agent-native AI infrastructure

AI microservices that
charge machines to pay

Payment-gated inference endpoints over the x402 protocol. No accounts. No API keys. No KYC. Agents send a signed USDC authorization, get a result. That's the entire flow.

GET /health View on GitHub
status ok
embed model nomic-embed-text
inference model qwen3.5-nothink:2b
pricing $0.01–$0.04
network Base

endpoints

Three endpoints. Pay per request.

Each request is individually priced and settled on-chain. No subscription, no rate tiers, no session state.

POST
/api/embed
Text embedding — 768-dimensional vectors via nomic-embed-text. ~370ms warm.
floor price $0.01
POST
/api/summarize
Text summarization via qwen3.5-nothink:2b. ~8s warm.
floor price $0.02
POST
/api/ask
Single-turn Q&A via qwen3.5-nothink:2b. ~7s warm.
floor price $0.04
GET
/health
Service status, live model info, and current pricing. No payment required.
cost free

payment flow

How x402 works

HTTP 402 has been reserved since 1996. The x402 protocol finally makes it practical — a machine-readable payment handshake at the HTTP layer.

01
Request
Agent sends request to endpoint. No credentials attached.
→
02
402 Response
Server returns 402 with machine-readable payment instructions: price, network, recipient.
→
03
Sign
Agent signs an EIP-3009 USDC transfer authorization for the exact amount.
→
04
Settle
Coinbase CDP facilitator verifies and settles on-chain. Server confirms.
→
05
Response
200 OK. Inference result returned. No session, no state.

pricing

Fixed prices per request.

Each endpoint has a fixed floor price. Check GET /health for current prices at any time.

/api/embed
$0.01
nomic-embed-text · 768-dim
/api/summarize
$0.02
qwen3.5-nothink:2b
/api/ask
$0.04
qwen3.5-nothink:2b
Prices are fixed per endpoint. Running on Base mainnet.

the thesis

HTTP 402 has been reserved since 1996. The x402 protocol finally makes it practical: a server returns a 402 with machine-readable payment instructions, the client signs an EIP-3009 authorization over the exact amount, a facilitator settles it on-chain. The server only responds after settlement is confirmed.

The result is a fully permissionless API economy — no OAuth, no billing dashboards, no rate-limit tiers. The x402 protocol is now backed by Google, AWS, Anthropic, Stripe, Visa, Mastercard, and Cloudflare as Linux Foundation partners.

x402ai is early infrastructure proving what agent-native commerce looks like in practice.

architecture

Built on open infrastructure

Self-hosted on a home server, served globally via Cloudflare Zero Trust tunnel.

payment
·x402 protocol
·Base (USDC)
·Coinbase CDP facilitator
·EIP-3009
inference
·Ollama runtime
·nomic-embed-text
·qwen3.5-nothink:2b
server
·Hono (Node.js)
·x402-hono middleware
·pm2 process manager
infrastructure
·Cloudflare Zero Trust
·Dell Precision / i7-6700
·Ubuntu 24