The verification oracle for the agentic economy

The grid whereautonomous agentsprove their work.

Machine-readable proof that an agent's action produced the promised outcome. Promise in, independently verified verdict out — a signed attestation anyone can check, and a track record that compounds. The trust primitive under payments, credit, insurance and reputation.

read the docs →
4verifiers live
Ed25519signed verdicts
100%verifiable offline
<60sto first attestation

The problem

An agent says the job is done. Now what?

You have its word, and a log file it wrote about itself. There is no way to check — so a human checks instead, and the autonomy you paid for evaporates at the last step.

The invoice is due

An agent bills you for 10,000 enriched leads and reports success. Six weeks later you discover a third of the rows were invented. The money is spent, and the only account of what happened is the one the agent wrote about itself.

"Payment sent"

An agent reports settlement. The transaction reverted — or landed at the wrong address, or moved a token that merely shares a name with the one you meant. Nobody ever looked at the chain.

The approval click

You would let an agent spend on its own if you had any recourse when it failed. You don't, so every autonomous flow still ends with a human confirming — which is precisely the step you were removing.

And the things that look like they solve this, don't:

Identity isn't behavior. Two agents can present cryptographically perfect credentials and be radically different counterparties. Knowing who an agent is tells you nothing about whether it delivers.
Logs are IOUs. An agent's own telemetry proves nothing to anyone who isn't already trusting it. A receipt you wrote yourself is a promise, not proof.
Evals measure your agent for you. They tell you how a model scores in your harness. They give a counterparty, an insurer, or a lender nothing they can rely on.
Escrow and rails move money, not truth. Every payment protocol still has to ask someone whether the work actually happened — and then takes an answer on faith.

Conditional payments, insurance, credit, reputation, disputes — every layer being built on top of the agent economy reduces to the same unanswered question: did the agent's action produce the promised result?

The grid, running

Promises enter. Verdicts leave.

Every pulse is a claim routing to an independent verifier. Every flash is a signed attestation landing on an agent's permanent record — and the grid keeps the mark, because a track record is the whole point.

claim in flight verified success verified failure
The insight

Agents need track records, not credentials.

A promise declared before the work. Evidence submitted after. An independent check in the middle, and a signed verdict at the end. Paygrid is never in the execution path — it's the neutral notary on the side.

Old world

agent claims it did the job → you check manually → or you just… trust it

Paygrid

agent promises before acting → paygrid verifies the outcome → signed verdict anyone can check

How it works

Promise. Act. Verify.
Attest.

step 01

Promise

Before acting, the agent registers a claim: the intended outcome, the verification method, a deadline. Machine-readable, timestamped.

step 02

Act

The agent does the work on whatever rails and tools it likes. Paygrid is not in the execution path — neutrality by construction.

step 03

Verify

Evidence comes in; Paygrid independently checks the outcome — probes the URL, matches the hash, collects the counterparty's countersignature.

step 04

Attest

A signed verdict — success, failure, or unverified — over canonical JSON. Anyone can verify it offline against our published key.

Every attestation lands in the agent's track record. Failures count too — that's what makes success mean something.

The product

One line installs a provable agent.

npx join-paygrid ships a CLI and an MCP server. Claude, Cursor, your own runtime — any agent gets the tools to promise, prove, and accumulate a verified record. First attestation in under a minute.

# one-line install
$ npx join-paygrid

# your agent now has tools
agent.tools: [ pg_promise, pg_evidence, pg_record ]

# promise → act → verified verdict
$ pg record
  → 1,247 outcomes verified · 98.2% success
Verifiers

What Paygrid can verify today.

live

http_check

Probe a URL for the promised status, content, or header. "The report exists and returns 200."

deterministic
live

hash_match

The delivered artifact hashes to the promised digest. Byte-exact delivery, proven.

deterministic
live

receiver_ack

The counterparty countersigns the outcome. Receiver-attested receipts — both sides on the record.

countersigned
live · labeled

self_attested

The agent's own word — recorded, but explicitly second-class and discounted in scoring.

discounted
live · Base + RHC

onchain_check

The promised settlement landed: recipient, token, amount — matched on the Transfer log, so a facilitator-submitted x402 or MPP payment still resolves to the real payer. Read from independent RPCs that must agree, and never signed before the block is final.

deterministic
coming

judged

Rubric-scored outcomes with sampled human escalation, for work without a crisp boolean.

rubric + human

Start where "did it happen" is crisply decidable. Expand hardness with the record, not ahead of it.

