed system.”

Tyler Bosmeny, Y Combinator

DESIGN PRINCIPLES FROM THE DVL FINAL REPORT
PRODUCT · ARCHITECTURE
DOMECOMMAND

ARCHITECTURE

system boundaries and feedback

DOMECOMMAND SITE DEPLOYMENT

EXTERNAL LINKAGES

SENSORS
Radar, EO/IR turrets, RF and SDR receivers, acoustic arrays, Remote ID and ADS-B receivers.
AIRCRAFT & EFFECTORS
MAVLink autopilots, DJI docks, interceptors, jammers. Commanded back through the same layer, under gates.
CENTRE & CONTEXT
Maps, zones and operator reports. Central C2 exchange is an integration path: tasking in, observations and outcomes back.
⇄TRACKS IN · COMMANDS OUT

INTEGRATIONS LAYER

PROTOCOL ADAPTERS
One adapter per protocol. Everything normalises to observations, detections and taskable assets. No proprietary bus.
MAVLINK MQTT DJI CLOUD SAPIENT ASTERIX MISB 0601 ADS-B REMOTE ID INTURAI SIM
⇄OBSERVATIONS IN · RELEASED COMMANDS OUT

DETERMINISTIC CORE

World model
Tracks, assets, zones, magazines, one time-aware store.
Fusion & tracker
Kalman-filtered association, confidence, CPA and TTI.
Solver
Optimisation over assets and threats. Plans in parallel.
Framing & rules
Facts become alerts and priorities on each tick.
Action gate
Checks authority and capability before release.
Runs & replay
Every run recorded, replayable, defensible.
FEEDBACK Platform telemetry updates the picture. Changed conditions trigger the next plan.
⇄PICTURE OUT · ACCEPTED INPUTS BACK

AGENTS & CONSOLE

DISTRI
Agents read the picture outside the deterministic core.
PROPOSE
Assess changes and draft rules, scenarios or plan adjustments.
REVIEW
Accepted inputs return to planning. Actions still cross the gate.
AUDIT & REPLAY
Decisions, approvals, refusals and outcomes remain recorded.
⇄PROPOSE · REVIEW

COMMAND

OPERATOR
Sets permissions. Approves effects. Owns every consequential action.
SIMULATION ENTERS THROUGH THE SAME INTEGRATIONS: DOME LITE · GAZEBO + PX4 · UNREAL
PRODUCT · SENSOR FUSION
DOMECOMMAND

SENSOR FUSION

every sensor is blind somewhere. Together they see.

RF onlyσ 126 m
→
+ drone cameraσ 89 m
→
+ site cameraσ 60 m
→
+ acousticσ 26 m
One bearing is a line, not a point. Each sensor from a new angle shrinks the uncertainty, and a track is never marked hostile from movement alone.
DOMECOMMAND TRACKER RECORD · CHANGI, DOME LITE SIMULATION · 95% ELLIPSES
PRODUCT
DOMECOMMAND

DECISION SUPPORT

multi-brain decision support, with an auditable, replayable event log

Solver DETERMINISTIC
Assigns effectors to threats under real limits: magazines, reach, geography, rules of engagement. Scored, ranked and repeatable.
Rules FACTS · WHEN · THEN
One-sentence rules on tracks, zones and assets decide what may be proposed, and keep every decision attributable.
Cognition AI · ON OR OFF
Turn cognition on or off for AI-assisted suggestions. Models from any vendor explain changes and draft plans. Proposals go to the operator.
Decisions queue. Plans wait for the operator: approve, reject or edit.
Event log. Every decision accounted for and auditable, with replay.
CAPTURED FROM THE DOMECOMMAND CONSOLE · SINGAPORE HARBOUR, DOME LITE SIMULATION
PRODUCT · AI NATIVE
DOMECOMMAND

AI NATIVE

command in plain language

SIMULATION · SINGAPORE HARBOUR · @BLUE-01 intercept @H1, drafted by DOME AI, released by the operator

Tag a vehicle, a contact or a zone with @. DOME AI turns the sentence into an order, and the operator confirms it.

FOR EXAMPLE

›@BLUE-01 take off to 40 m
›@BLUE-01 intercept @H1
›send the nearest aircraft after @H2
›@BLUE-02 watch the @north-fence
›mark @H3 hostile
›everyone back to base
FROM THE DOMECOMMAND DEMO FILM · COMMANDS FROM THE OPERATOR COMMANDS GUIDE
SIMULATION
DOMECOMMAND

BE PREPARED

simulate and train

  • Plan where every sensor and drone goes
  • Train operators on repeatable scenarios
  • Replay every run, it is all recorded
  • Improve with runs at scale over APIs

One scenario runs in three engines, through the same console as a real site.

