Open-Source LLM Security,
Enforced at the Token Level
AI systems have access to your most sensitive data — and they remain critically vulnerable. RESK builds the open-source stack that fixes this: a deployable LLM firewall, token-level filtering engines, agent observability, and deception for AI scanners.
Every Layer,
One Hub Each.
Deep guides per topic — threat models, mechanisms, and the open-source tools that enforce them.
LLM Security →
The complete threat model: prompt injection, data leakage, output filtering, access control.
AI Agent Security →
Tool permissions, least privilege, tool-call authorization, agent observability and audit.
MCP Security →
Mechanisms, layers, tool-call authorization and monitoring for the Model Context Protocol.
LLM Firewall →
Self-hosted, OpenAI-compatible enforcement: RBAC, policies, token-level filtering, Docker.
Prompt Injection →
Attack taxonomy, system prompt leakage, real examples and layered prevention techniques.
Enterprise AI Security →
Governance, compliance, per-role access control and audit trails for LLM deployments.
Catch the Threat.
Not Just the Prompt.
Prompt filters are jailbreakable. Post-hoc moderation acts too late. RESK enforces security where no injection can reach: inside the generation loop.
RBAC Bitmask
A 64-bit capability system per user — tools, code gen, DB access, PII, email — enforced at the token level.
Token Filtering
The resklogits engine (ShadowBan + Aho-Corasick) makes dangerous content physically ungeneratable. GPU-accelerated.
Admin Console
Users, roles, policies, providers, sessions, audit logs, and a D3 graph of your security topology.
Open Ecosystem
10 open-source libraries across Python, TypeScript, and Bun — from input-time detection to agent logging.
One Docker Command.
Full Pipeline Coverage.
ReskSafety deploys as a full-stack application and exposes an OpenAI-compatible endpoint — your existing code only changes its base URL.
git clone https://github.com/Resk-Security/ReskSafety.git
cd ReskSafety && docker compose up -dclient = OpenAI(base_url="https://your-firewall/v1")
# Every request: auth → RBAC → policies → token filter → auditOne Stack,
Every Layer
ReskSafety is the platform layer. Around it, specialized open-source libraries cover the full attack surface of LLM applications.
resk-llm · resk-llm-ts
Input-time prompt-injection detection for Python and Node/Bun — 11 attack-specific detectors, zero heavy dependencies.
reskSecure · resk-logits
Per-user firewalls with logits-level filtering — unauthorized content becomes physically ungeneratable.
resk-mcp · ReskPoints
Security for MCP servers and full observability for AI agents — every action logged and exportable.
resk-caching · ReskCrypt · honeycrawlpot
Semantic caching to cut costs, end-to-end encryption for AI conversations, and honeypots that feed fake secrets to AI scanners.
"AI systems have access to your most sensitive data. They remain critically vulnerable." RESK Security — Why We Build
Secure Your LLM Stack Today
Live demo, open-source tools, and security research — start anywhere. Running AI at scale in your organization? Explore our enterprise AI security services.