Your data
Raw and curated, never copied out
Where the line sits
The subject of the agreement is the right to remain a platform.
Not a feature, and not an advantage measured in kilograms. The right to choose your own computer, to own your models and your data, and to talk to your own customer — without keeping a permanent engineering team to pay for it. We work on your side of that line and leave nothing of ours behind.
02 / 04Your perimeter / What you license
Inside your perimeter — yours, permanently
Your labels
The annotation set you paid to build
Your weights
Every checkpoint, handed over
Your models
And every claim made about them
Your archive
Events and video for retraining
Your customer
The relationship stays yours
What you license from us
- The onboard layer on your computer
- The server layer, in your perimeter
- Over-the-air updates while it runs
- The MLOps toolchain that trains it
- The fleet registry and event store
- The engineers who do the work with you
03 / 04The six commitments
Six commitments, binding in every instrument we issue.
They survive the end of the agreement.
01
Weights handed over
Every checkpoint as it is produced, to any runtime or supplier you choose.
02
Your data never leaves
No copy reaches our infrastructure, and we obtain no access to it.
03
Your accuracy is your claim
We do not replace your model and we attribute no figures to it.
04
Nothing crosses between customers
No material from your programme reaches a manufacturer you compete with.
05
We build no competing model
We develop neither drones nor recognition models, so we combine manufacturers rather than displace them.
06
No exclusivity, either way
The catalogue of validated models stays open on non-discriminatory terms.
04 / 04Said plainly
Said equally plainly.
No cryptographic defence against satellite-signal spoofing exists — ours works by never consulting satellite navigation at any point in the chain.
Under full jamming nothing transmits, so data is written on board and sent when the link returns.
Target coordinates are carried through from your own source, so their accuracy equals the accuracy of that source.
We name no detection percentage in advance, because that is your model on your data.