Show HN: Forky Pig is a VM hypercompute fabric for AI Agents (github.com)

Forky PIG is an experimental “hypercompute fabric” that treats virtual machines like forkable processes: using QEMU, qcow2 copy-on-write overlays, and lightweight Python orchestration, it can snapshot and clone VMs in milliseconds. The system runs a host daemon (hostd) that controls local QEMU instances via QMP and a controller that issues fork/snapshot commands over gRPC. The current feature set includes reliable cold forking (snapshot a paused VM and instantly spawn children that share a base image but keep private writable overlays), stable snapshot chains, and ongoing experiments in hot forking, GPU passthrough, and multi-host orchestration. Implementation details: Python 3 asyncio + gRPC, qcow2 overlays, optional Linux KSM for memory deduplication, and simple run scripts—no Kubernetes, just processes and files. For the AI/ML community this matters because Forky PIG enables ultra-fast, deterministic cloning of complex, stateful environments—useful for agent training, reproducible experiments, debugging, and bursty multi-agent workloads where GPUs and orchestration are bottlenecks. Rather than rebuild or redeploy, researchers can branch off live environments, share a single base image to save storage, and iterate quickly. It’s not production-ready, but it’s a practical playground for exploring “git-like” workflows for machines and distributed hypercomputing primitives that could reshape how compute is provisioned for large-scale AI experiments.