Install Espada

The Agent Action Firewall

Deploy agents to production. Your data is truly protected.|

Espada is an infrastructure agent that runs inside your network — and the gate that lets Claude, GPT-4, or your own agents run terraform plan, kubectl apply, aws deploy in your real cloud. Hardware-key signature on destructive change. One self-hosted binary. Zero outbound.

curl -fsSL espadafirewall.com/install | sh See how it works

§ 01 · The Agent Action Firewall

Three steps.
Every tool call. Every time.

This is what Espada calls the Agent Action Firewall — a single self-hosted binary that sits in front of your AI agent's outbound calls. The gate runs the same three steps on every action, whether the agent wants to write a file, deploy a container, or destroy your production VPC.

  1. 01

    Intercept

    Your AI agent — Claude, GPT, Gemini, your own — tries to run a tool call. Espada catches it at the gateway, before a single packet reaches your cloud control plane.

    Example claude code → terraform apply → ESPADA

  2. 02

    Resolve

    We do not trust the arguments. We ask the cloud what will actually happen — and surface every consequence in plain text: resources destroyed, IAM widened, networks opened, cost lost.

    Example 41 destroys · 0 creates · 2 modifies · $48,200/mo lost

  3. 03

    Sign or block

    Destructive change waits on a human with a hardware key. Auto-approval is structurally impossible. Every decision — approve, block, sign — is hash-chained and signed into an audit log on your hardware.

    Example Press y to sign with YubiKey · n to block · r to re-plan

Step 02 is the part most security tools skip. Prompt classifiers ask whether the input looks dangerous; we ask the cloud what the action will actually do. The argument string never decides what gets destroyed — the cloud's own plan does.

Ready when you are.

One binary. One install. One hour to your first signed action.

brew install espada-firewall/tap/espada Read the security model