# About Inpherio — the open inference exchange

Source: https://inpherio.co.uk/about

About

# Most of the world's GPUs are *idle.*

Gaming rigs overnight, workstations between jobs, a Mac on a desk, a Raspberry Pi on a shelf. Meanwhile developers pay rate-card prices for open models that could run on any of them. Inpherio exists to close that gap — honestly.

What we are

## An exchange, not a cloud.

Inpherio connects people who already self-host AI models with people who need inference. Operators run a small connector next to the engine they already use and set their own prices. Developers reach every one of those models through one OpenAI-compatible API, routed to the best healthy machine for each request — weighing whether the model is loaded, reliability, price and speed — and moved to another if it fails.

We run the parts in the middle: routing, failover, metering, billing, payouts and the checks that keep everyone honest. For that we take a flat 15% from what providers earn — never a markup on what developers pay.

How we work

## Principles we hold ourselves to.

### Facts, not promises

Every fee, limit and payout rule on this site is read live from the platform's configuration. If the site and the software ever disagree, that's a bug and we fix it.

### Say the limits out loud

Consumer hardware has no secure enclave, residential nodes go offline, and we're in beta. We tell you, instead of selling an SLA we couldn't keep.

### Fair to both sides

Developers pay the provider's price with no markup, and we pay the card fee on their top-ups. Providers keep 85% of what they earn — and 100% during their fee-free period (12 months fee-free (up to £250 of earnings)); a payout fee of £2 + 1% per payout to a UK bank account, £2 + 2% to any other country applies each time they are paid. [Every fee, in full](https://inpherio.co.uk/pricing).

### Your traffic isn't our product

We don't train on prompts or outputs, and we don't keep their text. What we do record, and why, is in the [privacy policy](https://inpherio.co.uk/privacy).

### Measured, not claimed

Every model is benchmarked on the machine that serves it, token counts are checked against the text actually served, and reliability is scored from what the platform observes.

### Built in the open

The architecture, the security model and the money rules are all published — [how it works](https://inpherio.co.uk/how-it-works), [security](https://inpherio.co.uk/security), [pricing](https://inpherio.co.uk/pricing).

Where it stands

## In beta, and working today.

You can sign up, connect your hardware and earn, or create an API key and make requests right now: token-metered billing, audio billed by length, node rentals, operator self-serve, card top-ups and bank payouts in 33 countries all work. Inpherio is governed by the law of England and Wales.

Questions, partnerships or press: [get in touch](https://inpherio.co.uk/contact).

## Join from either side.

No waitlist. Connect a machine, or create an API key and make your first call.

 [Create an account](https://inpherio.co.uk/register) [Contact us](https://inpherio.co.uk/contact)
