Skip to main content
Cendra’s Knowledge Base stores property-specific information that AI agents use to answer guest questions accurately. When a guest asks about WiFi, parking, check-in procedures, or local restaurants, the AI retrieves the answer from your knowledge base — not from general training data.

​
How the Knowledge Base Works

  1. Auto-populated from PMS — when you connect your property management system, Cendra imports property descriptions, amenities, and basic details automatically
  2. Enriched by you — add house rules, check-in procedures, local recommendations, FAQs, and custom information
  3. Used in real-time — every AI response checks the knowledge base for relevant information before answering
Navigate to Agent Hub → Knowledge Base to manage property knowledge.

​
Conversation History Import

When you connect your PMS, Cendra imports up to 6 months of past guest conversations. This gives the AI immediate context about how your team communicates, what questions guests commonly ask, and how your properties handle different scenarios. This imported conversation history serves two purposes:
  • AI learning — the AI understands your team’s communication style and common responses from day one
  • MIRA discovery — MIRA scans these past conversations to identify knowledge gaps and suggest answers that were already provided manually but never added to the knowledge base
This means Cendra doesn’t start from zero — it learns from your existing operations before you even send your first AI-powered response.

​
Types of Knowledge

CategoryExamplesSource
Property detailsName, address, amenities, room configuration, photosAuto-synced from PMS
House rulesCheck-in/out times, noise policies, pet policies, smoking rulesManual entry
Check-in proceduresKey pickup, smart lock codes, access instructions, parkingManual entry
Local recommendationsRestaurants, attractions, transportation, grocery storesManual entry
Emergency contactsMaintenance, property manager phone, local emergency numbersManual entry
FAQsCommon questions and your preferred answersManual entry
Custom proceduresHow to handle early check-in requests, lost keys, complaintsManual entry

​
Data Hierarchy — How the AI Decides What to Say

When a guest sends a message, Cendra’s AI assembles context from multiple sources in a specific priority order. Higher-priority sources override lower ones when there’s a conflict.

​
Priority Order (Highest to Lowest)

PrioritySourceDescription
1. Business FlagsClassifier AgentCendra first classifies the message: is it an emergency? A complaint? An availability question? An upsell opportunity? These flags determine which tools and knowledge sources the AI uses.
2. AI Rules & GuardrailsYour configured rulesCustom rules and default safety guardrails are checked before any response is generated. A rule saying “never discuss refunds” overrides any knowledge base content about refund policies.
3. Conversation HistoryPast messages in this threadThe AI reads the full conversation thread to maintain context. It knows what was already discussed and avoids repeating itself.
4. Property DocumentsUploaded files & custom knowledgeDocuments you’ve uploaded (PDFs, house manuals, local guides) — highest trust knowledge because you explicitly provided it.
5. PMS KnowledgeAuto-synced from your PMSProperty details, amenities, booking rules, and descriptions extracted from your PMS — medium trust, system-extracted.
6. System KnowledgeCendra platform defaultsGeneral hospitality knowledge and platform guidelines — lowest trust, used as fallback only.
7. Web SearchReal-time internet searchIf enabled, the AI can search for weather, local recommendations, directions, and events — used when the knowledge base doesn’t have the answer.

​
How the Classifier Works

Before generating a response, Cendra’s Classifier Agent analyzes the guest’s message and sets business flags:
FlagWhat it meansAI behavior
EmergencyLife-threatening situationEscalates immediately with emergency contact information
Property-relatedQuestion about the property or amenitiesSearches property knowledge base
Availability-relatedBooking or date inquiryChecks calendar availability via PMS
Reservation-relatedQuestion about an existing bookingReferences reservation details
ComplaintGuest expressing dissatisfactionApplies complaint handling rules, may escalate
Upsell opportunityGuest interest in extrasOffers relevant upsell if configured
These flags determine which tools the AI activates — the classifier decision effectively overrides generic conversation context when specific patterns are detected. This ensures emergencies are always handled as emergencies, regardless of what the knowledge base says.

​
Why This Matters

The data hierarchy ensures that:
  • Safety first — guardrails and escalation rules always take priority
  • Your knowledge wins — property documents you uploaded override auto-synced PMS data
  • Context is maintained — conversation history prevents the AI from losing track of what was discussed
  • Fallbacks exist — if the knowledge base doesn’t have an answer, web search provides a safety net

​
Per-Property Knowledge

Knowledge is organized per property. Each property has its own knowledge base, so AI agents respond with property-specific information — not generic answers. If a guest at Property A asks about parking, the AI references Property A’s parking instructions, not Property B’s.

​
Missing Knowledge Detection

When the AI encounters a question it can’t answer from the knowledge base, it:
  1. Flags the conversation as “Missing Information” in the inbox
  2. You see what the guest asked and what knowledge was missing
  3. Add the missing information to the knowledge base
  4. Future similar questions are answered automatically
This feedback loop continuously improves your AI’s ability to handle guest inquiries.

​
Managing Knowledge

  1. Navigate to Agent Hub → Knowledge Base
  2. Select a property
  3. Browse existing knowledge articles
  4. Add new knowledge — enter the topic and content
  5. Edit existing articles to keep them current
  6. Delete outdated information
Changes take effect immediately — no restart or retraining required.