Scan, redact, and optionally block outbound requests to AI providers (OpenAI, Anthropic, Gemini, and more) for leaked secrets and PII. Review in an inbox and fail CI before a prompt leaks. SDK-agnostic, at the HTTP layer.