Overview
What Keep is, the problem it solves, its architectural assumptions, the four defensive layers, the two SKUs and the claim boundary, the three delivery modes, and how it composes with the rest of the stack.
Architecture
Moat, Barbican, Rampart, and Warden in depth — what each layer does, how a gated action is decided, and the precise distinction between the probabilistic layers and the correct-by-construction boundary.
Delivery modes
The three ways Keep deploys — Edge Filtration, Semantic Classification, and the Enforced Boundary — worked examples for each, and the zero-rewrite graduation path between them.
SDK Reference
@k9o/moat and @k9o/barbican — install, the Edge Filtration API shape, per-language availability, and current build status for each.
API Reference
The hosted Keep classification endpoint — POST /rampart/classify (Semantic Classification) — request/response shape, the receipt, and error codes.
Billing
The reserved-capacity billing model — Governed Operations, GOPS, capacity packs, the additive-weight event model, verdict-neutral metering, and how Keep's economics stay orthogonal to Kilimanjaro's other module meters.
Onboarding
Start with Keep standalone — no account, no Kilimanjaro plan — install the SDK and gate your first agent call in minutes.
Audit & Receipts
Every Warden gate decision is a signed, tamper-evident receipt correlated to the project's audit stream. The receipt shape, what it proves, how it relates to Rampart's advisory classification receipt, and how it relates to Kibo's audit ledger.