Skip to content
PRESHaiPRESHaiPRESHos Platform

Knowledge stores

Knowledge stores are the scoped reference data sessions read: knowledge bases managed for your agents. The knowledge surface is /ai/knowledge. Each store card shows sources, chunks, and token counts with an updated time; shared stores are tagged. Those counts are the size and freshness signals: a store with stale data or mystery sources gets re-scoped or removed, like stale access. Store mechanics beyond that are not documented in source yet.

  • Scope stored content like any context: named systems, named data, explicit exclusions. Blanket ingestion of “everything the team can see” is not a scope.
  • Stored content informs; it does not authorize. A document describing a system does not put that system in session scope.
  • Tenant and org boundaries apply to stored content exactly as to sessions. One tenant’s stores never leak into another’s.
  • Review stored content on the same rhythm as membership: stale or mystery content gets removed or re-scoped, like stale access.
  • Record what each workload may read: store, scope, review date. The pilot reads that register; the handoff hands it over.
Knowledge Stores: knowledge bases with sources, chunks, and token counts

Treat /ai/knowledge as the source of truth for knowledge-store surfaces. Context scoping in depth: Environments, Core concepts.