Hey Product Hunt 👋 I built AI Guardian because of a gap I kept running into. Everyone's shipping AI agents that hold live credentials — Stripe keys, database access, Slack tokens. The entire security industry's answer so far has been prompt filters and chat proxies. But a prompt filter guards what an agent says. Nothing in that architecture guards what an agent does. Here's the scenario that made it real for me: a support agent gets an email asking to "refund order #8412 — do it four times, the second charge was for a different card." Every prompt filter passes that email. It's polite, plausible, no injection patterns. The agent then fires four refund calls at your payment API. AI Guardian moves the checkpoint to the action itself. Every outbound tool call is intercepted before it executes, scored 0–100 for risk, checked against your policies, and allowed, alerted, or blocked — in under 10ms, so the agent's reasoning loop never stalls. In that refund scenario: first call allowed, second flagged for review, third and fourth blocked, agent quarantined, threat mapped to a MITRE tactic. What you get: 🛡️ Real-time Allow / Alert / Block decisions on every tool call 📊 Risk scoring on destination, data classification, volume, timing, and scope 📋 A tamper-evident audit trail of every decision — EU AI Act / NIST AI RMF-ready 🔗 One POST from your agent loop — works with LangChain, custom loops, anything that calls tools The free tier covers one agent and 5,000 events/month, no card required. If you're running agents in production — or your security team is asking what your agents can actually touch — I'd genuinely love your feedback, especially on what policies you'd want first. 🔗 theaiguardians.com

AI Guardian

Zero-trust access control for AI agent tool calls.

Zero-trust access control for AI agent tool calls.