Vixero Technology Enterprise · est. 2026

VIXERO.

Sarawak → Santa Clara

Physics-based network defense. We build systems that detect adversaries by the thermodynamics of their traffic — not its contents. Zero payload inspection. Pure math.

0.9952
F1 Score
99.50%
TPR
262μs
Latency
40Mpps
Throughput
✓ Anthropic CVP Certified ⬢ NVIDIA Inception ☁ AWS Activate · $10K arXiv:2604.02149
Vixero Technology Enterprise NVIDIA Inception Program Member
vix@aegis-01:~/AEGIS$ aegis_inference_test.py
CUDA 13.0 RTX 4090 BF16
AegisScan · Live in Beta · vixdev.cloud/scan

The first PCAP scanner that cannot read your traffic.

Drop a packet capture. AegisScan extracts 6-dimensional flow physics at the browser edge — packet size, inter-arrival time, direction, TCP window, flags, payload ratio — and returns a thermodynamic threat report. Your payload bytes never leave your device.

  • Privacy by architecture — payload stripped client-side before transmission
  • 262μs inference — TVD-HL-SSM running on AWS g5.8xlarge
  • VLESS Reality, GhostBear, AMOI morphing detection out of the box
  • Honeypot leaderboard — community captures train Tier V corpus
// capture-2026-04-23-vless.pcap
✓ Scan Complete
Packets Processed1,000,482
6D Sequences Analyzed1,000
Shannon Entropy H(X)3.2148 nats ▼
Manifold ProjectionPoincaré · boundary-anchored
Inference Latency262.27 μs
Threat Confidence 99.50%
! VLESS REALITY C2 TUNNEL · CONNECTION TERMINATED
Verified · Backed · Deployed
Anthropic
Cyber Verification Program
✓ CVP Certified
AEGIS is certified by Anthropic's Cyber Verification Program — verified safe for dual-use cybersecurity research. The only defense system of its kind to pass independent AI-safety review for autonomous threat detection.
NVIDIA
Inception Program
⬢ Member
Accepted into NVIDIA Inception for AI-first startups. GPU compute partner for Mamba-3 inference, BF16 mixed-precision training, and CUDA Graph optimization at line-rate.
Amazon Web Services
Activate · $10K Credits
☁ Active
AEGIS training and AegisScan inference running on AWS infrastructure. EC2 g5.8xlarge nodes for Mamba-3 Tensor Core compilation. Part of AWS Activate for bootstrapped startups.
The Stack

Three products. One thesis.

AEGIS is the engine. AegisScan is the product. HBEE is the future. Everything Vixero ships derives from a single conviction — that network defense belongs in physics, not in bytes.

Live · Research SOTA
// Detection Engine
AEGIS.
Thermodynamic Variance-Guided Hyperbolic Liquid SSM

A firewall that measures what your packets are, not what they say. TVD-HL-SSM fuses hyperbolic Poincaré embeddings, liquid time-constants, and Shannon entropy detection into a single linear-time core — immune to byte-level morphing.

F1 Score
0.9952
Latency
262μs
Throughput
40 Mpps
TPR
99.50%
Architecture ›
Beta · Shipping Now
// PCAP Threat Scanner
AegisScan
VirusTotal for network captures · privacy by architecture

Drop a PCAP. Get a thermodynamic threat report. Payload bytes never leave your browser — only the 6D physics tensor ever touches our inference API. Free tier is live at vixdev.cloud/scan.

Free Tier
15/day
Max Capture
50 MB
Pro
$19/mo
API
REST · Q3
Scan Now ›
Research Preview
// Sociotechnical Simulator
HBEE.
Human behavior entropy engine · preprint in preparation

Where AEGIS models what the packets say, HBEE models what the humans do. A multi-agent LLM-driven sociotechnical simulator for insider threats, phishing cascades, and trust collapse — the other half of the defense equation.

Agents
25
LLM Core
GLM-4.7
Runtime
RTX 3090
Target
USENIX '27
Request Preview ›
45 DAYS
Concept → SOTA
0.9952
F1 Score
99.50%
True Positive Rate
262μs
Bare-Metal Latency
40 MPPS
Theoretical Ceiling
Built in a dorm room. Published on arXiv. Now here.
AEGIS · Architecture

Four pillars, one core.

