For boutique and mid-size hotel groups
An operations team for your hotels that never sleeps.
AI agents answer guests in their own language, reply to reviews, nudge room rates and dispatch housekeeping. Your General Manager approves what matters from a single inbox.
Ops floor · Pebble & Pine · 10x
Event stream
Agents
- Dispatcher
- Concierge
- Reputation
- Revenue
- Ops
- GM Briefing
The problem
Thin teams, many properties, and nobody connecting the dots.
2 AM
is when guest messages arrive
Thin night teams across 14 properties mean a simple Wi-Fi question waits until morning, and a complaint grows into a review.
6
hygiene mentions nobody connected
Manali's one-star review was the sixth in 30 days. Each one was answered alone; none became a deep-clean job.
+31%
Diwali pickup, rates unchanged
Jaipur was filling faster than last year while competitors raised prices 9%. Rates were still set by gut feel.
Storylines from the fictional Pebble & Pine Hotels demo dataset.
How it works
Events in. Agents work. Guardrails decide.
Not a chatbot and not a fixed pipeline. A live stream of hotel events, a dispatcher routing each one, and specialist agents handing work to each other. Anything risky waits for a human.
1. Events stream in
Guest messages, reviews, bookings, ops issues, competitor rates, weather and flight disruptions.
2. The dispatcher routes each one
To Concierge, Reputation, Revenue, Ops, GM Briefing. Related spikes become an incident the agents work together.
3. Guardrails decide what happens
Rate floors, compensation caps and banned phrases are code. Safe actions run; risky ones wait for the GM.
A night at Pebble & Pine, in 90 seconds
The self-running tour plays these steps on the live demo. The recorded video lands here with the demo milestone.
- 1 Sign in as Kavya, Group Operations Head. The ops floor is live at 10x speed.
- 2 Inject "Goa flight cancellations". The dispatcher opens an incident and four agents light up.
- 3 Click the Concierge node: its trace shows every tool call and the message it drafted.
- 4 Switch to guest chat: the anniversary couple writes in Hindi and gets an instant reply.
- 5 Approval inbox: approve Jaipur's +12% Diwali rate after checking pickup and comp-set evidence.
- 6 Reputation: Manali's hygiene cluster, the drafted reply and the linked deep-clean ticket.
- 7 Briefing: the morning brief Sameer will read at 06:30.
AI everywhere
Six agents, one autonomy dial.
Each agent is a tool-using loop with typed tools and a six-step limit. Every action it wants to take becomes a proposal with reasoning, evidence and expected impact.
Dispatcher
AIListens to: Every event
Routes each event to the right agent, merges duplicates, and opens an incident when related events spike.
Default: Auto
Concierge
AIListens to: Guest messages, check-ins, flight disruptions
Replies in the guest's language, handles requests, offers upgrades and escalates complaints.
Default: Auto for info · Ask for upgrades
Reputation
AIListens to: New reviews on five platforms
Drafts replies, links each review to its stay and tickets, and clusters recurring issues.
Default: Auto for 4-5★ · Ask for 3★ and below
Revenue
AIListens to: Booking pace, cancellations, comp-set rates, events, weather
Proposes rate changes inside floors and ceilings, opens or closes inventory by channel.
Default: Auto within ±5% · Ask above
Ops
AIListens to: Housekeeping and maintenance issues, staff absences
Dispatches staff, tracks SLAs and batches preventive maintenance.
Default: Auto
GM Briefing
AIListens to: 06:30 IST daily, Mondays weekly
Writes each GM's morning brief and the weekly group report from read-only analytics.
Default: Auto (read-only)
The autonomy dial
Set per agent and per action type. Every level is logged.
- Off
- Suggest
- Ask
- Auto
Guardrails are code, not prompts
Rate floors and ceilings, a ₹2,500 compensation cap, banned phrases like "refund confirmed", and one rule above all: an agent can never approve its own proposal.
Calculated
Data at a glance
24 months of a 14-hotel chain.
From Udaipur's lakeside haveli to Mumbai's business district: Goa peaks in December, hill stations in summer, business hotels Monday to Thursday.
14
1,070
2,30,000
42,000
3,00,000
26,000
Built with
Three languages, one live system.
A Go simulator streams realistic hotel events into Redis Streams. A TypeScript agent runtime on Hono consumes them, calls typed tools, checks guardrails and stores every step as an event, which powers traces and replays. The SvelteKit ops floor follows it all live over WebSocket. To keep an always-on demo affordable, simulated events replay precomputed agent output; only visitor actions call the model.
Agents and API
- Node.js
- TypeScript
- Hono
- Drizzle ORM
- Zod tools
- WebSocket
Web
- SvelteKit
- Svelte 5
- Tailwind
- Svelte Flow
- MapLibre
- ECharts
Simulator
- Go
- Redis Streams
- YAML scenarios
- Seeded RNG
Data and AI
- PostgreSQL
- pgvector
- Redis
- Azure OpenAI
- Local embeddings
Built by AIGyde Solutions
We build agent systems that are safe to leave running: typed tools, hard guardrails, human approval where it counts, and a trace for every decision. If your properties run on thin night shifts, let's talk.