GPU datacenter
Open standard · Public registry · One resolver call

The price layer
of the agent
economy.

Every pricing page on the internet was written for humans. Ginkgo Protocol is an open, machine-readable feed standard + public registry — so an agent can answer "what does it cost, and how do I buy it" in a single HTTP call.

COMPLEMENTS x402 MCP-READY CC-BY-4.0 128 FEEDS · GPU VERTICAL
resolver.ginkgo.dev — live query LIVE
HTTP/2 · TLS1.3 — ms edge: iad-04
128
feeds indexed · gpu vertical
1.2M/day
resolver calls
41ms
median resolve latency
17verified
vendors cryptographically signed
01 Position in the stack

Discovery and comparison were
the missing layers. Not anymore.

The agent commerce value chain has three layers. Payment rails are solved. The two layers above them were empty — until now.

L1DISCOVER
Which services exist for this intent? Specs, limits, regions, SLA — as data, not marketing pages.
GINKGO ●
L2COMPARE
Which one should I buy given my budget and constraints? Deterministic ranking over live feeds.
GINKGO ●
L3SETTLE
How do I pay? Machine-native payment rails, already in production.
x402 · AP2 · Stripe AgentKit

// Ginkgo Protocol never touches money. We describe prices — settlement stays with the rails.

02 Why a standard

Everything else describes
the web. Nothing describes
what it costs.

Open standards with network effects are rare, and the agent economy has exactly one empty seat at that table. Whoever defines the price feed becomes its schema.org.

StandardDescribesBlind spot
schema.orgEntities, orgs, product metadataNo price, specs, or purchase action
OpenAPIAPI interfacesNo pricing or billing
x402 / AP2Payment railsNo discovery or comparison
agent.price.json Price · specs · limits · SLA · purchase —
03 The feed · 20 lines to adopt
.well-known/agent.price.json
{
  "$schema": "https://ginkgo.dev/schema/v1/feed.json",
  "publisher": { "id": "deepinfra", "domain": "deepinfra.com" },
  "services": [{
    "sku": "h100-ded",
    "name": "H100 SXM dedicated",
    "pricing": {
      "model": "per_gpu_hour",
      "unit_price_usd": 2.19,
      "spot_available": false
    },
    "specs": { "gpu_vram_gb": 80 },
    "sla": { "uptime_pct": 99.9 },
    "payment": { "methods": ["card", "x402"] },
    "purchase": { "action": "https://deepinfra.com/signup" }
  }]
}
One file, one convention A vendor drops this at /.well-known/agent.price.json — the registry crawls, validates and signs it. Twenty lines of JSON and any agent on earth can price you.

Publish once.
Be priced everywhere.

One static JSON file at a well-known location. No SDK, no lock-in, no revenue share on the standard itself. Agents resolve you; x402 settles the payment.

FREEStandard & registry access — forever, for everyone
VERIFIEDCryptographically signed feeds get ranking priority
OPENCC-BY-4.0 · governed by an independent foundation
04 Public registry · GPU inference vertical

128 feeds. Live prices.

Browse the registry →
VendorStatusH100 $/hrA100 $/hrRTX 4090SLAQueries 30d

// sorted by resolver query volume · prices are per-GPU-hour on-demand · updated 2h ago

05 For vendors

Your pricing page is invisible
to the buyers that matter next.

Agents don't read marketing pages. Every month we tell you how often agents looked you up, chose you — and exactly why they didn't.

STEP / 01

Publish or get crawled

Drop agent.price.json at your well-known path, or let our parser structure your public pricing page — human-reviewed before it ships.

STEP / 02

Get discovered

Your feed enters the resolver's ranking set. Agents with matching intent and budget see you — with your live price, not a stale screenshot.

STEP / 03

Win the query

The monthly vendor report: queried ×, selected ×, lost because Z. Claim your feed free — verify it for ranking priority.

06 Vertical rollout

One vertical, done to monopoly.
Then the rest.

NOW · 2026 Q4
GPU Inference
● live · 128 feeds
Q1 2027
LLM APIs
self-supply 100
Q2 2027
Vector DBs
claim loop
Q3 2027
Storage / CDN
claim loop
Q4 2027
Data APIs
claim loop
2028
SaaS Tools
match engine
GPU macro
Start here

Twenty lines of JSON.
Every agent can price you.

Adopt the spec → Query the registry