The TVD-HL-SSM architecture discards the Euclidean payload-reading paradigm entirely. It operates on 6-dimensional continuous-time flow physics, projected into a non-Euclidean manifold, modulated by liquid time-constants, and scored by thermodynamic variance.

01

6D Flow Physics

xᵢ = [Sᵢ, Δtᵢ, Dᵢ, Wᵢ, Fᵢ, Pᵢ]

Packet size. Inter-arrival time. Directionality. TCP window. Flag state. Payload ratio. No byte content — adversarial pre-padding is mathematically neutralized at the ingress layer.

§ III.A · AEGIS Paper
02

Hyperbolic Projection

φ(xᵢ) = Wₚxᵢ / (1 + ‖Wₚxᵢ‖ + ε)

Botnet topologies branch exponentially. Euclidean embeddings distort. Projection onto the Poincaré disk 𝔻ⁿ accommodates hierarchical routing structures without gradient explosion.

Nickel & Kiela · NeurIPS 2017
03

Liquid Time-Constants

dh/dt = −h(t)/τ(Δtᵢ) + f(h, x, t, θ)

Neural state decays proportional to microsecond IAT variance. The continuous-time ODE captures thermodynamic jitter that stateless surrogates cannot synthesize — exposing automated morphing.

Hasani et al. · AAAI 2021
04

Thermodynamic Variance

H(X) = −Σ P(xᵢ) log₂ P(xᵢ)

Sequence-wide Shannon entropy of hidden states. Human traffic has natural stochastic decay. Automated evasion produces rigid structural patterns — detectable as thermodynamic anomalies.

Vixero Proprietary · 2026
AEGIS · Live Inspector
inject a simulated 1,000-packet sequence into TVD-HL-SSM
interactive
H(X) Entropy
4.8128
Confidence
Latency
[aegis-v3] Ready. Awaiting 6D sequence injection...
Poincaré Disk · 𝔻²
‖x‖ < 1 · hyperbolic manifold · non-Euclidean separation
live
H(X)
4.81
Training Corpus · 400GB · 4-Tier

Trained on reality.

AEGIS was not trained on synthetic lab data. 400GB across four threat tiers — from trans-Pacific backbone traffic to proprietary adversarial captures that exist nowhere else on earth.

Tier I · Benign Normalcy

Planetary Baselines

WIDE MAWI · CIC-IDS-2017 · DoHBrw

300M+ packets of trans-Pacific backbone traffic, corporate enterprise flows, encrypted DNS captures. Teaches the Thermodynamic Sensor what natural, stochastic human decay looks like at planetary scale.

Tier II · Automated Swarm

IoT & Botnet Warfare

Aposemat IoT-23 (Mirai · Torii · Okiru) · CTU-13

Hardware captures of IoT devices infected with Mirai and Okiru botnets. Historical C2 beaconing DNA. Teaches the SSM to recognize forced, robotic sequence timing.

Tier III · Apex Predators

APT & Zero-Day Vault

Malware-Traffic-Analysis · BCCC-Mal-NetMem 2026

Seven-year in-the-wild archive: Emotet, ClickFix, Handala Wiper. Memory-resident rootkits and trojans. The radioactive material of the corpus.

Tier IV · Invisibility Cloaks

Proprietary Evasion

Vixero Custom Red Team · VLESS Reality · GhostBear · AMOI

VLESS Reality streams, GhostBear obfuscation, high-entropy VPN tunnels, AMOI adversarial morphing — traffic engineered to defeat classifiers, used to train the system that defeats it.

⚠ RAW PCAP CLOSED · 6D TENSORS GATED ON 🤗
400GB
Raw PCAP Analyzed
908,037
6D Sequences
10GB
.pt Tensor Distillation
4
Distinct Threat Tiers
Empirical Validation · 20% Stratified Holdout

Proven in silicon.

