Skip to main content
Meridian Inference is a decentralized inference router. Instead of holding an account with every model vendor, you point one client at one endpoint and pay per request in USDC. Payment happens over x402, the HTTP 402 Payment Required flow, so there is no billing portal, no monthly minimum, and no identity check between you and the model.

​
Vendors change, the API doesn’t

Model ids are namespaced as supplier/upstream-model-id. Adding a supplier adds rows to the catalog. It never changes your integration. Fetch GET /v1/inference/models and the current catalog, pricing, modality, and payment mechanics come back as registry data rather than constants you hardcode.

A single door
into every model

​
How a paid request works

1

Connect a wallet

One EIP-712 signature derives an in-session buyer key. The signature moves no funds and the key never leaves the browser tab. Your wallet stays the payer; the derived key becomes the spender.
2

Top up once

The top-up gateway takes a gasless EIP-3009 USDC authorization from any of 11 supported chains and credits the buyer balance on Base. No ETH required on your side.
3

Prompt without popups

Each paid request receives a 402 challenge from the supplier. The derived key signs it silently in the page and the request is retried with a PAYMENT-SIGNATURE header. Zero wallet popups per message.
The signature budget is the point of the design:

​
Own your conversations

Chat history lives in your browser, not in a company vault. Threads stay in your sidebar, searchable, and no platform owns them. There is no account to delete because there was never an account to create.

Your history,
not theirs.

​
Pricing

There is no Meridian subscription and no platform markup on chat. You sign the supplier’s own x402 challenge, and the supplier is the only payee. Per-token rates come from the supplier catalog and are shown in the model picker next to the per-request fee.

API Reference

Endpoints, request shapes, and payment kinds for the Inference API.