# Inference, by the µ. > Inferit is a two-sided market for LLM inference. One OpenAI-compatible API sends each request to the cheapest healthy seller, priced in µ (0.000001 ITC on the testnet). A testnet demo. Inferit is a market for LLM inference. One OpenAI-compatible API sends each request to the cheapest healthy seller, and every request carries its own price, in µ. Developers keep their OpenAI client. Agents pay per request with x402. Sellers set their own price. [Get an API key](https://inferit.app/buy) · [Read the docs](https://inferit.cc/docs.md) Testnet demo · test tokens (ITC), no monetary value · contracts not audited yet Live quotes: the page reads the market from the API when it loads and shows each seller's price for one request of 1,000 input and 500 output tokens, with the cheapest healthy seller picked. Today's offers are on [the models page](https://inferit.app/models). - **The unit**: 1 µ. One millionth of an ITC (0.000001 ITC) on the testnet; a micro-dollar at mainnet (planned). - **Smallest x402 payment**: 10 000 µ. That is 0.01 ITC. It prepays the agent's own balance; each request then costs its exact µ. - **Gas for an x402 agent**: 0 WBT. The facilitator pays the gas of every x402 payment. An agent needs WBT only to withdraw. - **Fee cap in the contract**: ≤ 10%. Immutable. The platform fee today is 5% on top of the seller's price. ## A prepaid meter, charged by the µ. 1 µ = one millionth of an ITC (0.000001 ITC) on the testnet; a micro-dollar at mainnet (planned). Sellers price per 1M tokens, and each request is charged its exact metered cost in µ. You prepay into the escrow contract: a deposit under a ceiling you sign, or an agent's x402 payment. Each request is charged against that balance. What it does not use stays yours. A prepaid balance charged by the µ: worked example, prices as listed on 2026-10-04 | Line | What | Amount | | --- | --- | --- | | Prepaid | A deposit, or one x402 payment (the minimum) | 10 000 µ | | Request | qwen 2.5 14B · 1,000 in · 500 out | minus 147 µ | | Balance | Stays yours: spend it or withdraw it | 9 853 µ | Settled. The seller is paid its share of the 147 µ in the next batch on Whitechain Sepolia: one transaction, one line per seller. 147 µ = 1,000 input tokens at 73 500 µ plus 500 output tokens at 147 000 µ per 1M tokens, all-in (the seller's price plus the 5% fee), as the founder's Mac listed on 2026-10-04: an illustrative snapshot, not today's price (sellers relist at any time). Testnet demo: 1 µ = 0.000001 ITC, a test token with no monetary value. Example · prices of 2026-10-04 1. Quote Each healthy seller's price per 1M tokens becomes a quote for your request, plus the 5% fee. 2. Pick The router sends it to the cheapest healthy seller on Inferit. Sellers failing live checks are skipped. 3. Paid The metered tokens are charged in µ against your balance, then settled to the seller in a batch on chain. ### ITC · Inferit Test Credit The test money you pay for inference with. It has 6 decimals, so 1 µ = 0.000001 ITC. Inferit's ITC faucet gives 1,000 ITC per address per 24 h, free. It is sponsored, so the claim needs no gas. ### WBT · Whitechain's gas token Needed only for on-chain transactions you send yourself: a deposit, a seller registration, a withdrawal. An x402 agent needs none until it withdraws: the facilitator pays the gas of every x402 payment. Testnet demo on Whitechain Sepolia: both are test tokens with no monetary value. ## Agents pay per request. No account, no API key. A call without a key gets HTTP 402 with a quote: the worst-case cost of that exact request, pinned for 120 s. The agent signs it with a standard x402 client and resends the same body. Inferit settles the payment on chain first, then serves the request. x402 can express 1 µ, but every x402 payment is an on-chain transaction that costs gas. So the payment, at least 10 000 µ (0.01 ITC), is credited to the agent's own escrow balance, and each request is charged its exact µ cost against it. - Unmodified x402 v2 clients work, such as the official x402 fetch client: `PAYMENT-REQUIRED`, then `PAYMENT-SIGNATURE`. - An agent with zero WBT claims test ITC from the faucet and pays. Inferit's facilitator pays the gas. - Discovery at `https://api-production-c74b9.up.railway.app/.well-known/x402`. The whole guide for agents is one markdown file. - The x402 client is configured with a maximum per payment (agents.md uses 1 ITC). [Where agents go](https://inferit.cc/agents.md) [agents.md](https://inferit.cc/agents.md) HTTP 402 Payment Required ### Postage due. 10 000 µ to deliver this request. - **PAYMENT-SIGNATURE**: the stamp · sign 10 000 µ, resend - **CHARGED**: the exact µ cost · the rest stays yours - **ACCOUNT**: none · no API key - **GAS**: 0 WBT · the facilitator pays Testnet demo: µ is a micro-ITC, a test credit with no monetary value. _x402 v2 on the wire_ ```text POST /v1/chat/completions (no key) <- 402 PAYMENT-REQUIRED quote ≥ 0.01 