OrchSec: runtime firewall that blocks unsafe AI agent actions
OrchSec intercepts AI agent tool calls and final messages, applying deterministic policies and optional LLM judging to decide ALLOW, BLOCK, or other actions. It normalizes actions, matches them against YAML policies, and can redact or log audit records. Designed for developers deploying agents who need a safety gate, it works as a self‑hosted proxy and integrates with existing services like DVEA without code changes. Compared to ad‑hoc scripts, it offers a structured, extensible policy engine with audit logging.
View on GitHub →huzaifaisavailable/orchsec-hackathon