niceley.ai · intake deliverable
Private AI intake plan
A useful intake ends with this information written down. The plan belongs to the buyer whether or not the implementation proceeds.
The day-one job
One real task, representative inputs, desired output, and the person responsible.
Data boundary
Data classes in scope, what must stay local, retention, and every permitted exception.
Quality floor
Examples of a passing answer, a failing answer, and the required human review.
Model and retrieval
Candidate model, why it fits, document sources, and when cloud quality is still needed.
Agent boundaries
Allowed reads/actions, approval gates, blocked paths, and the action-receipt location.
Hardware
Existing or dedicated machine, memory/storage assumptions, quoted-at-cost parts, and ownership.
Acceptance
Unplug test workflow, response-time target, backup check, training, and runbook handoff.
Scope
Milestones, dependencies, exclusions, schedule, setup price, and optional monthly care.