Show HN: AgentGate – Fast authorization gateway for AI agent tools
Category: security
Tags: ai-security, agent-safety, authorization, prompt-injection, microservice
Score: 7.8/10 (Innovation: 8, Technical: 8, Documentation: 8, Utility: 7)
AgentGate is an out-of-process authorization firewall for AI agent tool calls, using a fine-tuned Qwen 0.8B model to classify prompt injection, goal alignment, reversibility, and exfiltration risk in under 40ms. It combines a policy engine and human approval queue to provide deterministic, low-latency security for autonomous agents, addressing a critical gap in AI safety.
Target audience: AI engineers, backend developers, DevOps, security engineers
Repository: https://github.com/mohitdebian/AgentGate · TypeScript · MIT
View on Hacker News