Owner economics first.
No machine is purchased before a named workflow, accountable owner, baseline and acceptance test exist.
Starlight protocol / joint infrastructure / 01
A deal system for operators who can create AI leverage and owners who control operating demand, warehouses, solar, batteries, buildings or capital. The owner buys measurable capability. The operator earns by designing and running it. Capacity expands only when the receipts justify it.
00 / transaction thesis
Do not ask another company to fund your AI company. Construct a transaction in which it buys its own productivity, owns the productive assets, and grants you only the capacity and economics you have explicitly earned.
No machine is purchased before a named workflow, accountable owner, baseline and acceptance test exist.
Ownership, warranty, location, cost basis, capacity and exit are registered—not inferred from family trust or enthusiasm.
Business-critical demand, operator allocation, burst rights and revocation are explicit. Money never becomes accidental governance.
Recurring value, utilization, failure rates and founder attention decide the next asset purchase. A GPU farm is an outcome, not a starting identity.
The exchange
The protocol preserves separate ownership while composing the assets neither side could exploit as effectively alone.
Higher productivity, owned infrastructure, controlled private intelligence and optional residual-capacity yield.
Architecture, implementation and recurring operations revenue; only then defined capacity or performance upside.
Deal sequence
Each gate creates a receipt that makes the next commitment legible to the owner, operator, lawyer, accountant and lender.
Name the owner, the economic workload, the asset base, the operator and the decision rights. Reject deals whose only thesis is cheaper compute.
The owner commissions a fixed-scope Value Architecture. Baselines, economics, risks and procurement triggers are established before implementation begins.
Ship the smallest set of workflows capable of changing revenue, margin, working capital, asset productivity or management load within ninety days.
The asset owner buys only the machines justified by measured demand. Ownership, warranty, location, metering and capacity rights are explicit.
Workers receive bounded authority. Every consequential action has a human owner; every claim of value has a baseline and attribution rule.
Scale into additional buildings, workloads or external capacity only when recurring value, utilization and operating discipline clear the contract gate.
Where value is allowed to come from
Lead qualification, configuration, pricing, quotation, follow-up and customer lifecycle—without replacing the systems that already hold commercial truth.
Supplier intelligence, stock exceptions, price drift, receivables preparation and purchasing signals delivered as decision-ready exceptions.
Metering, maintenance, PV and battery telemetry, tenant or site services, remote operations and additional productive uses of physical space.
Document work, retrieval, models, creative production and internal agents routed across local and cloud capacity according to privacy, quality and economics.
The operating system
The same primitives support one family business, ten operator-originated deals or a multi-site portfolio. Tenants, credentials, memories, budgets and queues remain separate.
Procurement discipline
Provenance
No blockchain theatre. Versioned contracts, signed approvals, asset identities, run receipts and attributable baselines are enough—provided they are part of the operating path rather than assembled after the fact.
DEALSigned scope, economics, authority and expiry
ASSETOwnership, cost basis, location, warranty and capacity
WORKERRole, runtime, model, tools, permissions and budget
RUNInput lineage, actions, approvals, outputs and exceptions
VALUEBaseline, realized effect, attribution and confidence
DECISIONHuman owner, options considered and final authority
Contract stack
A lawyer localizes the templates. The protocol determines which questions the documents must answer and which rights expire with the underlying exposure.
Defines the paid diagnosis, data room, decision rights, fee and stop condition before implementation is discussed.
Separates services, liability, confidentiality, warranties and payment from any financing or ownership relationship.
Names the three workflows, acceptance tests, budget ceiling, owners, dates and evidence required for the 90-day gate.
Records owner, serial number, location, warranty, purchase basis, useful role and replacement trigger for every machine.
Reserves business-critical capacity, operator capacity, burst rules, metering and priority without implying equity or control.
Binds every worker to allowed tools, data classes, spend limits, approval gates, escalation and revocation.
Defines tenancy, secrets, retention, export, incident handling, subprocessors and the boundary between private and shared systems.
The owner keeps business data and bespoke deliverables; the operator keeps generalized platform primitives and unrelated ventures.
Fixes baselines, attribution rules, value owners, cost owners and the receipt format before anyone claims ROI.
Specifies credential handover, data export, worker shutdown, asset return, deletion evidence and continuity if the partnership ends.
For operators
The operator does not pitch “AI consulting.” They originate a bounded productive node, bring the Starlight contract and evidence system, and remain free to build their own portfolio because custom work is separated from reusable primitives.
Open the operator model ↗Find an owner with demand and assets—not a passive investor.
Translate workflows into baselines, payback gates and downside limits.
Separate services, assets, capacity, data, IP, authority and exit.
Route work across people, agents, local nodes and cloud providers.
Maintain workers, connectors, observability, incidents and evidence.
Earn the right to add hardware, sites and revenue streams from receipts.
Reference commercial model
Terms are configured to the company, jurisdiction and risk. The reference model keeps architecture, implementation, hardware and ongoing operation economically separate.
A decision-grade deal thesis, baseline, contract map, node design, economics and a go / revise / stop recommendation.
Three bounded workflows, owner-visible controls, worker receipts, acceptance tests and a measured scale decision.
Runtime stewardship, worker operations, observability, patching, evidence reports and controlled expansion.
Construct the first transaction
Compile the deal, issue the Value Architecture order, open a bounded data room and reach a go / revise / stop decision before anyone buys infrastructure.