Every OpenAI-compatible app today points at one of a handful of providers. That's convenient — and it means one company sets your price, your rate limits, your content policy, and your uptime. Inference is becoming infrastructure; infrastructure shouldn't have a single owner.
The marketplace thesis
There are millions of idle GPUs — gaming rigs, workstations, small datacenters. QueraIS turns them into a market: anyone can run a node, advertise the models they serve and the price they'll take, and earn $QAIS for every token they stream. Requesters call an OpenAI-compatible API and the matching engine routes each job to a node by price, reputation, and speed. Ninety-five percent goes to the node; five percent is the protocol fee.
It's BitTorrent for AI inference: no central server to depend on, the network gets stronger as more people join, and you only pay for what you use.
The hard part is trust
An open network means strangers run your prompts — so the protocol assumes nodes may lie. That's the whole game, and it's why QueraIS leans on four independent checks:
- Statistical re-runs — ~5% of jobs are re-run on oracle nodes and compared by embedding similarity.
- Format + length — cheap deterministic checks on 100% of jobs.
- Economic stake — every node posts slashable $QAIS collateral.
- Requester feedback — a soft signal that can never trigger a slash on its own.
A five-dimension reputation score compounds those signals, and proven fraud burns stake. Honesty ends up being the cheaper strategy — which is the only kind of honesty that scales.
Where we are
Today there's a trusted gateway doing matching and settlement — bounded (it can only settle at prices you signed, never steal deposits), and on the roadmap to be removed step by step: on-chain auction, a P2P mesh, a decentralized oracle. It's a testnet on Arbitrum Sepolia, so $QAIS has no real value yet. Point your OpenAI client at it and try it — see the quickstart.