← Work/AI Security · DLP Platform · 2025–2026
ZeusLock
The security layer for the AI era — DLP that stops data leaking into ChatGPT & Co.
Enterprise data-loss prevention built for AI usage: a browser extension and desktop agent feed a FastAPI engine where a jailbreak classifier, regex layer and a self-hosted multimodal LLM decide allow / alert / block in real time. Ships as SaaS and licensed air-gapped on-prem.
14+
AI providers covered
SaaS + air-gap
Editions
self-hosted LLM
Detection
AI in this product
Self-hosted multimodal LLM (llama.cpp on GPU) at the core of the detection pipeline — the AI never leaves the platform.
AI data-loss prevention
The problem
Employees paste customer PII, credentials and financial data into AI tools every day. Classic DLP never sees it — the data leaves through chat apps, desktop tools and MCP agents.
The outcome
EU-sovereign AI security in production — detection runs on our own GPU with open weights.
What we built
- Browser extension (Manifest v3) intercepting prompts and uploads inside LLM web apps; Electron desktop agent covering 14+ AI providers system-wide, with MCP monitoring and Shadow-AI discovery.
- FastAPI DLP engine: jailbreak classifier ‖ regex layer ‖ self-hosted LLM → allow / alert / block, with format-preserving anonymization.
- Multi-tenant PostgreSQL with row-level security, Keycloak OIDC, SIEM export (OCSF) to Splunk/Sentinel/Elastic.
- Kubernetes + GitOps delivery; one universal Helm chart drives SaaS and Ed25519-licensed on-prem installs.
A credit card caught before send
Architecture — three-layer detection
Stack
Next project
Mechanic Connect →