Compute is easier to trust when you can read the receipt.
Cloud marketplaces today ask you to trust a platform's ledger. Compute Contracts Protocol asks you to trust arithmetic instead: every request, bid, acceptance and receipt is a record you or anyone else can fetch, verify and replay, because it was never anywhere but your own repository and theirs.
- No broker. There is no marketplace operator holding both sides' records. The firehose is public infrastructure; anyone can run a relay.
- No trust, just proof. A receipt's CID binds the whole contract chain. Verifying it is a hash check, not a phone call.
- Built on what exists. It is ordinary AT Protocol records and the same firehose Bluesky and every other app already reads.
Not ready to run a requester?
The gateway requests the nodes of a Kubernetes cluster for you: a few calls in, a kubeconfig out, while every record it writes is still the same open protocol underneath.
Who does what
Four roles, none of them a central authority.
| actor | role |
|---|---|
| Requester | Authors the market.rfp and market.accept records. Wants compute. |
| Provider (bidder) | Authors market.bid and market.receipt. Has spare machines. |
| PDS | Each actor's Personal Data Server, holding their signed records. |
| Firehose / relay | Streams commits from every relevant PDS so participants can discover each other's records. |
Where this is headed
This is alpha software: the NSIDs live under a temporary namespace (com.publicdomainrelay.temp.*) and interfaces will
move. It already works for individual providers and requesters; the next stretch is organizations that hold compute
and want to hand it to their own members through the same records, and a scorer that goes past "lowest cost."
Read the code, not just the pitch
The record schemas, the bidder and the requester are all in the open repository.