All work
Solnest Stays/Short-Term Rentals/Prince George & Sun Peaks, BC/2026

The operating system for a rental portfolio.

Building the operating system for a short-term-rental portfolio.

AI AutomationSoftware DevelopmentBusiness Consulting
Live in productionapp.solneststays.com
app.solneststays.com
Solnest Stays
Real-time STR operations · 4 properties
Ask AI
This MonthLast MonthThis Year
Revenue · mo
$4k
▼ 58%
Occupancy
58%
▲ 20%
ADR
$318
▲ $88
Rating
4.9★
129 rev
On books
$14k
11 bkd
Revenue · 7d
Occupancy · 7d
ADR · 7d
RevenueGuestOperationsAnalyticsMarketing4/7 active

The live command center · app.solneststays.com

Prologue

The problem on Ryan's desk

Ryan runs four short-term rentals, two in Prince George, two up the mountain in Sun Peaks. Two markets, two seasons, four sets of keys, and a phone that buzzed at 1 a.m. when a guest couldn't find the lockbox code.

Every part of the operation lived in a different tab. Hospitable held the bookings, PriceLabs set the rates, Notion held the manuals, GoHighLevel passed for a CRM. None of them talked to each other, and none could answer the only two questions that mattered on a Sunday night: are we pacing ahead of last year, and is anything on fire right now?

This is the system that replaced the tabs. Seven agents and twenty-nine workflows quietly pull, price, answer and alert around the clock.

What shipped

The system, in four moving parts.

01

A revenue engine that prices itself

A deterministic Python pricing engine (fifteen modules, a hundred and sixty-four tests) reads comps, pace and orphan nights each night and files nearly two thousand pricing decisions. The model never sets a price; the numbers do.

02

A guest agent that drafts, never sends

Guest messages arrive HMAC-signed, get classified and answered against each property's knowledge base, then wait for a human tap. 273 messages handled, a 37-minute average response, 69% positive sentiment.

03

Operations & analytics briefs, every morning

Turnovers, arrivals and red flags synthesised into a daily brief; seventeen signal detectors that file written insights, then grade their own calls a week later.

04

One orchestrator, twenty-nine workflows

A CORTEX cycle runs the whole fleet on demand, but most days the agents fire on their own cron schedules. The portfolio runs unattended, and version-controlled as code.

How it’s wired

One pipeline. One source of truth.

Each tool writes into its own staging path, gets normalised, and lands in one canonical schema, so everything reads from a single, trustworthy source.

01 · Your tools
Sources
HospitablePriceLabsNotionGoHighLevelOpenRouter
02 · Orchestration
n8n

Pulls, normalises, schedules

03 · One source of truth
Supabase

Every record, reconciled

04 · In your hands
Live dashboard

Real-time, always current

Built on
ReactNodeSupabasen8nHospitablePriceLabsGHLOpenRouter
Results
7
AI agents in the fleet
29
n8n workflows orchestrating
4.9
guest rating · 155 reviews
Field notes

The parts that were harder than they looked.

The webhook nobody can forge

A guest message from the internet is never trusted on face value. The chain that carries it is HMAC-signed end to end, verified inside a five-minute replay window, deduped against a nonce, and rate-limited, verified before it was ever trusted with a real guest.

Draft, don't send

The most important product decision was that the agent does not send. "Most of the time" is exactly wrong for the message that tells a guest the wrong check-in date. The owner's reputation never rides on the model being right unsupervised.

The metrics that lived in the future

A rollup once wrote a thousand rows dated into next year. Nothing crashed; the numbers simply stopped meaning anything. A number that is wrong but confident is more dangerous than no number. The activity log made the bad rows findable.

It is the operating system for a short-term-rental portfolio, not a dashboard bolted onto six other tools.