Evaluated on the 20% holdout partition containing zero-day rootkits, VLESS Reality mimicry tunnels, and AMOI-morphed captures. Checkpoint locked at Epoch 10.

Receiver Operating Characteristic
AUC · 0.9998
AEGIS TVD-HL-SSM V3 ET-BERT under adversarial padding random classifier ★ operating point · 99.50% TPR @ 0.21% FPR
Confusion Matrix · Epoch 10
F1 · 0.9952
Pred. Benign
Pred. Malicious
Actual Benign
123,505True Negative
265False Positive
Actual Malicious
287False Negative
57,551True Positive
Adversarial Comparison · Under Attack

The benchmark gap.

What happens when the adversary fights back. Architectural comparison under identical adversarial conditions — not a product attack, a physics reality check.

F1 Score Under Adversarial Conditions
1.00
0.75
0.50
0.25
0.00
0.2568
ET-BERT
byte-level transformer · degrades under pre-padding
0.8500
Standard SSM
euclidean state space · susceptible to manifold shattering
0.9952
AEGIS V3
hyperbolic + liquid + thermodynamic · zero payload
Based on published adversarial evaluation methodology. ET-BERT baseline per Jing et al. (2025). Standard SSM references linear SSM without hyperbolic or thermodynamic components. All F1 values under identical adversarial padding + temporal morphing attack conditions.
CapabilityET-BERTStandard SSMAEGIS TVD-HL-SSM
Adversarial Padding Resistance✗ Vulnerable✓ Immune✓ Immune
Temporal Morphing Resistance— Not applicable✗ Susceptible✓ Resistant
Payload Inspection Required✓ Required✓ Required✗ Never
F1 Under Attack25.68%~85.00%99.52%
Inference Latency~500 ms~300 ms262 μs
Research Record · 2026

Publications & artifacts.

Every claim on this site is backed by a paper, a dataset, or a public benchmark. The research is the product.

0.9952
F1 · Adversarial
99.50%
TPR · Zero-Day
0.21%
FPR · Benign
262μs
Latency · RTX 4090
908K
Sequences Released
Publications · Preprints · Artifacts 3 tracked
arXiv:2604.02149
AEGIS: Adversarial Entropy-Guided Immune SystemThermodynamic State Space Models for Zero-Day Network Evasion Detection · cs.CR / cs.LG
Published arXiv ›
Vixero Tech · 2026
AMOI: Adversarial Morphing Obfuscating IntelligenceTraffic morphing for low-latency circumvention · cited in arXiv:2604.02149 ref [6]
Manuscript in Prep Embargoed
HuggingFace
AEGIS Adversarial Corpus908,037 sequences · 10GB tensors · 4-tier zero-trust release · gated
Released Dataset ›
Vixero Tech · 2026
HBEE: Human Behavioral Entropy EngineSociotechnical simulation framework for insider-threat modeling · 100+ tick runs completed
Research Preview Preprint Pending
Commercial Deployment

For enterprise.

AEGIS is built for integration, not replacement. Three pathways to deploy physics-based network defense.

Enterprise Licensing

Embed the AEGIS detection engine directly into your existing SOC or SIEM pipeline. Physics-based anomaly detection with sub-millisecond latency. No rearchitecting required — AEGIS integrates as a standalone inference module at your network perimeter.

Pilot Deployment

Deploy AEGIS on NVIDIA BlueField DPU hardware for data center perimeter defense. We run a structured 90-day pilot — your infrastructure, your traffic, our detection engine. Zero payload inspection means zero privacy liability during evaluation.

Research Partnership

Co-develop on next-generation AEGIS architectures (PELT V4) and gain access to the Vixero adversarial corpus — 908,037 sequences across 4 threat tiers. Ideal for academic institutions, national cybersecurity agencies, and advanced threat research teams.

Privacy by Architecture — Not by Promise.

AEGIS operates exclusively on 6-dimensional flow physics: packet timing, size, TCP flags, directionality. Payload bytes are deliberately and permanently discarded at the eBPF ingress layer before any neural processing occurs. This means:

  • Zero payload stored
  • Zero TLS interception required
  • Zero user data retained or processed
  • GDPR and PDPA compliant by architectural design
  • No man-in-the-middle liability for your organization

Unlike Deep Packet Inspection or TLS proxy solutions, AEGIS cannot leak user data — because it never reads it.

