European AI Models.
Built to Protect.
Patronus builds Purpose-built security models optimized for real-time endpoint inference. Detect prompt injection, sensitive data exposure and agent abuse before threats reach the model.
Security Models Designed For Runtime Protection
Unlike general-purpose foundation models, Patronus models are purpose-built for AI security, governance and runtime protection. Our detection engine combines multiple layers of analysis for fast, reliable, deterministic decisions.
Ensemble Architecture
Multiple specialized detection layers, heuristics, gradient-boosted trees and transformer models, combine for robust, accurate security decisions.
Sub-5ms Latency
Deterministic heuristic rules provide near-instant classification without model inference, critical for real-time endpoint protection at scale.
On-Device Inference
All analysis runs locally on the endpoint. No prompts, responses or agent data leave your system, ever.
Heuristic Detection Engine
Deterministic heuristics and protocol-aware detection rules, near-instant detection for known patterns, extremely low-latency classification without model inference.
Machine Learning Detection
Gradient-boosted decision trees (LightGBM) evaluate structured risk signals from prompts, responses, tools and agent activity, excellent performance, highly efficient on endpoint devices.
Transformer Security Models
Compact transformer models trained exclusively for security classification: BERT-style architectures optimized for local deployment and real-time inference.
Multi-Task AI Architecture
No single model is optimal for every security problem.
Patronus uses a multi-task AI architecture inspired by mixture-of-experts systems. Instead of relying on one large general-purpose model, Patronus routes security tasks to specialized detection layers optimized for speed, accuracy and memory efficiency.
Reduced Memory Footprint
Multi-task models detect multiple threat classes within a single compact model, reducing memory usage while still covering prompt injection, sensitive data exposure, policy violations and agent risk.
Ensemble Decisions
Patronus combines signals from heuristics, gradient-boosted trees and transformer models to improve threat detection, reduce false positives and provide more robust risk decisions.
Intelligent Model Routing
Inspired by mixture-of-experts architectures, Patronus dynamically routes each interaction to the most suitable detection layer: heuristics, machine learning classifiers or compact transformer models.
HuggingFace
Wolf Defender
Wolf Defender is our open AI security model for prompt injection detection, built to identify jailbreaks, instruction overrides and agent manipulation attempts before they impact AI systems. Lightweight, security-focused and optimized for real-world deployment.
Try Wolf Defender5,000+ / month
Downloads on Hugging Face
BERT-based
Architecture
On-Device
Inference
European
Built & Maintained
Open Source in Numbers
25,000+ downloads and counting
The Patronus security models have passed 25,000 total downloads on Hugging Face; the Wolf Defender models alone are downloaded more than 5,000 times per month.
25,000+
total model downloads
5,000+
monthly downloads, Wolf Defender
6
open-source collections
Open Model Catalog
All Patronus Models
One main model from each Patronus open-source collection, built for fast, local AI security. Beyond these six you will find further models, variants and quantizations in our Hugging Face organization.
Browse the full collection on Hugging FaceFAQ