Skip to main content
AGENT-FIRST INFRASTRUCTURE

Originary & AI

Built for agents, not browsers

Every agent call emits a verifiable PEAC-Receipt. Agent-to-agent (A2A) messages carry proof, payments, and preferences by default.

Works with A2A networks, MCP tool calls, and HTTP-402 payment rails out of the box

Agent-native trust

Responses ship with a signed PEAC-Receipt so downstream agents can verify who did what, when.

A2A-ready

Receipts attach to inter-agent messages, enabling zero-trust hand-offs without brittle webhooks.

Preference-aware

We snapshot AIPREF at call time so usage rights are auditable, not "best effort."

Pay-as-you-go rails

Built-in HTTP-402 flows (L402/LSAT, x402, Stripe) so agents can pay and proceed - no accounts, no PDFs.

Provenance-friendly

Receipts can reference C2PA manifests to keep media and model outputs traceable end-to-end.

Why “agent-first” matters

Most AI pages talk to humans. Originary AI talks to agents.

Your graph may include solvers, crawlers, planners, retrievers, and buyers - often from different vendors. We make those hops safe by stapling a cryptographic receipt to each hop.

Any party can verify it offline using your published keys.

How it works

1

Challenge/Access (optional)

If a resource is gated, we answer with HTTP 402 (Payment Required). Agents respond via L402 (Lightning), x402 (stablecoin), or Stripe.

2

Execute + Snapshot

We run the call and snapshot effective AI-usage preferences (AIPREF) at that moment.

3

Issue a Receipt

We sign a compact JWS (PEAC-Receipt) that binds the result, enforcement method, payment evidence, AIPREF snapshot, timestamp, and key id. It travels with the response - and can be forwarded A2A.

For agent-to-agent (A2A) networks

Attach receipts to messages

Include the JWS token alongside your payload so the next agent can verify provenance and policy.

Trust decisions become code

"Process only if receipt.valid && receipt.aipref.train_ai == false" is now a one-liner, not a policy memo.

Replay-resistant

Receipts are content-bound (hashes) and timestamped; tampering breaks verification.

For MCP & tool calls

Per-tool receipts

Wrap every MCP tool invocation with a receipt (tool id, param hash, issued_at).

Privacy-first

Hash sensitive params; keep the proof, not the secrets.

Drop-in

Keep your tool schema; we add the proof layer.

Learn how to integrate MCP with PEAC-Receipts

View MCP Documentation

Payments that agents understand

L402 / LSAT

Invoice + macaroon; return with preimage → access granted.

x402

Stablecoin-backed HTTP-402 for instant settlement in USDC.

Stripe

Classic cards behind 402 challenges for human-in-the-loop agents.

All rails normalize to a single payment{ rail, amount, currency, evidence } in the receipt.

What you can build

Concrete patterns for agent infrastructure

Proof-carrying crawlers

Your crawler pays via 402, collects data + receipts, and forwards both to retrievers.

Verifiable purchases

Agents buy datasets/API calls; downstream analytics accept only inputs with valid receipts.

Compliant fine-tuning

Filter corpora by AIPREF in receipts so training jobs don't ingest "no-train" content.

B2B agent marketplaces

Sellers require receipts on every consumption event; buyers verify before paying invoices.

Dev notes

Response header:
PEAC-Receipt: <compact JWS>
Discovery:
/.well-known/peac.txt
Verification:
Offline with your keys or POST the token to /verify
Streaming:
Use HTTP/2 trailers to deliver the receipt at end of stream

Ship agent-grade trust in a day, not a quarter

Every agent step. One verifiable receipt.