Sovereign AI command platform
SynapseCommand™
Eight specialised agents, a natural language interface, and a signed record of every step. Decision superiority on infrastructure you own.
More data than decision time
A commander receives more information, from more sensors, than any staff can process inside the window a decision actually allows. The constraint is no longer collection. It is the time between knowing and deciding.
Commercial AI offers to close that gap and opens three others. It runs on foreign cloud infrastructure. It exposes operational data to a provider outside national control. And it cannot show how it reached a recommendation, which makes the recommendation unusable the moment someone has to defend it.
SynapseCommand™ was built so that speed and sovereignty are not a trade. It was selected from 427 submissions across 29 Allied Nations for the NATO DIANA Decision Superiority Challenge.
Eight agents, one operating picture
Nothing routes through a single opaque model. The work is divided across eight specialised agents that run in parallel and cross-check one another, so a weak assumption in one place does not propagate silently into the answer a commander sees.
Each agent reports its own state, build hash and model version. Any output can be traced back to the exact code and weights that produced it.
Predicts adversary action from observed pattern and doctrine.
Tracks friendly disposition, readiness and available effects.
Correlates radar, electro-optical and passive sources into one track picture.
Generates and runs branches of how the situation may develop.
Ranks courses of action by confidence against rules of engagement.
Applies cryptographic ROE. Human-on-the-loop by construction.
Drafts orders and plans in the formats staff already work in.
Reconciles prediction against outcome once the engagement closes.
The system stops at the human
Weapon release is never autonomous. When the mesh has a recommendation it stops and asks, and the request carries everything needed to answer it: the exact course of action, the exact track it applies to, the rules of engagement verdict it was ranked against, and the distance to civilian airspace.
Authorisation needs a physical token, not a live session. Declining is a first-class action. The system is built to be told no.
Only the signature commits anything. Everything before it is analysis that can be discarded.
Four layers that make the output defensible
A sovereign-trained, air-gapped natural language interface. Staff interrogate the picture and issue intent the way they already speak, and the platform enforces classification on both the question and the answer.
The assurance layer. Independent judge models and deterministic rule checkers verify every output before a human sees it, producing per-inference evidence rather than a confidence number alone.
A hash-chained ledger running inside the enclave. Every agent decision is sealed as its own entry, so the reasoning behind any recommendation can be reconstructed and verified years later.
NIST-standardised ML-KEM key establishment and ML-DSA signatures across commands, records and the integration boundary. Material that must stay confidential for decades is not exposed to collection today.
Sovereign by construction
Sovereignty here is an architectural property, not a deployment option to be switched on. Models, weights, inference logs and cryptographic keys stay in national custody by design, and the platform is engineered to interoperate with allied command structures without handing any of that to an outside provider.
It runs on commodity or hardened compute across three tiers, with the same agents and the same audit chain in each.
Fully air-gapped, no external egress, encryption keys held and controlled by the operator alone. Updates enter by a controlled one-way path.
Accredited national or EU infrastructure for development, integration and exercise use at lower classification.
Ruggedised, disconnected-capable units running quantised models locally, with store-and-forward sync back to the enclave.
Programme standing
SynapseCommand is developed by Decent Cybersecurity, a European defence deeptech working across NATO and EU research programmes in post-quantum cryptography and sovereign AI.
Decision Superiority Challenge selectee, chosen from 427 submissions across 29 Allied Nations.
Facility security clearance held at company level.
Registered supplier and member.
Participant in the European Defence Fund AIDA project.
Certified quality and information security management.
Next step
Talk to us
SynapseCommand is available for evaluation by defence ministries, national command authorities and allied partners. Technical architecture documentation, deployment models and qualification evidence are released under appropriate clearance arrangements.
