What are AI receipts?

An AI receipt is a signed record of declared fields about an inference, agent action, model call, or payment. The relying party must use the verifier and trust key named by the selected receipt profile. It is not proof that the declared action occurred unless the receipt profile and external evidence establish that fact.

What are agent receipts?

Agent receipts turn each route decision, tool call, model choice, and action an AI agent takes into a signed receipt chain. That's verifiable proof of what an autonomous agent actually did, for provenance and accountability.

What are signed model calls and model receipt verification?

A signed model call is a model inference wrapped in a receipt that declares provider, model, prompt hash, response hash, and timing under a signature. Verification checks the signature and payload under the named profile. Offline checking requires the matching verifier, trust key, receipt, and retained resources.

What are x402 AI payments and AI payment receipts?

x402 AI payments let an agent pay per action in USDC on Base using the HTTP 402 payment_required flow. Each settlement emits an AI payment receipt: signed proof of what was paid, for what action, on which rail.

What is an AI audit trail?

An AI audit trail is the chain of signed receipts an AI system leaves behind: inference, agent actions, model calls, and payments, all pulled into an exportable record that an auditor, partner, or regulator can check again independently.

The proof layer for AI systems

Signed records for AI requests
and actions.

Hive receipt APIs create signed records for supported AI and payment integrations. A receipt identifies the fields its issuer signed. Verification depends on the receipt format, supported algorithm, and trusted key. Check the endpoint documentation for the profile your integration uses. Hive signs the call metadata it records. Providers do not sign the model output shown here. A direct provider path and a routed path have different custody boundaries; a signed model identifier remains an assertion unless separate evidence establishes execution.

A log says “trust us, this happened.” An AI receipt says “check it yourself.” Hive attests what the AI did: the conditions and the action, and nothing more. It never claims the output was correct.

Run a free model receipt → Start Receipting Now → See agent receipts live →

Receipt types and integrations

Find the receipt type or integration you need. Each link opens its product page or technical explanation.

Thread 01

AI receipts

A portable signed record of the fields captured by a supported integration. Check the endpoint’s algorithm and verification requirements.

Cryptographic AI receipts → Thread 02

Agent receipts

Record selected agent routing decisions, tool calls, and actions in a linked receipt chain. Coverage depends on the integrated capture points.

Agent receipts, live → Thread 03

x402 AI payments

Agents pay per action in USDC on Base through the HTTP 402 payment_required flow. Settlement is live today, and every payment emits a signed receipt.

x402 checkout → Thread 04

AI payment receipts

Record a payment’s declared amount, purpose, rail, and transaction reference. Verify settlement against the relevant payment record.

Emit a payment receipt → Thread 05

Model receipt verification

Check the signature and declared fields on a model call under its named receipt profile. Offline checking requires the matching verifier, trust key, receipt, and retained resources.

Verify a model receipt → Thread 06

Agent provenance

Bind the recorded agent, tool, model, or human identifiers to signed fields. Verification establishes the signed statement; identity assurance depends on the evidence behind those identifiers.

Verify surface → Thread 07

AI audit trail

The chain of signed receipts an AI system leaves behind, assembled into an exportable, independently re-verifiable record for an auditor, partner, or regulator.

Assurance postures → Thread 08

Cryptographic AI receipts

Selected receipt profiles can use ML-DSA-65 under NIST FIPS 204. Verification requires the matching verifier, trust key, and retained resources.

The cryptography → Thread 09

Verifiable AI agents

Agents whose every action is receipted by default. Choose the agent, the receipt level, the fleet size, and run: from one agent to a hundred.

Agent fleet → Thread 10

Signed model calls

Prompt through Hive to a model and get the response plus a signed receipt for the call path. Mistral Leanstral is live today; the adapter layer targets more providers.

Signed model calls → Thread 11

AI action receipts

This demonstration shows selected routes, tool calls, and decisions with its named receipt profile. The receipt establishes only the signed fields and checks supported by that verifier.

Watch actions receipt → Thread 12

Agent payment proof

On-rail proof that an agent paid for what it used: x402 settlement in USDC on Base, with each task receipted. Usage you can prove, not just a bill you have to trust.

Agent payment proof → Thread 13

MCP receipts

When an MCP-compatible agent operates a platform through its tools, each operation becomes a signed receipt: which agent, which tool, against which workspace.

Receipt every agent operation → Thread 14

API agent receipts

Every API operation an agent performs, each model, tool, and approval stop on the route graph, gets captured as a signed route receipt.

SmartAgent route graph →

Try a supported receipt flow

Every page below is real and running, not a mock. Start with a free model receipt, pay and prove with x402, or run an agent trip.

live · free demoModel receipts/model-receipts/Signed proof for every model call: provider, model, prompt/response hashes, timing, and a signature that verifies in-browser. live · x402x402 checkout/x402-checkout/Fetch a live USDC quote on Base, prepare a wallet payment, run a signed receipt through the payment_required flow, and provision. live · receipt chainAgent trip/agent-trip/Seven route decisions stream in, each minted as a real ML-DSA-65 agent receipt and verified in your browser. Agent fleet/agent-fleet/Verifiable AI agents by the fleet. Choose the agent, the receipt level, the fleet size, and run. Start with one, scale to one hundred. live · relayReceipt Relay/receipt-relay/Receipt any external action, API call, agent step, payment claim, or chain transaction, with honest evidence tiers so a receipt never claims more than it can prove. Hive Ledger/hive-ledger/The searchable index for every signed receipt: discovery, search, and audit. The ledger is convenience. The receipt is the proof. Payment Healing/payment-healing/Sorts out rail-specific recovery for money movement: pre-broadcast stop or revoke, escrow or quarantine hold, issuer or custodian freeze where integrated, counter-transfer, or evidence-only after settlement. Verify/verify/Check the supported Ed25519 mesh receipt and countersignature forms. Typed ML-DSA and hybrid envelopes require their profile-specific verifier. Activate/activate/Request a tk_live activation key to meter fleet usage and scope a pilot across your own AI-receipt workloads.

Here's the receipted money-movement loop: Receipt Relay signs the event, Hive Ledger searches, exports, and checks it, and Payment Healing sorts out rail-specific recovery. Hive proves what happened early enough to act on it where the rail allows, and proves what happened after the fact when it doesn't. The Circle/USDC-compatible receipt schema is live today. Direct Circle integration comes next.

Request an AI payment quote

This hits Hive’s public, no-login x402 quote endpoint and returns a live settlement envelope with an ML-DSA-65 (FIPS 204) signed quote receipt. No key, no secret, nothing to install.

POST · https://receipts.thehiveryiq.com/v1/x402/quote

curl -sS -X POST https://receipts.thehiveryiq.com/v1/x402/quote \
  -H 'Content-Type: application/json' \
  -d '{"agent_did":"did:example:agent","profile":"nano"}'

Read the technical articles

Field notes and explainers on the proof layer for AI, from the Hive blog.

What cryptographic AI receipts are, and why inference needs proofThe base object: signed proof for AI inference that you can check offline. AI agent receipts: verifiable proof of what an agent actually didRoute decisions, tool calls, and model choices as a signed receipt chain. x402 agent payments: receipts for agentic commerceAgent payment proof on the HTTP 402 rail, in USDC on Base. Human-origin attestation: third-party proof of expert-data provenanceProvenance for the human data behind AI systems. AI assurance for regulated workflowsFinance, healthcare, legal, government: the AI audit trail as evidence. The Hive blog →Comment letters, position papers, and engineering field notes.

Signed records for AI requests and actions.