Skip to content

Domain: Knowledge Bases

Knowledge Bases (KBs) are structured sources of truth used by AI and automation across the platform.

A knowledge base contains:

  • approved facts (products/services/pricing, locations, policies)
  • tone/voice rules and brand guidelines
  • compliance constraints (disallowed claims, required disclaimers)
  • mappings that connect KB concepts to channel/app entities
  • KB content must be versioned or at minimum have a publish/review boundary (MVP can start simple).
  • AI outputs must reference only approved KB content for “facts” (where required by policy).
  • KB access must be permissioned and audit logged for publish events.
  • Knowledge bases are workspace-scoped objects: a KB belongs to a workspace and must not be visible across workspaces unless the user is authorized for that workspace.
  • The platform provides an org-wide KB discovery surface (route-level concern) that aggregates KBs across all workspaces a user is authorized to access, enabling cross-workspace search/filtering without breaking workspace isolation.
  • Draft → Review (optional) → Published
  • Publishing triggers:
    • downstream cache refresh
    • app configuration refresh (Comment Responder, Chatti Live)

Any KB state change must have explicit expectations for:

  • Audit events: who changed what, when, and in which workspace/org context.
  • Analytics events: follow Analytics Events (MVP); avoid KB content and PII in payloads. At minimum, KB publish must be represented as kb.published.
  • Bypassing workspace scoping (e.g. “global KB table” without authorization constraints).
  • Encoding domain logic in UI-only behavior without a domain source of truth.
  • Skipping audit expectations for publish/edit actions.
  • Emitting analytics with message content, KB content, or PII.
  • “Fixing” correctness by widening permissions or weakening isolation.