MODEL GARDEN

Pre-trained models.
Retrained on fresh data.

Seven production-ready cybersecurity models spanning URL analysis, malware detection, network behavior, and browser threats. Each model is automatically retrained when new labeled data lands — so they track the evolving threat landscape.

4 ML models (fast, cost-effective)
3 DL models (higher accuracy)
Auto-retrained weekly
ML Models

Gradient boosting and random forest models. Sub-10ms latency. Lower hosting cost. Best for high-throughput pipelines where speed matters more than squeezing the last 2% of accuracy.

DL Models

Transformer and LSTM-based deep learning models. Higher accuracy on complex patterns (obfuscated code, redirect chains, sequential behavior). Slightly higher latency and hosting cost.

ML Models
Fast · Cost-effective · Sub-10ms

URL Phishing Detector

Gradient Boosting (XGBoost)
ML

Classifies URLs as phishing, suspicious, or benign. Trained on 28M+ labeled URLs. Updated weekly with new campaign data.

F1 Score
96.2%
Latency
< 5ms
Use cases
Browser extension protectionEmail security gatewayProxy filtering

URL Lexical Classifier

Random Forest + TF-IDF
ML

Lightweight classifier using only URL text features — no external lookups required. Ideal for high-throughput pipelines where latency is critical.

F1 Score
94.8%
Latency
< 2ms
Use cases
Real-time proxyLog analysisBatch URL scoring

PE Malware Classifier

LightGBM on PE header features
ML

Classifies Windows PE binaries as malware or benign using static file features. No execution required. Trained on 2.4M labeled samples.

F1 Score
95.4%
Latency
< 10ms
Use cases
AV augmentationEDR triageFile upload scanning

Browser Extension Risk Scorer

XGBoost on manifest + static analysis
ML

Scores browser extensions by risk level (0–100) using manifest permissions and static JS analysis. Trained on 18K+ labeled extensions.

F1 Score
93.6%
Latency
< 15ms
Use cases
Enterprise browser policyMDM / device managementExtension store vetting
DL Models
High accuracy · Complex patterns · 80–200ms

Malicious JavaScript Detector

Transformer (BERT-style on JS tokens)
DL

Detects malicious JavaScript in web pages and browser extensions. Handles obfuscated, minified, and AI-generated code. Trained on 890K labeled samples.

F1 Score
92.1%
Latency
~80ms
Use cases
WAF augmentationExtension vettingWeb scraper analysis

Data Exfiltration Detector

LSTM on network flow sequences
DL

Identifies data exfiltration patterns in network flow data. Trained on 4.1M labeled flows from honeypot environments.

F1 Score
89.7%
Latency
~120ms
Use cases
NDR/NTA platformsSIEM enrichmentDLP augmentation

Malicious Redirection Detector

Graph Neural Network on redirect chains
DL

Detects malicious redirect chains used in phishing and malware distribution. Models the full redirect graph, not just the final URL.

F1 Score
91.3%
Latency
~200ms
Use cases
Email link scanningAd fraud detectionBrowser safety
DEPLOYMENT OPTIONS

Run it your way.

Hosted API

Call the model via our REST API. We handle serving, scaling, and retraining. Pay per prediction.

No infrastructure required. Best for getting started quickly.

Export Weights

Download model weights (ONNX, PyTorch, or scikit-learn pickle). Run inference in your own infrastructure.

Full control. Air-gapped deployments supported.

Fine-Tune & Deploy

Start with our base model, fine-tune on your labeled data, and deploy via our API or export the weights.

Best of both worlds: our base + your proprietary signal.

FREE TIER

One ML model. 100 predictions/month. Free.

Free tier gives you access to the URL Phishing Detector (ML) with 100 API predictions per month. Upgrade to unlock all models, DL variants, and unlimited predictions.

Start free Compare plans