Pre-flight safety checklists for autonomous AI agents — risk checks, implied requirements, and post-action verification.
What it does
AgentUtil Think provides intent security for autonomous AI agents. Before an agent executes a high-risk action (deleting records, sending emails, deploying code, processing payments), it can call the /v1/check endpoint with a short action identifier and natural-language description. The service returns a structured checklist of severity-ranked safety checks, risk factors, and domain-specific considerations — the kind of things a human would think about but an agent might miss, such as backup expectations, permission checks, rollback plans, and compliance needs. The service uses pure data lookup with no LLM in its path, claiming sub-50ms response times.
The API exposes four endpoints. POST /v1/check generates a pre-flight safety checklist for a proposed action. POST /v1/expand surfaces implied requirements that humans assume but never state explicitly. POST /v1/review produces a post-action verification checklist to confirm rollback readiness, stakeholder notification, and monitoring. GET /v1/actions lists all recognized action types across domains (database, communications, file_system, deployment, financial, user_management, api, content, commerce, infrastructure) with their risk levels and check counts.
Pricing follows a free tier of 10 queries per day per IP with no signup required. Paid access uses the x402 protocol, settling in USDC on Base at $0.002 per check request, $0.003 per expand request, and $0.002 per review request. The /v1/actions endpoint is free. Discovery metadata is available via agent-service.json and agent.json at well-known paths.
Capabilities
Use cases
- —An autonomous agent calls /v1/check before deleting database records to get a checklist of backup, permission, and rollback considerations
- —An agent receives a vague human instruction and calls /v1/expand to surface implied requirements like notifications and compliance needs
- —After completing a deployment, an agent calls /v1/review to verify monitoring, rollback readiness, and stakeholder notification
- —A developer integrates /v1/actions to display all recognized risky action types and their severity levels in an agent dashboard
Fit
Best for
- —Autonomous AI agents performing destructive or high-risk operations
- —Agent frameworks that need structured safety guardrails before executing actions
- —Workflows requiring compliance and audit trails for agent-initiated actions
- —Teams building agents that interact with databases, deployments, or financial systems
Not for
- —Real-time content moderation or text safety classification
- —General-purpose LLM inference or text generation
- —Agents that only perform read-only, low-risk operations
Quick start
curl -X POST https://think.agentutil.net/v1/check \
-H "Content-Type: application/json" \
-d '{
"action": "delete_records",
"description": "Delete all inactive users",
"domain": "database"
}'Example
Request
{
"action": "delete_records",
"domain": "database",
"context": {
"table": "users",
"estimated_rows": 15000
},
"description": "Delete all inactive users older than 2 years"
}Response
{
"summary": "High-risk bulk deletion of user records. Ensure backup, authorization, and rollback plan before proceeding.",
"checklist": [
{
"why": "Bulk deletes cannot be undone without a backup",
"check": "Verify a recent backup exists for the users table",
"category": "backup",
"severity": "critical"
},
{
"why": "User data deletion typically requires elevated privileges",
"check": "Confirm authorization to delete user records",
"category": "permissions",
"severity": "critical"
},
{
"why": "If criteria are wrong, affected rows need restoration",
"check": "Ensure a rollback plan is in place",
"category": "rollback",
"severity": "high"
}
],
"risk_level": "high",
"action_type": "delete_records",
"risk_factors": [
"Bulk data deletion",
"Affects user accounts",
"Potentially irreversible"
]
}Endpoint
Quality
Full OpenAPI schema with request/response definitions, clear pricing, and a working landing page. However, the x402 probe did not return a 402 challenge (returned 400 instead), so liveness via x402 is unconfirmed. No dedicated docs page exists. The example response is inferred from the schema rather than captured live.
Warnings
- —x402 probe returned HTTP 400 on POST rather than the expected 402 challenge — endpoint may require a valid request body to trigger the payment flow, or x402 gating may not be active on this path
- —No /docs, /api, /pricing, or /README pages found — documentation is limited to the landing page and OpenAPI spec
- —Example response is inferred from the OpenAPI schema, not captured from a live call
Citations
- —Pre-flight safety checklists for autonomous AI agents with sub-50ms response and no LLM in pathhttps://think.agentutil.net
- —Free tier: 10 queries/day per IP, no signup. Paid: x402 protocol (USDC on Base)https://think.agentutil.net
- —POST /v1/check costs $0.002, POST /v1/expand costs $0.003, POST /v1/review costs $0.002, GET /v1/actions is freehttps://think.agentutil.net
- —OpenAPI spec describes four endpoints with full request/response schemashttps://think.agentutil.net/openapi.json
- —Supported domains include database, communications, file_system, deployment, financial, user_management, api, content, commerce, infrastructurehttps://think.agentutil.net/openapi.json