Jack uses AI to understand guest intent, generate natural responses, and decide when to escalate to staff. This page explains how the AI system works.
How It Works
When a guest sends a message, Jack:
- Classifies intent — Determines what the guest wants (question, request, complaint, etc.)
- Retrieves context — Searches the knowledge base for relevant information
- Generates response — Creates a natural reply using the AI model
- Decides routing — Either sends the response or escalates to staff
Supported Providers
Anthropic Claude
Excellent at understanding context and nuance. Great for complex guest requests and natural conversations. Recommended for most deployments.
Models: Claude 3.5 Sonnet, Claude 3 Haiku
OpenAI GPT
Fast and reliable. Good balance of quality and speed. Well-suited for high message volumes.
Models: GPT-4o, GPT-4o-mini
Local AI
Runs entirely on your server using Transformers.js. No API costs, complete privacy, works offline. Best for privacy-focused deployments.
Uses: Xenova models via Transformers.js
Ollama
Run open-source models locally with more power than Transformers.js. Requires Ollama server running separately.
Models: Llama 3, Mistral, and others
Autonomy Levels
Control how independently Jack responds to guests:
| Level | Behavior |
|---|---|
| Full Auto | Jack responds to all messages automatically. Only escalates when confidence is low or guest explicitly asks for staff. |
| Suggest | Jack drafts responses but waits for staff approval before sending. Good for training and quality control. |
| Off | AI is disabled. All messages go directly to staff. Use during sensitive situations or high-stakes conversations. |
Change autonomy level in Settings → AI → Autonomy or per-conversation from the inbox.
Knowledge Integration
Jack uses RAG (Retrieval-Augmented Generation) to provide accurate, property-specific responses:
- Guest message is converted to an embedding (vector representation)
- Similar content is retrieved from your knowledge base
- Retrieved context is included in the AI prompt
- AI generates a response grounded in your property's information
This ensures Jack gives accurate answers about your specific check-in times, amenities, policies, and local recommendations.
Response Quality
What Jack handles well
- Frequently asked questions (check-in, WiFi, parking)
- Room and amenity inquiries
- Local recommendations
- Simple requests (extra towels, late checkout)
- Booking confirmations and directions
What gets escalated
- Complaints and negative feedback
- Billing and payment issues
- Requests outside knowledge base
- Complex or unusual requests
- Guest explicitly asks for a human
Cost Estimates
AI API costs depend on usage:
Typical costs for a 20-room hotel with ~500 messages/month:
- Claude (Sonnet) ~$5-15/month
- OpenAI (GPT-4o-mini) ~$3-10/month
- Local AI $0
Configuration
See AI Providers Configuration for setup instructions.