SINGAPORE, IN SIMULATION

SIMULATION
Changi. Three kinds of sensor, one fused track.
SIMULATION
The harbour. Plan approved, two intercepts.
SIMULATION
The city centre. The operator approves the plan.
SIMULATION
50 v 50. Every engagement assigned.
UNREAL / AIRSIM · CAMERA IMAGERY
GAZEBO + PX4 · REAL AUTOPILOTS
DOME LITE · FAST RUNS, REPLAY
CAPTURED FROM THE DOMECOMMAND CONSOLE, SEPTEMBER 2026
SIMULATION · END TO END
DOMECOMMAND

END TO END

three defenders, three raiders, fully automated

SIMULATION · SINGAPORE CBD, GAZEBO + PX4 · FROM THE DOMECOMMAND DEMO FILM, 3× SPEED, NO SOUND
HOW IT WORKS
DOMECOMMAND

HOW IT WORKS

a super OODA loop

  • Every track in parallel, from a wide variety of sensors and a blend of algorithms
  • Mission command: centralized command, distributed control, decentralized execution
  • Built to connect to your existing C2, over open standards such as Cursor on Target
CENTRAL COMMAND

EACH SITE RUNS ITS OWN LOOP AT THE EDGE, ON ITS OWN BOARD

EVERY TRACK IN PARALLEL MANY LOOPS · ONE AUTHORITY
ObserveFUSION
Radar, EO/IR, RF, acoustic and Remote ID, fused by a blend of algorithms: clustering, Kalman filters, bearing fusion.
OrientFUSION + RULES
Every track classified and identified on every update. Agents add assessment on top.
DecideDECISION SUPPORT
Solvers author ranked plans, scored on leakers, cost and coverage.
ActPERMITTED
Commands go out up to the level the commander sets: six levels, Observe to Destroy.
SUPER OODA LOOP: MICHAEL RASKA, RSIS, 2025 · MISSION COMMAND: USAF DOCTRINE PUBLICATION 1-1 · C2: CURSOR ON TARGET / TAK ADAPTER BUILT, LIVE TWO-WAY LINK NEXT
PRODUCT
DOMECOMMAND

DOMECOMMAND

modular, customisable and nimble

Sensor agnostic

Radar, RF, EO/IR, acoustic, Remote ID, ADS-B and Inturai sensing, fused into one picture. Use the sensors a site already has.

Connects to most protocols

One common way to connect sensors and command vehicles. A new integration takes days, not months.

MAVLINK · MQTT · DJI · INTURAI · SAPIENT · ASTERIX · MISB 0601 · ADS-B · REMOTE ID

Customisable

Sites, zones, rules, solver weights and planning profiles, all over APIs. Algorithms extendable at every level.

DEPLOYS IN MANY WAYS · THE SAME SOFTWARE EVERYWHERE

Orange Pi 5, about $200. A gate or a compact site. 50 v 50 in real time.
NVIDIA Jetson. Mid-size sites that need local AI.

On a drone

The same software as companion computing, for autonomy on the aircraft.

Your own cloud

Private cloud or site servers, in the customer's own Azure or AWS tenancy. Data stays with the customer.

BUILT IN RUST · EXAMPLE HARDWARE, SIZE TO WORKLOAD
USE CASES
DOMECOMMAND

USE CASES

one system, military and civil

USE CASE 01 · COUNTER-UAS

Centrally coordinated, locally executed

Headquarters sets priorities. Each site responds to local conditions within delegated authority.

LOCAL EXECUTIONTWO-WAY INFORMATION

USE CASE 02 · CRITICAL INFRASTRUCTURE

Airports, energy, ports

Runs on the sensors the site already has, with effects set for civil law.

PASSIVE RFREMOTE IDEO/IR

USE CASE 03 · MULTI-DOMAIN C2

Departments on one picture

Air defence, police, airport operations and emergency services each get their own channel on one fused picture.

CHANNEL PER DEPARTMENTAUDIT TRAIL
FROM THE PRODUCT DECK · ALSO: HUMAN SENSING WITH DOME SENSORS, POWERED BY INTURAI
STATUS
DOMECOMMAND

SEPTEMBER 2026

current project status

PRODUCT

  • Advanced simulation runs on multiple engines
  • Several integrations are already built, giving a sensor-agnostic picture
  • A single box can defend an edge location

COMPANY

  • Acquired by Inturai, a company listed in Canada (CSE: URAI)
  • Commercial discussions under way in several countries, for civil use cases

DOMECOMMAND

The future is about harnessing AI.
DomeCommand makes it possible.

Fuse. Reason. Gate.

Thank you.

VIVEK GUDAPURI | DOMECOMMAND.AI

DOMECOMMAND.AI · INTURAI.COM
APPENDIX · A1
DOMECOMMAND

APPENDIX · A1

the team

Vivek Gudapuri

  • Large-scale route and fleet optimisation for freight forwarders, across countries, containers and trucks: the basis of the solver
  • Built an AI gateway, and Distri, the agent runtime DomeCommand's AI layer runs on
  • CTO at several companies, on several founding teams
  • Raised capital from Google and Sequoia

Sashank Donkena

  • Ten years of embedded and platform engineering
  • Cisco, then Microsoft
  • Builds agentic experiences for Windows
DISTRI IS VIVEK'S AND IS LICENSED TO INTURAI, NOT SOLD
APPENDIX A1
APPENDIX · A2
DOMECOMMAND

APPENDIX · A2

permissions: the commander sets the line

LevelExample actions
0 · ObserveAlert, cue a sensor
1 · ManoeuvreTake off, move to, return to base
2 · WatchSurveil
3 · ShadowFollow, warn, mark suspect
4 · DenyJam, spoof
5 · DestroyIntercept, mark hostile
  • Every sensor and vehicle declares its actions, and each action has a level
  • The operator sets how far actions run without asking. Above that line, actions wait for approval
  • A deployment sets a ceiling: what exists at that site at all
  • Shadow, jam and intercept ship switched off. A site enables each one separately
AUTOMATE WHAT YOU WANT, KEEP HUMAN CONTROL OVER WHAT MATTERS
APPENDIX A2