@irisrun/host

Same image, a different host — made explicit. A host is just {name, capabilities, store, scheduler}. Once the engine is pure and durability lives in the journal, "run anywhere" stops being a slogan: any host that supplies a store + scheduler can resume the same image, and a capability diff decides — at deploy time, loudly — whether it's allowed to. Host-side; the core stays pure.

What it is

HostAdapter is the four-field surface (name, capabilities, store, scheduler). runTurnOn(adapter, opts) runs ONE turn on that host's store + scheduler — a thin call into the engine's runTurn with the adapter's ports injected, defaulting the writer holderId to the host name.

The deploy gate is two functions over an image's CapabilityProfile: diffCapabilities returns the structured CapabilityGap[] (booleans like filesystem/websockets, plus tool_locality ranked remote < local < in-process against the host's ceiling); assertDeployable throws — joining every gap's message — if the image can't run. checkHostCapabilities is the narrower tool-level boolean refusal. Refuse LOUDLY, never silently degrade. Zero external deps; depends on @irisrun/core + @irisrun/agent only.

Use it

import { runTurnOn, assertDeployable, type HostAdapter } from "@irisrun/host";

assertDeployable(image.requires, host);        // throws naming every gap, or returns
const outcome = await runTurnOn(host, { sessionId, defDigest, program, performers, clock });

See docs/Architecture and docs/Deploy.

Part of Iris — own, portable, verifiable state.