FluxyChatHow-to Guides
Room memory for operators
Persistent facts extracted from room conversations (PH-132).
Room memory for operators
Room memory stores durable facts from the conversation: decisions, tasks, FAQs, user context, and key facts. Each entry links to source messages for audit.
Query
GET /rooms/:roomId/memory?limit=12
Authorization: Bearer <member-jwt>Optional filter: ?kind=decision
Keyword search: ?q=pricing (D1 LIKE). If you bind Vectorize + Workers AI (VECTORIZE, AI), extract also upserts embeddings and q re-ranks hits that already matched D1. Unbound Vectorize is a no-op. Not Cloudflare AI Search. Not Notion/Drive connectors.
Extract on demand
POST /rooms/:roomId/memory/extract
Authorization: Bearer <member-jwt>Requires a configured LLM provider on the Worker. Returns extracted entries and insert/update counts.
Dashboard
Rooms → Advanced → Room memory. Search box calls ?q=. Extract still needs an LLM on the Worker.
Memory kinds
| Kind | Use |
|---|---|
decision | Choices made in the room |
task | Action items and TODOs |
faq | Repeated questions and answers |
user_context | Preferences or facts about participants |
key_fact | Durable technical or business facts |
sentiment | Tone signals (use sparingly) |
Pair with Live knowledge graph for relational context.