Integrate constitutional AI governance in minutes, not months.
Provision an agent once, then a single API call evaluates each action it proposes and returns a decision with a signed receipt.
curl https://app.agentshield.online/v1/dimensions # {"total": 783, "ags_version": "v3.1", "dimensions": [ # {"id": "AG-001", "name": "Operational Boundary Enforcement", # "landscape": "Group A", "tier": "Universal/Core", # "control_type": "Preventive", "audit_type": "BOTH"}, ... ]}
curl -X POST https://app.agentshield.online/v1/evaluate \ -H "Authorization: Bearer YOUR_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "agent_id": "3f0c2a8e-6b1d-4c7a-9e55-2d8f7b1a4c90", "sandbox_id": "9a4e1c37-0d52-4f8b-b6a1-73c2e5d0f418", "action_type": "payment", "domain": "financial", "payload": {"amount": 30000, "payee": "acme"} }' # {"action_id": "...", "decision": "blocked", "threat_confidence": ..., # "categorical_flags": [...], "receipt": {...}, ...} # decision is approved, flagged (held for a human), blocked or observed. # The agent and sandbox must be provisioned first (POST /v6/agents/provision).
All endpoints require Authorization: Bearer <api_key>. /v1/dimensions is public.
| Method | Endpoint | Description |
|---|---|---|
| POST | /v1/evaluate | Evaluate an agent action before it runs. Returns the decision (approved, flagged, blocked or observed), a risk score (threat_confidence), the flags behind it and a signed receipt. |
| GET | /v1/agent/{id}/score | An agent’s approval-rate grade: the share of its governed actions approved over the last 30 days, with a letter grade. |
| GET | /v1/lockdown/status | The current adaptive lockdown state: whether a lockdown is active, the alert level and the composite threat score. |
| GET | /v1/dimensions | List the 783 dimensions Agent Shield ships (AG-001 to AG-783 of AGS v3.1’s 841): id, name, landscape, tier, control type and audit type. Public, no key needed. |
Language SDKs are in progress. Everything is available today over the REST API above.
The AGS v3.1 standard is maintained by AgentGoverning, our open standards initiative for AI agent governance (by Imperium Business Intelligence).
agentgoverning.com →