Development Timeline

Roadmap.

From dorm-room research to sovereign cybersecurity infrastructure. Four phases, one mission.

Phase 1
01
Research & Validation
2026 Q1
CURRENT
  • AEGIS V3 · arXiv:2604.02149
  • 908K-sequence corpus released
  • NVIDIA Inception + Anthropic CVP
  • AWS Activate $10K approved
  • AegisScan beta shipped
  • IEEE TIFS submission · in progress
Phase 2
02
Architecture & Hardening
2026 Q2–Q4
  • PELT (AEGIS V4): Permutation Entropy + CfC + TTT-Linear
  • HBEE human entropy engine · preprint
  • AWS A10G benchmark validation
  • AegisScan Pro · Team tiers
  • USENIX Security 2027 submission
Phase 3
03
Production Deployment
2027
  • TensorRT compilation for BlueField DPU line-rate
  • GTC 2027 poster + live demo (Santa Clara)
  • First enterprise pilot deployment
  • AegisScan Enterprise · SOC integration
  • Sdn Bhd incorporation · MD Status
Phase 4
04
Scale
2027–2028
  • ASEAN sovereign cybersecurity positioning
  • Multi-vendor SOC integration (CrowdStrike, Palo Alto)
  • Government and defense contract pipeline
  • OEM licensing at scale
Why Now · Why Us

For investors & partners.

Vixero is a research-first deep-tech company with a shipping product, a peer-reviewable paper, and institutional backing. Not a deck. Not a demo.

01
Published Research
arXiv:2604.02149 indexed on Semantic Scholar, DeepDyve, and cs.CR feeds within 8 days of submission. Cited in academic trackers. The paper is the defensible moat — the corpus is the flywheel.
02
Strategic Partners
NVIDIA Inception member (GPU compute & GTC access). Anthropic CVP Certified (dual-use safety review passed). AWS Activate approved ($10K infrastructure credits). Each gate was reached on merit, not introductions.
03
Shipping Product
AegisScan is live at vixdev.cloud/scan — the first PCAP scanner that cannot read your traffic. Free tier, Pro $29/mo, Team $99/mo. Every scan harvests a 6D tensor for the Tier V corpus. Product is the flywheel.
04
Defensible Moat
The Tier IV corpus (VLESS Reality, GhostBear, AMOI captures) is closed and proprietary. The AMOI adversarial methodology is unpublished. Competitors must either rebuild both, or license from us. The moat is the work itself.
The Builder

Founding myth.

Vickson Ferrel
Founder & Lead Architect · Vixero Technology Enterprise
Kuching, Sarawak · Malaysia · UNIMAS FCSIT

AEGIS was not born in a well-funded lab. Built on a rented cloud server. Trained on real malware. Submitted to NVIDIA on the same day as Jensen Huang's GTC keynote. AMOI was built first — its failure modes studied, then reversed into a defense system. 45 days. Rented server. Published on arXiv.

For Humanity, From Humanity.
✓ Anthropic CVP Certified ⬢ NVIDIA Inception ☁ AWS Activate · $10K 📄 arXiv:2604.02149 🐱 GitHub Education ☁ Cloudflare Enterprise
◈ Build Pipeline · 45 Days
AMOImorphing framework
Ayaka AH-MSIwhite-box adversary
AEGIS V1reversed defense · 0.9604 F1
AEGIS V3 APEX0.9952 F1 · 262μs
arXiv:2604.02149published · cs.CR + cs.LG
AegisScanshipping product
40 MPPSTheoretical Ceiling
☁ On AWS

For humanity, from humanity.

Vixero is accelerating AEGIS toward production-grade TensorRT compilation for BlueField DPU line-rate deployment in next-generation data centers. We are open to partners, pilots, and principled capital.

Anthropic CVP Certified
⬢ NVIDIA Inception
AWS Activate