One artifact, without code changes, executes across laptop, edge, and cloud — with predictable state and observable telemetry.
VYR is not another orchestrator. It occupies the execution fabric layer — portable, vendor-neutral, offline-tolerant. KYX sits above it as the systemic intelligence plane.
Artifacts execute as Wasm/WASI by default — portable across laptop, edge VM, and cloud without OS or vendor runtime dependencies.
VYR Node RuntimeEvery artifact is addressed by content digest with signature verification. Reproducibility and provenance from the first line of code.
Artifact ModelTask-scoped ephemeral state with explicit binding model. Workloads declare state class — no implicit shared memory.
State KernelTraces and metrics flow via OTLP on the critical path from the first iteration. Instrumentation is not retrofitted — it is structural.
Telemetry PlaneTopology model built from accumulated VYR telemetry. KYX requires real history — it does not start until VYR is operational.
KYX — post-MVOStructural Causal Model with intervention and counterfactual evaluation. Advisory first — automation is staged after measured precision.
KYX — post-MVOThe exclusion list is as important as the inclusion list. Both appear in Whitepaper v0.1. Explicit non-goals prevent scope drift and establish trust with technical readers.
Digest addressing + signature verification. Local storage only; no registry service.
No multi-node, no placement engine. Submit, execute, terminate.
Single consistency class. No eventual, no strong state.
submit_task, cancel_task, create_state, bind_state, stream_traces, stream_metrics.
OTLP traces and metrics on critical paths.
First adopter can run a full cycle without source access.
event transform + disconnected collector across laptop / edge / cloud.
Numbers vs K8s baseline. Technical document with explicit non-goals.
Requires accumulated telemetry history that does not exist at MVO stage.
Adds coordination complexity without new proof for the portability thesis.
Require Raft and CRDT implementation — separate phase.
No sidecar proxies, no L7 routing, no inter-service mTLS.
Dilutes positioning; increases surface without strategic gain.
HA, automatic failover, SLA guarantees — phase 24–30 months.
Compatibility stub only; not a supported execution path at MVO.
Each layer has a single exit criterion. Development of the next layer begins only when the criterion is met. The critical risk: OTel hooks must be embedded during runtime development, not retrofitted.
Pre-production targets. All results will appear as raw numbers in Whitepaper v0.1 — not as assertions.
T0 = May 2026. MVO covers Phase 1. Subsequent phases are referenced here — each has a separate architecture document.
The full MVO specification — dependency graph, assembly layers, KPI targets, and explicit non-goals — is in the docs.