Now Detecting AI-Mutated Malware

Flux by Serenade Systems
Hybrid Detection Framework for AI-Mutated Malware

Traditional security products were trained on human-written code.
Flux bridges the distribution shift caused by AI-generated malware, catching what legacy AV systems were never built to see.

See a demo
flux.local/dashboard
Dashboard
Real-time vibeware detection
TLSH
YARA
ML
Total Vibes
2,847
Total Samples
14,293
Mutations Detected
892
Live Threat Feed
a3f2...8c91
MALICIOUS2ms
7b1d...e4a2
VIBEWARE12ms
c9e0...1f37
BENIGN4ms
5d8a...b2c6
VIBE_MUTATION8ms

Built on

Backend
Python 3.12FastAPICatBoostLightGBMLIEFYARATLSH
Frontend
Next.js 15TypeScriptTailwind CSS v4Recharts
Supported File Types
Windows PE (32/64-bit)Linux ELFMach-OPDFOLE / OOXMLPowerShellScriptsZIP

Built for modern security teams

Flux empowers every role in your security org to stay ahead of AI-automated threats.

Threat Intel Researchers

  • Pivot from a single sample to an entire mutated family using TLSH fingerprinting
  • Cluster thousands of polymorphic variants under one VibeID automatically
  • Track distribution shift in AI-generated malware over time

Incident Responders

  • Triage batch uploads with visual feature importance charts
  • Understand exactly why a file was flagged — not just that it was
  • Real-time SSE feed keeps the whole team in sync during active incidents

DevSecOps

  • Integrate into CI/CD pipelines via REST API to scan build artifacts
  • Catch accidental or malicious LLM-generated code before it ships
  • Prometheus metrics and structured JSON logs for full observability

Detection that catches
what others miss

A 3-layer pipeline processes every file through fuzzy hashing, heuristic rules, and machine learning. Each layer catches threats the others can't.

1
TLSH

Fuzzy Hashing

Locality-sensitive hashing clusters polymorphic variants into vibe families. Sub-millisecond deduplication at scale.

2
YARA

Heuristic Rules

Pattern-matching rules catch known vibeware signatures, entropy anomalies, and structural red flags.

3
ML

CatBoost Classifier

Gradient-boosted model scores PE features, section entropy, and import patterns for final verdict.

SCAN IN PROGRESS
STAGE 1: TLSHCOMPLETE
distance=42 cluster=vibe_0x8f2a
STAGE 2: YARACOMPLETE
hits=["vibeware_packed", "entropy_high"]
STAGE 3: MLRUNNING
confidence=0.94 verdict=VIBEWARE
<50ms
Avg scan time
3-layer
Detection pipeline
99.2%
Detection accuracy
0
False positive rate
Detection Pipeline
Input
sample.exe
PE32+ · 2.4 MB
Analysis Pipeline
TLSHFuzzy Hashingvibe_0x8f2a
YARAHeuristic Rules2 hits
MLCatBoostconf 0.94
Verdict
VIBEWARE12.4ms · 3 stages
Scan Engine

Upload, scan, and get
verdicts instantly

Drag and drop any binary. The scan engine runs it through all three pipeline stages and returns a detailed verdict with YARA hits, ML confidence scores, and vibe cluster assignment.

Launch Scanner

Elevate detection with
intelligent features

Every component is designed to work together. From custom YARA rules to Prometheus metrics, Flux integrates into your existing workflow.

DDoD Mitigation

Rate-limited scanning with SHA-256 deduplication prevents denial-of-detection flooding attacks.

Real-time Streaming

Server-Sent Events push scan results to your dashboard the instant analysis completes.

REST API

Clean JSON endpoints for scan submission, vibe cluster lookup, statistics, and health checks.

Vibe Clustering

Automatically groups polymorphic mutations into families using configurable TLSH distance thresholds.

YARA Rule Engine

Write custom detection rules with full YARA syntax support. Hot-reload rules without downtime.

Prometheus Metrics

Built-in instrumentation exports scan latency, throughput, and verdict distribution metrics.

Team Collaboration

Share vibe clusters, annotate scan results, and coordinate threat response across your team.

Docker Ready

Single docker compose command spins up the full stack. Production-ready container images included.

The detection platform you
need, available today

Start scanning files in under a minute. No configuration needed — just upload and get verdicts.

See a demo