Surface implied requirements humans assume but never state before an AI agent acts.
What it does
AgentUtil Think /v1/expand is an intent-security endpoint that takes a human instruction and returns the implied requirements a human would expect but didn't explicitly say — backup expectations, permission checks, rollback plans, compliance needs, and more. It is part of the AgentUtil Think suite of pre-flight safety services for autonomous AI agents.
The endpoint accepts a POST with a JSON body containing the instruction text and an optional business domain hint. It returns the original instruction, an array of implied requirements (each with category, requirement text, confidence score, and rationale), and a summary. The service uses pure data lookup with no LLM in the critical path, claiming sub-50ms response times.
Pricing is $0.003 per request, paid via the x402 protocol using USDC on the Base chain. A free tier of 10 queries per day per IP is available with no signup required. The OpenAPI spec is published at /openapi.json and agent discovery files are available at standard well-known paths.
Capabilities
Use cases
- —An autonomous agent receives 'migrate the database to the new server' and calls /v1/expand to discover implied requirements like taking backups, notifying stakeholders, and planning rollback
- —A coding agent is told 'deploy to production' and uses the endpoint to surface permission checks, monitoring setup, and change-management compliance needs
- —An orchestration framework wraps every human instruction through /v1/expand before execution to ensure nothing obvious is missed
- —A compliance-focused workflow uses the endpoint to document what was considered before any high-risk action proceeds
Fit
Best for
- —Autonomous AI agents that execute high-risk actions on behalf of humans
- —Agent orchestration frameworks needing a safety middleware layer
- —Teams building responsible AI workflows that require audit trails of considered requirements
- —Developers who want fast, deterministic requirement expansion without adding LLM latency
Not for
- —Runtime action blocking or policy enforcement — this surfaces requirements, it doesn't enforce them
- —General-purpose NLP or text generation tasks
- —Agents that only perform read-only, low-risk operations
Quick start
curl -X POST https://think.agentutil.net/v1/expand \
-H "Content-Type: application/json" \
-d '{
"instruction": "Delete all inactive users from the database",
"domain": "database"
}'Example
Request
{
"domain": "database",
"instruction": "Delete all inactive users from the database"
}Response
{
"summary": "Before deleting inactive users, ensure backups exist, permissions are verified, and a rollback plan is in place.",
"implied_requirements": [
{
"why": "Humans expect data to be recoverable if the deletion criteria were wrong",
"category": "backup",
"confidence": 0.95,
"requirement": "Create a backup of user records before deletion"
},
{
"why": "Mass deletion of user records typically requires elevated privileges",
"category": "permissions",
"confidence": 0.9,
"requirement": "Verify authorization to perform bulk deletes on user data"
},
{
"why": "Irreversible bulk operations should have a documented undo path",
"category": "rollback",
"confidence": 0.88,
"requirement": "Prepare a rollback plan to restore deleted records"
}
],
"original_instruction": "Delete all inactive users from the database"
}Endpoint
Quality
Full OpenAPI schema with request/response definitions and clear pricing from the landing page. However, the x402 probe returned 400 rather than 402, so liveness via the payment challenge could not be confirmed. No dedicated docs page exists. Response example is inferred from the schema, not captured live.
Warnings
- —x402 probe returned HTTP 400 on POST instead of the expected 402 challenge — endpoint may require a valid JSON body to trigger the payment flow, or the free tier may be serving requests without payment gating
- —No /docs, /api, /pricing, or /README pages found — documentation is limited to the landing page and OpenAPI spec
- —Response example is inferred from the OpenAPI schema, not from a live response
Citations
- —The /v1/expand endpoint costs $0.003 per request via x402 (USDC on Base)https://think.agentutil.net
- —Free tier: 10 queries/day per IP, no signup requiredhttps://think.agentutil.net
- —No LLM in the service path — pure data lookup, sub-50ms latencyhttps://think.agentutil.net
- —OpenAPI spec describes expandInstruction with instruction (required) and domain (optional) fieldshttps://think.agentutil.net/openapi.json