> For the complete documentation index, see [llms.txt](https://docs.icme.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.icme.io/home.md).
# Home
Welcome to your team’s developer platform
AI Agent Actions Anyone Can Verify
Your logs only prove what you say happened. Today, showing that an AI agent followed policy means asking an auditor, customer, or counterparty to trust your logs, your vendor, or your model. PreFlight checks each agent action against your written policy before it runs and issues a cryptographic receipt. Anyone can verify that receipt on their own, without seeing your policy or your data.
***
Introducing ICME's Preflight
{% columns %}
{% column %}
### Logs Ask Others to Trust You
Most AI controls produce records that only the operator can vouch for: logs, dashboards, traces, and an LLM judge's explanation. When a regulator, auditor, customer, or counterparty asks whether an agent followed the rules, the answer is your word plus your logs. Logs can be edited, and models change their answers. Neither is evidence an outsider can check.
AI agents make this harder. The same prompt can produce different actions, so replaying the model does not show what it decided or why. For regulated industries, and for anyone accepting transactions an agent started, "the model said so" is not an answer. "Our logs say so" is not much better.
PreFlight gives every agent action a receipt. The policy is compiled into formal logic, a solver checks the action before it runs, and the result is wrapped in a zero-knowledge proof. The receipt is tamper-evident, independently verifiable, and shareable with any third party or agent. It proves the policy was applied to the action as submitted. It does not prove that the facts in the action are true..
{% endcolumn %}
{% column %}
#### If someone outside your company needs to trust what your agent did, this is how you prove it.
#### What a receipt gives you that a log cannot
1. **Independent verification.** Anyone can verify a receipt in under a second, without trusting you or ICME.
2. **Privacy.** The verifier learns the result, not your policy, your prompts, or your data.
3. **Portability.** Download the proof when it is generated and it travels with the action, to an auditor, a regulator, a court, a counterparty, or another agent.
{% endcolumn %}
{% endcolumns %}
***
{% columns %}
{% column %}
### Get started in 4 steps
**Write your agent policy in plain English.**\
Example: *"Never issue a refund over $500 without manager approval." "Never share a confidential document outside the firm."*
**PreFlight compiles it to formal logic.**\
Your policy is translated to SMT-LIB and stored. No prompt engineering required.
**Check each agent action before it runs.**\
Send the proposed action to the API. You get back SAT (allowed) or UNSAT (blocked).
**Share the receipt.**\
Every decision, allowed or blocked, comes with a cryptographic proof. Hand it to whoever needs to trust the decision: your auditor, your customer, or the agent on the other side of the transaction..
Get startedAPI reference
{% endcolumn %}
{% column %}
{% code title="In your terminal." overflow="wrap" %}
```bash
// Sample call.
curl -s -X POST https://api.icme.io/v1/verifyPaid \
-H 'Content-Type: application/json' \
-d '{
"policy_id": "f6e3cd15-9e28-45c4-9f4c-683edd63e468",
"action": "Send 1000 USDC to an unknown wallet."
}' | jq .
// JSON response after agent pays.
{
"check_id": "90484dac-f41e-47ee-9758-a2f63a4900b7",
"action": "BLOCKED"
}
```
{% endcode %}
{% endcolumn %}
{% endcolumns %}
No human-in-the-loop required
Human review works at low volume and risk. It breaks at scale.
AI agents already buy inventory, execute trades, book services, issue refunds, and negotiate contracts at machine speed. When an agent takes hundreds of actions a minute, a review queue is a bottleneck. When the reviewer approves faster than they can read, it isn’t oversight at all.
PreFlight checks every action against your policy before it runs and issues a receipt for each one. Your team, your auditors, and your counterparties can verify any decision afterward without re-running it or seeing your policy. People stay accountable for the rules. The receipts show the rules were applied.
***
{% columns %}
{% column width="50%" %}
{% endcolumn %}
{% column width="50%" %}
### Want to go deeper?
Our blog covers real agent failures, why checks that rely on a model’s judgment can be talked out of a decision, and how teams in finance, legal, and payments use verifiable receipts to answer auditors and regulators.
If your agents move money, touch sensitive data, or act on someone else’s behalf, and you need to show an outsider what they were allowed to do, start here.
ICME Labs builds the infrastructure that makes AI agent actions verifiable by anyone, not just the company running them.
Blog
{% endcolumn %}
{% endcolumns %}
Join a community of builders making guardrails for agents that do not fail.
:github:
GitHub
Fast zkVM born at a16z Crypto substantially adapted by ICME Labs (NovaNet) for verifiable machine learning. ⚡
We use zero knowledge machine learning for agentic guardrails.
