p2ptokens
A distributed, peer-to-peer network for LLM inference where every peer both serves and consumes completions, metered by a BitTorrent-style upload-to-download ratio instead of money, brokered by a coordinator that never sees the request content.
p2ptokens is a single binary that both serves and consumes LLM completions across a peer-to-peer swarm, applying the seed-and-leech incentive model of the BitTorrent protocol to inference instead of file sharing: a peer seeds completions from a model it already runs, such as one served through Ollama, an OpenAI-compatible endpoint, or Claude, and leeches completions from other peers when it needs more capacity than it has. Access is governed by an upload-to-download ratio rather than a payment, intended to discourage free-riding among peers who do not know or trust each other.
A central coordinator brokers matches between peers but is designed to never see the content of a request, since inference bytes flow directly between the two peers involved. The project describes signed co-receipts that bound how much either side of an exchange can misreport usage, signed heartbeats intended to prevent a peer from registering under another peer's identity, and an index designed to match peers by model in constant time at scale.
It is written in Rust, built on libp2p, and exposes an OpenAI-compatible interface.