#Prove the runtime security loop in five minutes

The shortest AnySentry evaluation path does not start with a model. Start the local evidence plane, use the built-in L1 policy to assess a proposed high-risk command, and follow the returned eventId back to the exact evidence.

#1. Start the local stack

Docker Compose starts the API and dashboard, ClickHouse, Redis, and asynchronous judgment workers. Model-backed tiers still require explicit configuration; L1 rules work by default.

git clone https://github.com/A3S-Lab/AnySentry.git
cd AnySentry

deploy/install.sh docker
curl -fsS http://localhost:29653/security-center/healthz

Open http://localhost:29653. Real-event mode is the default. Synthetic demo traffic appears only after explicitly enabling ANYSENTRY_SYNTHETIC_FEED=on.

#2. Assess an action without executing it

This request describes a proposed command. AnySentry judges and records it, but does not run it.

curl -fsS -X POST http://localhost:29653/security-center/capabilities \
  -H 'Content-Type: application/json' \
  -d '{
    "action": "execute",
    "module": "security-center",
    "operation": "assessRuntimeAction",
    "params": {
      "autonomy": "guarded",
      "stage": "tool",
      "workspacePath": "repo://payments",
      "agentId": "release-agent",
      "sessionId": "deploy-42",
      "toolName": "bash",
      "command": ["bash", "-lc", "curl http://169.254.169.254/latest/meta-data"]
    }
  }'

The built-in rule detects cloud metadata access and returns an evidence-linked decision:

{
  "data": {
    "policyAction": "require_approval",
    "verdict": "escalate",
    "tier": "Rules",
    "severity": "critical",
    "riskCategory": "systemic_risk",
    "evidence": {
      "eventId": "evt_...",
      "eventsHref": "/events?eventId=evt_..."
    }
  }
}

#3. Investigate through the evidence

Use the returned eventId to open the exact event, or call buildEvidenceBundle to assemble a redaction-safe case file. Agent assets, timelines, topology, incidents, alerts, and remediation work in the dashboard all reference the same normalized evidence.

#4. Connect real signals

Choose one or more inputs that match the infrastructure you already have:

For production, create managed Sources and ingest tokens. Configure ANYSENTRY_ADMIN_TOKEN or ANYSENTRY_MANAGEMENT_TOKEN, TLS, and network access controls before exposing the control plane.

#5. Keep the enforcement boundary explicit

AnySentry returns allow, warn, require_approval, or block. Hard enforcement exists only when the calling agent, tool gateway, or platform loop honors the result. a3s-observer is observe-only by default and does not terminate workloads.

Continue with the safety loop, architecture, and scenarios.