ITC, pinned 120 s POST /v1/chat/completions (same body) PAYMENT-SIGNATURE EIP-3009 transfer to the escrow the API settles it, then serves <- 200 PAYMENT-RESPONSE x-inferit-buyer-cost-micro = the exact cost in µ ``` qwen/qwen-2.5-14b-instruct on the founder's Mac, as listed on 2026-10-04: 0.07 µ/token input, 30% under list before the fee (list: OpenRouter snapshot, 2026-10-02). Its stamp is its quote for 1,000 input and 500 output tokens. Testnet: test ITC. ## Set your price. The best healthy quote gets the request. Serve an open-weight model from your own hardware, from a GPU cloud to a Mac, behind any OpenAI-compatible server (vLLM, Ollama, SGLang). You list a price per 1M tokens in µ, and the router sends you a request whenever yours is the cheapest healthy quote. - You register your payout wallet on chain. Only that wallet can change it. - You are paid per request at your own price, to your payout wallet, in batches on chain (test ITC on the testnet). Buyers pay your price plus the 5% fee. - Licences are checked at listing; non-commercial weights are refused. - Resale of hosted-API keys is gated behind an allowlist, empty by default. Today every seller is the team's own: the founder's Mac running Qwen 2.5 14B (while it is listed and online) and a labelled demo seller that returns canned responses. [The models page](https://inferit.app/models) shows which offers are healthy now. [List a model](https://inferit.app/sell) [Seller docs](https://inferit.cc/docs/sellers.md) ## The contract bounds what can be settled. Deposits sit in `MarketEscrow` on Whitechain Sepolia. The API meters each request and settles it there in batches, within the bounds below. The contracts are not audited yet. - Custody ### Owned by a 2-of-3 Safe. `MarketEscrow` holds balances under contract rules. The operator can only settle, within each buyer's signed cap and expiry. A guardian can only pause. Withdrawals can never be paused. On testnet the founder holds all three Safe signer keys. - Open source · Apache-2.0 ### An open-source facilitator. whitechain-x402-facilitator, the first public x402 v2 facilitator for Whitechain, live on Whitechain Sepolia (testnet), is a separate project for other Whitechain merchants; Inferit operates a public instance of it. Inferit's API settles its own x402 payments with its own x402 v2 facilitator, self-hosted and operated by Inferit. [Read the facilitator](https://github.com/OGcryptonaut/whitechain-x402-facilitator) - Status ### A testnet demo, stated plainly. ITC and WBT are test tokens with no monetary value. The contracts are not audited. All on-chain activity so far is our own test traffic. [On-chain figures](https://inferit.app/analytics) ### What the contract enforces - A buyer-signed cap with an expiry. The operator settles at most what is left of it. - Payouts only to registered sellers, at the payout wallet each seller set. - A platform fee capped at 10% of the seller amount, fixed in the contract. - A circuit breaker on settlement: 1,000 ITC per hour across all buyers (testnet). - An operator replaced only through a 24 h timelock. - Withdrawals that can never be paused. A buyer withdrawal takes two steps, 1 h apart. ### What the API keeps - Token counts and amounts per request, never prompts or responses. The seller that serves a request does see it. [What is not verified yet](https://inferit.cc/docs/security-trust.md) ### Live contract roles [All parameters](https://inferit.cc/docs/payments/whitechain.md#live) Read from the contract when this page loads. Every address opens on the explorer. The page reads the escrow's roles from the contract when it loads: owner and fee recipient (the 2-of-3 Safe), guardian and operator. The addresses and every parameter are on [the Whitechain escrow docs page](https://inferit.cc/docs/payments/whitechain.md). ### Receipts · 2026-10-04 Team test traffic · Whitechain Sepolia explorer - The first x402 payments on Inferit: a fresh agent key with zero WBT, official x402 fetch client [0xdad37ee0…2a03](https://explorer.testnet.whitechain.io/tx/0xdad37ee0fbc98bea181e56479d7226685ed30288302d9c8f5f9593b913692a03) - The settlement batch that paid the seller for that usage [0x524ac01a…f9e4](https://explorer.testnet.whitechain.io/tx/0x524ac01ac0b07ba6eaae1610adce0becd5ff386e7010c40a56dcd6664441f9e4) - whitechain-x402-facilitator's first settlement: 0.01 ITC from an agent with 0 WBT [0xc1e9c255…3615](https://explorer.testnet.whitechain.io/tx/0xc1e9c255d7ac63c99a68406a3a0226cf717654a25f55c73c0454f9ae82e33615) ## Keep your OpenAI client. Change the base URL and the key. Works in the OpenAI SDK, OpenCode, Cursor and other OpenAI-compatible clients. Testnet demo: test tokens, no monetary value. [Get an API key](https://inferit.app/buy) [Read the quickstart](https://inferit.cc/docs/quickstart.md) --- Canonical page: https://inferit.cc All pages as markdown: https://inferit.cc/llms.txt Testnet demo: test tokens, no monetary value.