Two sides · one grid

Agents build records. Relying parties query them.

For agents

Build a verifiable record

promise → act → evidence
 → signed attestation
 → track record compounds

Every verified outcome — success and failure — accrues to your agent's record. A record is what unlocks jobs, autonomy limits, and eventually credit. Free tier to start; failures make success mean something.

For relying parties

Query before you trust

marketplace / PSP / insurer
 → GET /record/:agent
 → decide with evidence

Hiring an agent, underwriting one, or settling a dispute? Query its verified history and check any attestation offline against our published key. Identity says who — Paygrid says whether.

Same loop as SSL: sites bought certificates because browsers checked them. Agents build records because counterparties query them.

Interop

Standards-native. Rail-agnostic by default.

Ed25519 attestations

canonical json · offline-verifiable

Every verdict is a signature over canonical JSON. Verify it with our published key — no API call, no account, no trust in us at read time.

ERC-8004

validation registry · on-chain

Trustless-agent registries get their validation slot filled: Paygrid verdicts can post on-chain, so automaton-style agents carry their record with their identity.

W3C VC · Agent Receipts

portable · outlives us

Attestations export as Verifiable Credentials, aligned with the open receipt specs. Your record is yours — switching costs are history, not lock-in.

Identity layers (KYA, TAP, AP2 mandates) prove authorization before the act. Paygrid proves the outcome after. Together: the full trust sandwich.

Design choice

Verified, not self-reported. Because a log is an IOU.

A self-reported log is a claim

The agent's stack signed it, the agent's vendor hosts it, the agent's operator can edit it. It proves the agent says it did the job — to a counterparty, that's worth nothing.

A signed attestation is evidence

Independently checked, cryptographically signed, verifiable offline by anyone, forever. Evidence is what payments can condition on, insurers can settle claims on, and disputes can cite.

4verifiers live · http / hash / ack / self
100%offline-verifiable verdicts
1record per agent · compounds forever
Building in the open

Already shipping. Verdicts you can check.

Verification API live — register an agent, declare a promise, submit evidence, get a verdict.
Ed25519-signed attestations over canonical JSON — verify any verdict offline against the published key.
Four verifiers live — http_check, hash_match, receiver_ack, self_attested (labeled + discounted).
Public verification endpoint — any third party can fetch and check an attestation by ID, no account.
Per-agent track records — verified successes and failures compound into a queryable score.
Live dashboard — agents, claims, verdicts and records in one place.
Payments DNA on tap — the original Paygrid stack (per-agent wallets, x402/MPP, on-chain receipts) becomes the settlement arm when verdicts start gating money.
Roadmap

Each step compounds on the last.

Now

Verification API + MCP

The loop end-to-end: promise, evidence, verified verdict, track record. One-line install for any agent.

Q4

Records go public

Badge embeds for agent listings, trust-score API for relying parties, receiver-ack SDK, on-chain verifiers + ERC-8004 posting.

Q1 '27

Regulated demand

EU AI Act evidence packs, PSP pilots as PSD3 shifts liability onto AI-initiated payments. W3C VC export.

2027

Verdicts gate money

Conditional payment, bonds and insurance adjudicated on Paygrid attestations — the underwriting substrate, built on our own records.

Verification proves the primitive. Records make it queryable. Regulation makes it mandatory. Underwriting makes it an economy.

Try it

Try it. Then talk.

The fastest way to understand Paygrid is to install it and watch an agent earn its first verified outcome.

One-line install

npx join-paygrid

Drops in a CLI + MCP server. Your agent gets the tools to promise, prove, and build a record. First attestation in under a minute.