Agent Service Provider · OKX.AI

Know who your agent pays before it pays.

Agent commerce moves before reputation can form. Vera Trust turns a stranger's endpoint, contract, or wallet into a deterministic verdict. Every finding is attached to its source.

Inspect a counterparty Browse 44 signals

The exposure

Commerce moves faster than reputation forms.

A marketplace listing is only an assertion. The moment an agent can move money, that assertion needs a testable record behind it.

01 / CLAIM

A listing is only a claim.

An advertised refund tool may not exist. A payout address may not be the address an agent was shown.

02 / CODE

A contract can change.

Upgradeable proxies can switch logic after payment; privileged owners can pause, mint, or reclaim.

03 / HISTORY

New markets have no memory.

Reputation takes time. Evidence is available now—and can be evaluated again later.

The decision boundary

The model does not decide.

Checks are deterministic lookups and static analysis. A model may explain the sealed result, but it cannot alter it. That is why a receipt means something. Anyone can re-run the evidence and reproduce the verdict.

Collect factscontract · address · live MCP probe

Map each fact to a trust signal44 codes · v2026-07-16

Score by policy, not by impressionseverity × confidence

Seal a replayable receiptre-derive it yourself

Agent surface

Call the evidence, not an opinion.

The signal catalogue is public. A paid verdict costs $0.01, settled per call over x402 on X Layer.

POST /check

The request

POST /check
{ "kind": "asp",
  "endpoint": "https://some-asp.example/mcp",
  "claimedTools": ["get_quote", "send_payment", "request_refund"] }

402 payment required

Unpaid → challenge

HTTP 402 Payment Required
{ "x402Version": 2, "accepts": [{
    "scheme": "exact", "network": "eip155:196",
    "amount": "10000", "payTo": "0x354a…2b04" }] }

200 sealed receipt

The verdict

HTTP 200
{ "verdict": "block", "risk": 45,
  "findings": [{ "code": "ASP_PAYTO_MISMATCH", "source": "x402-probe" }],
  "receipt": { "hash": "874ffb2a…e944", "replay": "/verify/874ffb2a…" } }

Questions

What agents — and the people who build them — ask.

Direct answers about what the check does, what it costs, and what it will not claim.

What does Vera Trust actually check?+

For a service: it triggers the endpoint's own x402 challenge. It compares the live payment destination, network, and amount against what you expected. It also compares the live MCP tools against what the listing advertises. For a contract or wallet: public attributes such as proxy upgradeability, mint and pause authority, and ownership. Technical signals only — not financial or legal advice.

Does a language model decide the verdict?+

No. Every check is a deterministic lookup or static analysis. A model may write the plain-English summary afterwards. It cannot change the verdict. That is the only reason the result is reproducible.

What do I get back?+

A verdict, a risk score, and a list of findings. Each finding cites the upstream response it was read from, with a digest of those bytes. All of it is sealed inside a signed, replayable receipt.

Why should I trust a result that comes from you?+

You don't have to. Re-derive any receipt's hash in your own browser at /verify — we are not consulted. The ledger tip is also committed on-chain to X Layer, so we cannot quietly rewrite past results.

What does it cost?+

The signal catalogue is public and free. A paid verdict is $0.01, settled per call over x402 on X Layer in USDT0 — no account, no subscription.

What will it not tell me?+

It reports what is observable now, not a guarantee of future behaviour — an endpoint can change after you check it. It is not AML, sanctions, investment, or legal advice. When something can't be verified, it says so rather than guessing.

Checks remember what changed. Call again later and Vera Trust compares structured evidence instead of repeating itself: a new owner, an upgraded proxy, or a tool that appeared since your last check becomes a meaningful diff.