RESK Security Projects
Comprehensive AI Security Solutions
Our Mission
At RESK Security, we develop cutting-edge solutions to protect AI systems and infrastructure from evolving cyber threats. Our projects span from logits-level filtering and bitmask access control to agent observability and threat detection.
LLM Security Core
reskSecure
Bitmask-based LLM Security Firewall
Policy-driven LogitsProcessor that restricts model output based on capability bitmasks. Uses GPU-accelerated Aho-Corasick pattern matching with YAML policy configuration, hot-reload, and tool call prevention at the token level.
Key Features
- Capability bitmask permission system
- Dual severity mode: hard block and bias penalty
- Tool trigger blocking at generation time
- YAML policy with hot-reload and caching
resk-logits
GPU-Accelerated Logits Processor
Shadow ban system for filtering dangerous content during LLM generation. Uses vectorized Aho-Corasick automaton for O(1) token filtering with GPU acceleration. Compatible with HuggingFace and vLLM.
Key Features
- Vectorized Aho-Corasick with danger mask
- Multi-level shadow ban penalties
- Automatic EOS on complete matches
- vLLM and HuggingFace adapters
ReskPoints
The AI Agent Logger
Track every action your agents take with full context. Ship logs to Datadog, Prometheus, OpenTelemetry, webhooks, or file. Framework agnostic with decorator-based API.
Key Features
- Multi-destination logging
- Async batch processing
- Decorator and context manager APIs
- Framework agnostic
Multi-Platform & Integration
resk-llm
Robust Python library for securing LLM API interactions with protection against prompt injections, PII leaks, and malicious URLs.
View Toolresk-mcp
Security and management layer for the Model Context Protocol (MCP) Python SDK with monitoring and protocol validation.
View Toolresk-llm-ts
TypeScript security toolkit for LLM integrations with OpenAI-compatible API wrapper and built-in defenses.
View ToolUpcoming Projects
RESK Monitor
A monitoring solution for LLM deployments with real-time instrumentation and behavioral deviation detection.
In Development
ReskCrypt
Conversation encryption and model monitoring with AES-256-GCM and ECDSA digital signatures.
Learn MoreGNOM
Graph Neural Network monitoring for real-time threat detection and network topology analysis.
Learn MoreReskLayer
Advanced prompt injection detection using ModernBERT with DiffTransformer attention mechanisms.
Learn MoreTechnology Stack
Core Technologies
- GPU Computing: CUDA, PyTorch, vectorized Aho-Corasick
- Security: AES-256-GCM, ECDSA, Argon2, capability bitmasks
- Observability: Datadog, Prometheus, OpenTelemetry
- ML: Graph Neural Networks, ModernBERT, DiffTransformer
- Languages: Python, TypeScript/JavaScript
- LLM Frameworks: HuggingFace, vLLM, LangChain, OpenAI
Research and Development
RESK Security Research Lab
Our mission goes beyond security tools. RESK aims to become a leading research laboratory in AI security, advancing the science of protecting artificial intelligence systems through rigorous research, open collaboration, and innovative solutions.
Research Areas
- Prompt injection detection and prevention
- Token-level logits filtering and shadow banning
- Capability-based access control for LLMs
- Agent observability and monitoring
- Graph-based security analysis
Get Involved
Join our community of security researchers, developers, and AI practitioners.
Contact UsStay Connected
Follow our latest developments and contribute to our open-source projects.
- GitHub: /Resk-Security
- LinkedIn: /company/resk-security
- Email: contact[@]resk.fr