INFERENCE
IS A COMMODITY.
PRICE IT.
A marketplace where GPUs bid for prompts. Clients pay providers in xBZZ, payment escrows on Gnosis, and every byte routes over Swarm. No middleman, no rate limit, no key.
HOW IT WORKS
Four moves. No central operator owns any of them. Selection, payment, reputation, and liveness all live on-chain.
No middleman. Direct settlement. No key. Pay per AI token, refund on miss.
-
Providers run inference jobs.
Anyone with a GPU and an OpenAI-compatible backend (Ollama, vLLM, llama.cpp…) runs the T4T provider container. It registers on-chain, stakes xBZZ, and publishes per-million-token prices for each model it serves.
-
Gateways post inference jobs.
The gateway container exposes a normal OpenAI /v1/chat/completions endpoint locally. It picks a provider from the on-chain directory, escrows the maximum payment, and ships the encrypted prompt over Swarm PSS.
-
Provider serves, claims on delivery.
The provider runs inference locally, returns the response over Swarm, and claims payment from escrow. If the deadline lapses, the client gets a refund. Repeated faults slash the provider's stake.
-
No middleman, no API key.
No central API gateway. No marketplace operator. No rate limit. Just a contract, a stake, and a hash for every byte exchanged.
→ Full protocol: docs/spec.md
LIVE DIRECTORY
Pulled live from ProviderRegistry on Gnosis. Only providers with a fresh heartbeat (≤ 10 min) and at least one offering appear.
GET STARTED
Two containers. Pick a side. Both ship with onboarding UIs that mint a wallet and bind a Swarm postage stamp on first run.
As a gateway.
Run the gateway container. Point your existing OpenAI SDK at http://localhost:8080/v1. Pay per token in xBZZ.
As a provider.
Run the provider container against your local Ollama / vLLM. Stake xBZZ. Earn per-token revenue from clients worldwide.