Pool local Macs for MLX workloads.

R&D. The transport substrate is published as Go modules: mlx-go-iroh ships peer discovery, gossip, and content-addressed blobs (v0.3.0, with opt-in global discovery). Quorum-gated DiLoCo sync and peer-to-peer weight sharing run over it as live, untagged datapaths. In a small measured run, two machines fine-tuned, exchanged, and combined LoRA adapters with a merge rule designed to tolerate bad contributions; both wrote byte-identical files across three rounds (the write-up). Coordinated training across the mesh is still being built.

• Memory-weighted scheduling — a 128GB Mac and a 48GB Mac pool their capacity automatically. Pipeline parallelism proportional to each node's unified memory.
• Content-addressed weight sharing — model weights move peer-to-peer over LAN, Apple AWDL, and a global iroh mesh, verified against Hugging Face SHA-256 digests rather than trusting any peer.
• iroh transport — QUIC links and signed gossip via mlx-go-iroh. LAN and gossip by default; pkarr/relay global discovery is opt-in.
• Quorum-gated sync — DiLoCo-style updates apply only when a quorum of distinct peers publishes the same signed digest. Fail-closed: the live datapath stays off unless explicitly enabled.

Research. mlx-go-iroh is published and tagged; weight sharing and a quorum-gated DiLoCo sync loop run on top of it. The next validation gates are a complete multi-node training run, cross-device latency measurement, and heterogeneous-compute consistency.

mlx-go runs the local computation; mlx-mesh coordinates the machines and moves model state between them. The intended use is a model that does not fit on one Mac, or a training workload that must remain inside an operator's custody domain. cove can isolate individual workloads when they need a disposable Mac.