Every payment online needs permission. A card. An account. A subscription. A country the seller is willing to serve. People get stuck behind those walls every day, and the fastest-growing use
Every payment online needs permission. A card. An account. A subscription. A country the seller is willing to serve. People get stuck behind those walls every day, and the fastest-growing users of the web, AI agents, can't get through them at all. An agent can't open a bank account.
Today we're launching two things that take the permission out of paying:
x402 Work, an open payment scheme for HTTP 402. Instead of sending money from a wallet, the client pays for each request with a few seconds of proof of work.
Hashference, the first service on x402 Work: 45 live AI models through OpenAI- and Anthropic-compatible APIs, paid for by mining, with QUAI or Qi, or one request at a time with work.
Both are built by Dominant Strategies, the team behind Quai Network, and both run on Quai, where energy becomes money.
HTTP has had a status code for payments since the 1990s: 402 Payment Required. For most of the web's history nobody used it, because there was no native way to pay inside a request. The x402 standard revived it: a server answers 402 with a price, the client pays, and the client retries the same request.
Standard x402 pays with a stablecoin transfer from a funded crypto wallet. x402 Work keeps the same headers and the same flow, and swaps the payment. The price is a small proof-of-work puzzle, a shrunken Quai mining job. The client solves it on its own machine and attaches the answer. There's nothing to sign up for, nothing to fund, and nothing that links back to you.
It takes two ordinary HTTP requests with a small puzzle in between:
Ask. The client sends a normal request.
402. The server answers 402 Payment Required with a PAYMENT-REQUIRED header: a price and a small mining job for its algorithm.
Work. The client searches for an answer. There's no shortcut, and checking it takes one try.
200. The client retries the identical request with a PAYMENT-SIGNATURE header. The server verifies the answer, spends it once, and responds with a PAYMENT-RESPONSE receipt.
Prices are set in expected guesses, so a seller prices each route by what it costs to serve and how much friction they want. A laptop browser makes roughly 50,000 SHA-256d guesses a second: a price of 65,536 is about one second of work, and 1,048,576 is about twenty seconds.
More private. A payment carries no wallet address, account or card, so a request can't be traced back to you by how it was paid.
No wallet needed. Anyone with a device can pay: a person in a browser, an agent on a server, a script on a laptop. There's nothing to fund first.
Made for agents. An agent buys exactly what it needs, one request at a time, inside a work limit you set. No sign-ups, keys or checkout.
Fair to sellers. Scrapers that took pages for free now pay for each request, and that work goes to the seller instead of being wasted.
The work has value because the puzzles are real Quai Network mining jobs. Quai is solo-mineable, so each job pays the seller's own address. Most answers only prove the buyer did the work, but occasionally an answer is good enough for the network itself, and the block reward goes to the seller. Those rewards are paid by the network, not by x402 Work, and a receipt is proof of an access payment, not a promise of a reward.
Paying for any request that offers work is one call:
import { payWithWork } from "x402-work-client"; const { response } = await payWithWork( "https://hashference.com/x402/v1/chat/completions", body, { maxPrice: 500_000 }, // the most work you'll accept );Or from a terminal: x402-work pay <url>. Charging for your own routes is just as short. Pass your Quai payout address to workSeller, call seller.charge() in your handler, and the hosted facilitator builds each job for your address, so you run no node or verifier. Install with npm i https://x402.qu.ai/x402-work-client-0.1.0.tgz, and read the full guides at x402.qu.ai/docs.
x402 Work supports any proof-of-work algorithm through profiles. Quai's are live today: SHA-256d (CPU, GPU or ASIC), Scrypt (ASIC) and KawPoW (GPU).
The x402 Work marketplace launches with five services: Hashference, a JSON workbench, a text toolkit, a model catalog export, and a demo that unlocks a cat picture through a real HTTP 402, solved by your browser. Try the demo.
Hashference is an inference API that 45 live models sit behind, from Anthropic, OpenAI, DeepSeek, Moonshot (Kimi), Qwen, Z.ai (GLM), xAI (Grok), Meta (Llama), Mistral and Google (Gemma). It speaks the APIs your tools already use:
OpenAI-compatible: Chat Completions (with streaming and tools) and the Responses API at https://hashference.com/v1.
Anthropic-compatible: the Messages API, so Claude Code and the Anthropic SDKs point straight at it.
Works with your tools: Claude Code, Cursor, VS Code, Codex, OpenCode, Cline, Roo Code, Kilo Code, Hermes Agent, OpenClaw and any OpenAI-compatible app.
Claude Code is one line:
ANTHROPIC_BASE_URL="https://hashference.com" ANTHROPIC_AUTH_TOKEN="hf_..." ANTHROPIC_API_KEY="" claudeMine for credits. Point a GPU, ASIC or rig at Hashference over Stratum. Accepted shares become credit for any model, spendable as soon as the pool accepts them.
Pay per request with work. No key and no balance? An agent can pay for a single request with a few seconds of its own machine's work, using x402 Work, at /x402/v1/chat/completions, /x402/v1/responses and /x402/v1/messages.
Buy credits. Top up with QUAI or Qi on Quai Mainnet, or with USDT or USDC on Ethereum or Base, ETH or BTC. Every API key in your workspace spends the same balance.
Stake QUAI. The Starter plan stakes 100,000 QUAI for $20 of credit every 30 days, and Pro stakes 1,000,000 QUAI for $200. Your QUAI stays in the staking contract. Plan credit is funded by Hashference; it isn't consensus staking or a cash yield.
Workspaces and keys. One shared balance per workspace, as many API keys as you need, and a monthly budget per key.
Honest metering. 1,000 credits equal $1. Each request reserves credit, then settles at actual usage, and failures before generation are refunded.
MCP server. https://hashference.com/mcp gives agents tools to list models, check balance, chat and create funding invoices. Run it locally with hashference mcp and agents also get x402_chat, which pays with your machine's work.
CLI. hashference setup claude-code (or cursor, codex, opencode and more) writes your config. hashference x402 ask "…" pays for one answer with work, and hashference fund --usd 10 creates a QUAI or Qi invoice.
Playground and statistics. Try any model in the Playground, and follow usage on the public statistics page.
AI turns electricity into intelligence. The agents that run on it need to buy compute, data and answers from each other, one request at a time, without asking anyone's permission. Cards and accounts were never built for that. Work is: every device can do it, it's priced in something physical, and on Quai it settles as energy-based money.
x402 Work is the payment rail, Hashference is the first service on it, and Quai is the network underneath.
Read the docs and try the demo at x402.qu.ai
Get an API key, or mine for credits, at hashference.com
Related: Energy based money: what it is and how Quai built it · What is QI? The energy dollar for the AI economy
