Screen Flows — Knowledge Base

Screen IDs reference screens-knowledge-base.md.

Audience: Tenant admins (Dashboard) and DM Portal reviewers. Platform: Web only. Entry point: Dashboard Settings sidebar → Knowledge Base → /settings/knowledge-base

Status: [To Build] — The entire Knowledge Base / RAG system is new. No KB screens exist in the current live app. All flows below are specifications for what needs to be built.

1. Overview → Dataset Management

2. Dataset Detail — KB2

KB2 is a three-tab screen for a specific dataset.

3. Upload File Modal — KB3

4. Retrieval Sandbox — KB4

5. Dataset Configuration — KB5

Note: Embedding model is read-only after dataset creation. Changing the model would invalidate all existing vectors — users must create a new dataset instead.

6. Website Crawl Settings Modal — KB6

Schedule options: Weekly (every Monday at 3am) · Monthly · Manual only.

7. RAG Context in Other Apps

Knowledge Base content surfaces in two other apps when agents use rag_search.

Dashboard — Campaign Detail Activity tab

Activity cards show a collapsible “Retrieved context” section listing:

• Each RAG query the agent made
• Dataset searched, chunk count, relevance scores
• Chunk preview text and source file

DM Portal — Activity Detail (P3)

Same “Retrieved context” section appears in the right metadata pane alongside cost, token count, and model info. This lets reviewers identify poor retrieval quality and flag that a re-crawl or re-index is needed.

Manage — Tenant Detail Knowledge Base tab (M3)

Super admins see a read-only dataset summary per tenant:

Admins can trigger re-index or clear a dataset on behalf of a tenant; all actions are written to audit_logs.

8. Screen Index