Run the work on machines you hold.
Status: Running
Alpha is the infrastructure a community can hold, audit and operate.
A request is made
Summarise this month’s workshop notes.
Policy decides where it runs
Placed on the community’s own machines.
The answer comes back
Notes summarised.Where it ran, and what it cost, recorded.
Illustrative example · an example request, placement and record
Policy decides where work runs, before it runs.
Not habit, and not a vendor.
Read the full project overview
This is an experiment on the layer: Alpha is the infrastructure a community can hold: a policy-driven execution platform where identity, routing, infrastructure state, and automation work as one coherent system.
What it is
Alpha goes beyond a hardware cluster. It places workloads by reliability and performance class, routes AI requests by policy, and keeps repeatable operations executable and versioned. The substrate that makes "run it on your own systems" a real option rather than a slogan, which is what a credible right of exit requires from infrastructure.
End-state concept
Canonical infrastructure model that can drive deployment and audit.
Secure access model with centralized identity, policy, and traceability.
Deterministic automation plane where repeatable operations are executable and versioned.
Tiered compute and data platform that places workloads by reliability and performance class.
Policy-aware AI routing that can evolve from rules to model-based routing safely.
Operating principles
Strict role separation. Compute, storage, and network are distinct, governed planes.
Deterministic by default. Automation is repeatable; non-determinism is explicit and bounded.
Routing before execution. The system decides where work runs before it runs.
Least privilege. Every actor gets the minimum access needed.
Namespace-first. Resources are organized by namespace rather than by convention.
Observability is architecture. Every layer exposes its own health and cost signals.
Progressive escalation. Risky operations step up in confirmation as blast radius grows.
Resource pooling. Compute is shared, tiered, and metered.
Cost-aware routing. Routing decisions account for the cost of the execution target.
Repetition becomes automation. Any operation performed more than once becomes a versioned, reusable artifact.
Who it is for
Platform engineers who need a coherent substrate for GPU and CPU work, LLM routing, and data services.
Operations teams who need deterministic, auditable automation instead of tribal knowledge.
Communities that want their AI and data workloads on infrastructure they control, with policy rather than trust deciding where work runs.