What is an agent firewall?

Your agent has your API keys. It makes HTTP requests. It calls tools that read files, query databases, and fetch web pages. Any of those can leak credentials, get prompt-injected, or exfiltrate data. An agent firewall sits between the agent and everything it touches. It scans traffic in both directions before anything gets through. Not a guardrail inside the model. Not a policy engine that checks tool names. A proxy that inspects requests and responses before they reach either side. Why agents need firewalls Traditional apps don't have this problem. A web app talks to a database and an API. We understand the attack surface, and we've had decades to build WAFs, rate limiters, and network policies around it. Agents are different. They decide at runtime which tools to call, what URLs to fetch, and what data to send. You can't write a static allow list for something that improvises. Three things go wrong: Credentials leak outbound. The agent has API keys in its environment. A prompt inject