On your platform · with your models · in your perimeter

The operating system for an unmanned platform.

An onboard layer where your AI models run on every frame at full resolution, and a server layer that holds that across the whole fleet. We supply the layer between your hardware and your models — and nothing else changes.

Airframe, payload, sensors, radios, autopilot: untouched. Your models, your data, your archive, your customer relationship. We do not resell computers, we do not replace your models, and we obtain no access to your data.

Your fleet and your unit

yours

Finished targets in real time — to the crew and into your command system.

Asterioniq · server layer

in your perimeter

Registry of aircraft and model versions · distribution with staged rollout and rollback · event store · retraining.

Asterioniq · onboard layer

on the computer you chose

Decode · inference of your model on every frame · track · sensor fusion · event · recording · secured perimeter.

Your aircraft

unchanged

Airframe · payload · sensors · communications · autopilot.

Two layers of one system · neither works alone · licensed per platform

Onboard AI stopped being an extra feature. The adversary already fields it.

Frame rate, range and latency are fixed by the hardware chosen before the first sortie. The only open question is whether that choice can wait. Machine vision on NVIDIA Jetson modules has been recovered from several classes of enemy munition.

01

Geran-2, MS series

A modernised Shahed-136 with a camera and AI-based guidance; an NVIDIA chip among the components.

Defence Intelligence of Ukraine · War&Sanctions portal

02

V2U strike drone

A Jetson Orin operating as the terminal guidance system — in effect, a seeker head.

Debris analysis, 2025–2026

03

S-71M 'Monokhrom' cruise missile

A Jetson Orin and a camera previously identified in the jet-powered 'Geran-4'.

Defence Intelligence of Ukraine

04

Autonomous target selection

Jetson Orin recovered from the debris of drones used in trials of independent target selection on the Zaporizhzhia axis.

The New York Times, August 2026

The same model, on a computer that gives it the whole frame.

Within a chosen board the result is decided by three quantities, and only three: how often the aircraft looks, how far away there is anything to see, and how long it takes to say so.

2.8 m

between processed frames at 300 km/h

vs 16.7 m at 5 fps

Where the track holds a target through a manoeuvre. No model compensates for the gap between frames.

2,750 m

detection range on a 4 m target, from the full frame

vs 917 m once the frame is downscaled to a 640 input

Range is geometry — pixels on the target, not TOPS.

247 ms

from the target entering the field of view to a confirmed event on board

vs 794 ms at 7 fps

Routed through the ground, the radio link alone adds ~200 ms — a link the aircraft does not need.

How each of these is calculated →

The mission chain has to close on the aircraft, with no help from outside.

  1. Target

    enters the sensor's field of view

  2. Frame

    full resolution, hardware decode

  3. Detection

    your model, on every frame

  4. Track

    held through the manoeuvre and after loss

  5. Confirmation

    three frames of five, not one trigger

  6. Terminal guidance

    on board, with nothing asked of the outside

No link in that chain passes through the radio channel, the ground, or satellite navigation. That the chain closes on the aircraft is the condition for working under jamming — not an extra feature switched on separately.

Three ways to put AI on an aircraft. They differ in who owns the product afterwards.

A neutral layer on top of your computer and your models barely exists on the market.

ComparisonVertically integrated systemBuild it yourselfAsterioniq
Onboard computerTheirsYoursYour choice, from 22 qualified modules
AI modelsTheirsYoursYours — we help build and prove them if you want
Flight controller & radiosReplaced by theirsUnchangedUnchanged
Mission coverageMostly the terminal phaseWhatever you buildReconnaissance through terminal guidance
Where the data staysUnder their rulesWith youInside your perimeter, no access for us
Time to first deploymentWeeksAbout two yearsWeeks
Who keeps the layer aliveThe vendorYour team, permanentlyUs, over the air to the whole fleet
What it costs youYou stop being the platform~15 engineers, ~$2.03m, ~2 yearsFrom $281.04 per platform per year