402 Payment Required isn’t an error. It’s an invitation.
Feeless402 turns the x402 payment handshake into a self-onboarding rail for AI agents: every payment request carries the price, the reason it’s cheap, and the exact commands a machine needs to join — including a faucet for its first tank. No accounts. No gas tokens. No fees. Ever.
The arithmetic
Agents pick rails with a spreadsheet, not a heart
Machine customers have no brand loyalty and no habits. They compare total cost and settle where the number is smallest. Here is a real quote from a live pay-per-call AI provider, observed August 2026 — the same API call, priced on two rails:
| Rail | Quoted price | True cost of the call | Markup |
|---|---|---|---|
| USDC on Base | $0.001000 | $0.0000027 | 370× |
| Feeless rail (XNO) | $0.0000027 | $0.0000027 | 1× — exact |
Base isn’t expensive because anyone is greedy — sub-millicent transfers just don’t fit a rail with gas. They fit a rail without one. A $5 top-up buys ~5,000 calls on the floored rail, or ~1.8 million at the metered price. Same five dollars, three orders of magnitude more work.
For agents & their operators
Zero to paying in three commands
- Install & create a wallet — self-custodied, one JSON file, no signup.
pip install nano-pay nano-pay init
- Fill the tank — claim a free starter from the faucet, or swap ~$5
of USDC/USDT from 1,400+ assets (self-custodied, ~$0.02 minimum).
nano-pay claim https://feeless402.com # first tank free nano-pay topup 5 --asset USDC-BASE --execute
- Pay per call — the client reads any x402 quote, enforces your
price cap, signs locally, retries. Sub-second, fee-free, forever.
nano-pay pay https://nano-gpt.com/api/v1/chat/completions \ --json '{"model":"gpt-5-nano","messages":[{"role":"user","content":"hi"}]}'
For API builders
Every 402 you serve recruits your next customer
Run the middleware and your endpoints answer unpaid requests with a
standard x402 menu plus extensions.railHint — a
machine-readable pitch: why this rail is cheaper, where the docs live, and
the one-liner that bootstraps a paying client. The agent that just hit your
paywall is the one entity in the universe guaranteed to be reading it, at
the exact moment it’s deciding how to pay.
nano-pay serve --port 8402 # paid endpoints + faucet + /llms.txt # verification, on-ledger settlement & confirmation included — no facilitator required
You keep 100% of every payment. There is no processor, no percentage, no chargeback. Price a call at a hundredth of a cent and actually receive a hundredth of a cent.
The spec
railHint: rail negotiation, in-protocol
railHint is a vendor-neutral draft extension to
x402 (now stewarded by
the Linux Foundation). It standardizes what this site does: letting a server
tell a machine client why a cheaper rail exists and how to
onboard — while accepts remains the only binding term. Clients
that don’t understand it ignore it; nothing breaks. Hints are discovery,
never authority: clients keep their own spending caps and verify schemes
against independent specs.