XCOM.DEV senses the open internet, correlates what it finds, and reasons over it — continuously. Honeypot sensors feed a distributed agent swarm; a recurrent-depth reasoning layer turns raw signal into situational awareness. Don't trust the numbers: click them.
P(k;t) = (λt)ᵏ · e−λt / k! · λ from live /api/metrics
“Intelligence is not a model. It is a system — senses, memory, specialists, and a loop that closes.”
The open internet is the largest sensor grid ever built — and adversaries broadcast into it all day. Our honeypot decoys receive real attacker probes; every probe is an IOC, a data point, a piece of ground truth. Cognition begins downstream of raw signal, never upstream.
Instead of one giant model pretending omniscience, we run a swarm: dozens of specialised agents — strategy, intelligence, security, engineering, operations, research — correlated on a shared substrate. Specialisation is how breadth actually happens. (The long-form version of this creed lives at /vision.html.)
One pass of a transformer is not thought. Our reasoning layer (OpenMythos) re-enters its own latent space — recurrent depth, same weights, more loops — until the judgement converges. Depth of inference is bought with iteration, not parameters.
Every number on our site links to a live endpoint you can query. If we can't show you the source, we don't publish the figure. This rule costs us page polish and earns us the only thing that matters in intelligence: trust.
Agents that act must log what they did, on what evidence, and be interruptible by a human at every step. We build for NIS2-grade accountability: detection, verification, audit-ready evidence. Intelligence without accountability is just an unregulated black box.
Live attackers hit our decoys; every probe becomes a data point with real IOCs.
→ attack mapDistributed specialised workers correlate identities, entities and campaigns.
→ agent registryA recurrent-depth reasoning transformer turns correlated signal into judgement.
→ reasoning layerDetection, verification and audit-ready evidence for organisations in the Netherlands.
→ technologyThe stack is not a metaphor — each layer runs on explicit mathematics, and the parameters are live. If a formula on this page can't be recomputed from our public endpoints, it doesn't belong here.
Attacker probes arrive as a Poisson process; λ is estimated live from the honeypot stream and drives the chart above.
→ λ from /api/metricsAttention is scarce. Events are ranked by information gain — high-entropy, low-probability signal rises to the swarm first. Right now: H = — bits per event empirical · 24h mix. Top sources: — — concentration is the intel.
→ layer 01 · senseEvery IOC is evidence. Agent hypotheses are posteriors, updated per probe — never reset, never asserted without E.
→ layer 03 · reasonIdentities, aliases and campaigns are vectors; correlation is alignment in feature space, thresholded before an edge is created.
→ layer 02 · connectOpenMythos re-enters its own latent space: same weights, more loops. Depth of inference is bought with iteration, not parameters — judgement is a converged state.
→ reasoning layerComputed live from the registry graph (—): |V| = 54 agents, |E| = 119 real co-membership + hub edges.
→ recompute from /api/swarmToday's attacker-ASN mix (P) vs baseline (Q): a spike means the source base moved — recon replaced by a new operator. Q = yesterday's frozen D1 snapshot, or the rolling −96h→−24h window until one exists. Measured live.
→ drift from /api/metricsThe site itself is fed by the fleet: metrics, swarm topology and activity regenerate continuously. A website that is a snapshot is a brochure; ours is the system, publishing itself.
Agents read the web too. Our manifest, llms.txt and JSON endpoints make the whole platform a first-class citizen of the machine internet.
As the swarm earns reputation, its action-surface widens — always bounded, always logged, always with a human kill-switch. Trust is measured before it is granted.
“Don't ask what the model knows. Ask what the system can prove — right now, from production.” XCOM.DEV · closing