What you get
29 detection patterns — from real CVE research across NVIDIA, PyTorch, HuggingFace
Hourly process scan — checks running processes against ML attack patterns
Local dashboard — HTML monitor included, runs in your browser
Zero interference — no blocking, no shell changes, no performance impact
One-command install — unzip, run install.sh, done
We protect secrets. We don't collect them.
No file contents — we never read your code, data, models, or documents
No credentials — we never see passwords, tokens, API keys, or environment variables
No network traffic — we don't sniff, capture, or inspect your connections
Metadata only — process names and the first 80 characters of command lines. That's it.
Local first — everything logs to your machine. Connected mode is opt-in, not default.