AI Defender — documentation and CLI

@grovetech/defender is runtime protection for AI applications. Deployment is one line: app.use(d.web()). It blocks prompt injection, API key leaks, data theft and cost attacks at runtime.

CLI commands

• defender init — detects your framework and wires the middleware into your app.
• defender check — scans your build (./dist) for leaked secrets and personal data. In CI it exits non-zero on a hit, so the deploy fails before a key ever ships.
• defender proxy — a stack-agnostic reverse proxy for when you'd rather not touch the code.
• defender replay — replays a captured attack against your target to verify the protection holds.

What Defender watches

Prompt injection in inputs, secrets in responses (API keys for 25+ providers), access to sensitive paths (.env, .git, .ssh, .aws), cost anomalies in LLM calls, and NIS2 audit logs.