Signed agent-to-agent commerce — across four regulated verticals.

What is A2A — the open standard NVIDIA pushes heavily

A2A (Agent-to-Agent) is an open protocol — originally authored at Google, now actively advanced by NVIDIA — that standardises how autonomous AI agents discover each other, exchange messages, and delegate tasks. The A2A project on GitHub defines the canonical Task and Message types, a gRPC/HTTP service, and a streaming interface for real-time agent coordination.

NVIDIA has adopted A2A as the coordination layer across its AI Blueprint reference implementations — including the AI Virtual Assistant, RAG blueprint, and Retail Agentic Commerce blueprint. When one NVIDIA blueprint agent sends a Task to another, A2A is the wire format. The gap A2A leaves open: there is no signed receipt. An agent can lie about what it sent, when it sent it, or whether it sent it at all.

NVIDIA A2A blueprint reference implementations

Hive ships drop-in adapters for three NVIDIA blueprint repos. Each adapter is a single @hive_signed decorator placed on the agent's primary method.

How Hive plugs in — the @hive_signed decorator

The @hive_signed decorator wraps any NVIDIA blueprint agent method. It hashes the input payload (SHA-256), mints a pre-call Hive receipt via POST https://thehiveryiq.com/api/receipts, waits for the receipt to be anchored on Base 8453, then proceeds with the original method call. The receipt ID and Base explorer URL are attached to the response.

# Original NVIDIA RAG blueprint — src/nvidia_rag/rag_server/server.py
from fastapi import APIRouter
from nvidia_rag.rag_server.main import NvidiaRAG
+from nvidia_blueprint_adapter import hive_signed            # ← line 1

@router.post("/generate")
+@hive_signed(                                               # ← line 2
+    blueprint_id="nvidia-rag",                              # ← line 3
+    agent_id="did:hive:agent:rag-retriever",                # ← line 4
+)                                                           # ← line 5
async def generate(request: RagRequest, rag: NvidiaRAG):
    return await rag.aquery(request.messages, request.use_knowledge_base)

The same 5-line pattern applies to the AI Virtual Assistant blueprint's /generate handler and to Retail Agentic Commerce's NAT @register_function components. See nvidia_blueprint_adapter.py for all three diffs.

The middleware approach

For deeper integration — where you want receipt-gating at the task level before any agent call is forwarded — use HiveA2AMiddleware. The middleware wraps A2A's canonical Task and Message types with Hive receipt fields, mints the receipt, and only yields the signed task to the receiving agent after the anchor is confirmed.

from hive_a2a_middleware import HiveA2AMiddleware, A2ATask, A2AMessage

# Initialise once — pick your NVIDIA blueprint ID
middleware = HiveA2AMiddleware(blueprint_id="nvidia-rag")

# Every send is gated on a Hive receipt
async with middleware.send(
    task=my_task,
    sender_id="did:hive:agent:pricing-001",
    receiver_id="did:hive:agent:rag-retriever-001",
) as signed_task:
    response = await nvidia_rag_agent.process(signed_task)
    await middleware.verify_inbound_task(response)

Live demo

See two NVIDIA-style agents — Pricing and Inventory — exchange A2A messages, each producing a Hive receipt with Base 8453 anchor. The demo runs in your browser against live Hive endpoints.

What an NVIDIA A2A blueprint receipt looks like

Every receipt carries NVIDIA blueprint attribution in the nvidia_blueprint_id field. The envelope is CBOR-canonical, JSON-rendered here for inspection.

Pricing — NVIDIA A2A blueprint tier

One price across every NVIDIA A2A blueprint integration. Settlement is USDC on Base 8453 via x402. Treasury 0x15184bf50b3d3f52b60434f8942b7d52f2eb436e is on-chain.

Full pricing ladder on the pricing page including monthly flat tiers for high-volume NVIDIA blueprint deployments.

Integration steps for any NVIDIA blueprint engineer