AgentArmor: A Technical Deep Dive into LLM Security Proxies

How an open-source proxy in front of your LLM works, what it blocks and where it stops

In this post, I want to look at AgentArmor, an open-source reverse proxy that sits between your application and an LLM provider. It scans prompts and responses, controls where the proxy can connect, and adds a few less common ideas, like a canary token inside the system prompt.

This post describes the project as it was in mid-May 2026. It has changed a l…