Knowledge Base CMS · Documentation

Architecture

Knowledge Base CMS's pipeline, its owned data, the events it emits/consumes, and what is out of scope.

Knowledge Base CMS
author+ versionauto-tagmeteredpublishowner-gatedchunk→ groundingmetered · LLMdeterministic · $0
Live diagram — authoring, publish and chunking are deterministic; only the auto-tag assist is metered.
author$0$auto-tagmeteredpublish$0chunk$0COST LEVER · auto-tag is optional
Live diagram — spend accrues only on the optional auto-tag; authoring and the publish-to-grounding path are free.

Pipeline + owned data

Authoring, search, versioning, and chunking are deterministic; the one metered stage is auto-tag. The app owns knowledge_article in its demo_eco_c3_article table; a published article's body re-chunks into demo_eco_c3_kb_chunk (the concierge grounding unit). It reads the shared host #01 catalog for context. The data invariant holds: owner writes canonical (credential_id NULL); viewer writes credential-scoped, ephemeral, reset-clearing; the DO-UPDATE scope guard blocks a viewer write from clobbering canonical.

Metered AI + dual-mode, out of scope

The one metered stage (autoTagArticle, an AI-assist) runs dual-mode: Cloud claude-haiku-4-5, cost-capped at $0.05 per session and fail-closed — over the cap it falls back to recorded OSS (qwen2.5-7b-instruct) at $0, recorded on local hardware and never fabricated. Each run writes a cost_ledger row and a demo_cache replay key. Publish emits article.published to the durable event log. Out of scope and clearly labelled: auto-tag is a deterministic keyword-hint simulation in the prototype; content is synthetic; no production prompts appear. Honesty is inspector-only.

Architecture · Knowledge Base CMS · Abhishek Saxena