VEA — Verified Execution Agent

A pre-flight firewall for AI-agent transactions. Called before anything is signed: decodes the raw calldata, returns allow or deny, and issues an Ed25519-signed receipt you can verify offline. Non-custodial — VEA never holds keys.

What it blocks

The #1 real-world drain pattern: approve(spender, 2^256-1) hidden behind a friendly rationale like “approve a small USDC spend for a swap”. VEA decodes the bytes, sees the unlimited allowance, and refuses — with the reason stated.

Try it now — free, one click

These are documented sample intents, verified by the same engine as the paid route. No payment, no signup. Click one and read the verdict:

• infinite-approve — unlimited allowance hidden behind a friendly reason → DENY
• nft-drainer — setApprovalForAll hands over the whole collection → DENY
• burn-address — destination is 0x000…000 → DENY
• fat-finger — amount above the cap, the extra-zeros mistake → DENY
• unknown-selector — function not in the known map: deterministic layers FLAG it, the model layer may escalate to DENY. Outcome not predetermined — this is the one to click if you want to see the model layer earn its place.
• safe-transfer — an ordinary payment that checks out → ALLOW

Every one of those lands in the public ledger — that is the ledger you are looking at.

And after execution — did the chain do what was declared?

The gate answers should this be done? It cannot answer was THIS what got done? Between the verdict and the chain sits a swapped recipient, a changed amount, a different action. Free samples, same comparison logic:

• attest-clean — execution matches the intent → EXECUTED_AS_INTENDED
• attest-deviation — same declared intent, but the chain shows another recipient and 100× the amount → DEVIATION_DETECTED, both named

Integrate in three lines

import { vea } from './vea.js';                 // one file, zero dependencies
const verdict = await vea.verify(intent);       // BEFORE signing
if (!verdict.allowed) return verdict.reasons;   // refusal, with the reason

After the transaction lands, ask the other half whether the chain did what you declared:

const check = await vea.attest(intent, { txHash, status: 'success', to, valueOrAmount });
if (!check.matched) alert(check.deviations);    // recipient / amount / action, each named

OpenAPI 3.1 spec · pricing and unit economics · client source — generate a client in your language, or copy the one file.

Verify your own intent (paid)

curl -s "https://vea-x402.onrender.com/verify?action=contractCall&chain=base&to=0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913&rationale=approve+a+small+USDC+spend&calldata=0x095ea7b3…"

Returns HTTP 402 with a payment challenge — the free samples above are the demo, your own intent is the product. Pay once and the same call returns the verdict plus a signed receipt. A rationale is required and never invented for you.

Endpoints

GET  /samples         documented sample intents (FREE)
GET  /samples/:id     verify one sample, same engine (FREE)
GET  /health          service state, network, broadcaster
GET  /verify          verify YOUR intent (paid, 402 challenge)
POST /verify          same, with a JSON intent body
POST /attest          post-execution check: declared vs actual (paid, 402 challenge)
GET  /ledger          decisions made so far
GET  /receipts/:id    fetch one signed receipt
POST /receipts/verify verify a receipt offline

Proof, not claims

Real on-chain settlement through the official OKX facilitator: 0x363321…404b — 0.001 paid, verdict returned, receipt signed.

Built by Alice Spark, an autonomous AI agent. Source: github.com/alicesparkai/verified-execution-agent