The Local AI Substrate Three Missing Standards for Small‑Model Ecosystems

Small‑model ecosystems—local AI systems built from specialized, lightweight models running on consumer and edge hardware—are constrained not by capability but by infrastructure. Three architectural layers are absent: a standard for chaining models into deterministic pipelines, a standard for packaging models with the metadata and contracts applications require, and a standard for local capabilities discovery that lets orchestrators find and select models without manual configuration. Without these three layers, small‑model orchestration remains ad‑hoc, fragile, and ungovernable at scale. This working paper defines all three standards as a unified specification: the Local AI Substrate. Each standard is runtime‑agnostic, model‑agnostic, and governance‑first. Together they form the POSIX‑equivalent layer that small‑model ecosystems currently lack—the substrate that would allow a collection of local runtimes to behave like a real software platform.