The Shared Build is a collaborative technical proposal for P2P-LECS, a peer-to-peer compute and data commons for open AI systems.
The proposed first phase is modest: users contribute consumer hardware to run approved open-model inference jobs. A daemon detects resources, accepts signed manifests, and routes work through a peer mesh. Arbitrary remote code is excluded. More difficult phases would add consented data contribution, embeddings, LoRA adaptation, governance credits, and research into federated or distributed training.
The proposal avoids tradable tokens. Compute credits are nontransferable, basic access remains free, and governance weight is capped or sublinear. A contributor chamber would govern training mixture and rights questions, while a technical chamber would govern scheduling and resources.
Its data-governance section requires explicit consent, contributor provenance, rights to the material, and limits on sensitive, student, minor, or third-party conversations. The authors acknowledge that privacy is imperfect and that untraining from existing weights is not currently solved. P2P-LECS is a working specification and call for builders rather than an implemented public network.