Knowledge Library
Knowledge Library manages durable source collections, indexing state, search, pinned libraries, recovery, and per-library settings without pretending that a visual component owns ingestion, embeddings, storage, or permissions.
Knowledge library
Durable sources with explicit update and recovery state.
Product decisions
A knowledge base is a durable, named boundary around approved sources. It is not a folder alias, a chat attachment list, or a decorative database card.
- Every library exposes source count, freshness, indexing phase, and a recovery path.
queuedandrunningare distinct because waiting for capacity and actively indexing require different expectations.- A failed update preserves the last usable searchable version. Destructive rebuilds require an explicit host flow.
- Automatic updates need a stated policy, minimum interval, pause behavior, and a visible manual recovery action.
- Source selection, authentication, parsing, chunking, embeddings, vector storage, retrieval quality, and access control remain host responsibilities.
Controller contract
Accessibility and ownership
Library items remain ordinary named buttons in a list; arrow keys, Home, and End move between visible items while Enter or Space opens details. Status always includes text and never depends on color. Detail tabs, when present, use the Tabs contract. The host owns authorization, ingestion jobs, retries, source validation, storage, retrieval, deletion recovery, and localized user data.
