Overview
ICME PreFlight is a policy enforcement layer for agentic AI that stops autonomous agents from taking actions you never authorized. As enterprises let AI agents send email, move money, modify files, and call external APIs, a single manipulated or misaligned action can carry real consequences. The problem with prompt-based safety checks, LLM judges, and reasoning chains is that the enforcement mechanism is itself a model, which means it can be jailbroken, prompt-injected, and reasoned around. PreFlight sits in front of every consequential action an agent attempts and checks it against your policy before it executes, so unsafe actions are blocked at runtime rather than flagged after the fact.
What makes PreFlight different is how policies are defined and enforced. You write your rules in plain English, such as "never send funds to an unverified wallet" or "never approve transactions over $10,000 without a second confirmation." PreFlight compiles those rules into formal logic (SMT-LIB), and an automated solver evaluates each proposed action against them, returning an allowed-or-denied decision in under a second. Because enforcement is grounded in formal verification rather than heuristics or model judgment, the same policy produces the same decision every time, and the guardrail cannot be talked out of its own rules.
Every decision also produces a zero-knowledge proof: a tamper-proof, independently verifiable cryptographic receipt that your security team, auditors, and compliance stakeholders can check without exposing the underlying data or policy internals. This removes the human from the enforcement loop without removing accountability, which matters when agents run hundreds of actions per minute and a manual review queue becomes a bottleneck rather than a safety net. For teams deploying agentic AI under real security, audit, and regulatory requirements, PreFlight turns agent governance from a matter of trust into something you can mathematically prove.
Highlights
- Jailbreak-proof runtime enforcement for agentic AI. PreFlight checks every consequential agent action against your policy before it executes and blocks unauthorized ones at runtime, so prompt injection and manipulation can't talk your guardrails out of their own rules.
- Plain-English policies, formal verification under the hood. Write rules in natural language and PreFlight compiles them to formal logic (SMT-LIB). An automated solver returns an allowed-or-denied decision in under a second, giving the same result every time.
- Zero-knowledge proofs your auditors can verify. Every decision produces a tamper-proof cryptographic receipt that security, audit, and compliance teams can independently verify without exposing your data or policy internals, turning agent governance into something you can prove.
Details
Features and programs
Financing for AWS Marketplace purchases
Pricing
Dimension | Description | Cost/unit |
|---|---|---|
Agentic Commerce x402 | Agentic Commerce meters this unit to bill buyers the seller's x402 quote price. | $1.00 |
Vendor refund policy
There are no refunds.
How can we make this page better?
Legal
Vendor terms and conditions
Content disclaimer
Delivery details
API-Based Agents & Tools
API-Based Agents and Tools integrate through standard web protocols. Your applications can make API calls to access agent capabilities and receive responses.
Additional details
Usage instructions
API
How to pay for this API
This API is sold pay-per-request through AWS Marketplace using x402. Charges are billed in USD to your AWS account under AWS Marketplace. To call it, set up Amazon Bedrock AgentCore payments in your AWS account (a Payment Manager and a payment session with a spend limit), which your agent uses to obtain the AWS Marketplace payment token.
Send your request to the endpoint below. The API returns HTTP 402 Payment Required with an AWS Marketplace x402 price quote (extra.quoteToken) in the PAYMENT-REQUIRED header. Pass that 402 response to AgentCore payments (ProcessPayment, or the AgentCore payments plugin for Strands/LangGraph), then retry the same request with the payment token in the PAYMENT-SIGNATURE header. A successful call returns 200 with your result, and the PAYMENT-RESPONSE header contains your AWS Marketplace payment ID.
How to call this API
ICME PreFlight checks a proposed AI-agent action against a policy you wrote in plain English and returns SAT (allowed) or UNSAT (blocked), with a verifiable zero-knowledge proof. Decisions are based on formal logic, so they cannot be jailbroken or prompt-injected.
Prerequisite - a compiled policy. Each check runs against a policy_id. Create one once by writing your rules in plain English at https://docs.icme.io ; you receive a reusable policy_id. You compile a policy once and reuse its policy_id on every call (not per request).
Method + URL POST <https://api.icme.io/agentpay/check>
Required headers Content-Type: application/json (The PAYMENT-SIGNATURE payment header is added automatically by AgentCore payments, as described above.)
Body parameters
- policy_id (string, required) - a compiled policy's ID (UUID), from https://docs.icme.io
- action (string, required) - plain-English description of the action to check
Example request
curl -s -X POST <https://api.icme.io/agentpay/check> \ -H 'Content-Type: application/json' \ -d '{ "policy_id": "530af065-7225-4f46-8fa9-94f0a53fef31", "action": "Release a payment of 500 to the freelancer. No dispute is active and 3 days have passed since deposit." }'Example response (200 OK)
{ "check_id": "5faf6a70-525f-4b7a-938f-a62a20d21507", "result": "SAT", "detail": "Satisfiable", "verification_time_ms": 2393, "proof_id": "51c5a5a8-bbb2-44aa-8abe-387f11397d5c", "proof_url": "https://api.icme.io/v1/proof/51c5a5a8-bbb2-44aa-8abe-387f11397d5c" }result: SAT = the action complies with the policy (allowed); UNSAT = it violates the policy (blocked). Actions the solver cannot map to the policy fail closed (blocked by default). proof_id / proof_url: a verifiable zero-knowledge proof of the decision.
Support
Vendor support
Email us at help@icme.io for support.
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.