7 roles
row-level security enforced server-side — a technician's own data only, never the company's
Your whole shop, in one governed layer —
and everyone sees only their part.
One data warehouse pulled from the tools an HVAC company already runs, exposed two ways: a live owner & controller dashboard, and a role-secured MCP connector for any AI assistant. Every number is scoped to the person asking by the same security layer, so a technician never sees the company's payroll.
1 source
the dashboard and the AI read the same governed gold layer, so their numbers always agree
~$4.5M
a full simulated Massachusetts HVAC company — jobs, payroll, vendors, customers, fleet
Two ways in, one governed layer
Synthetic data · Meridian Comfort Systems📊
Live
Owner Dashboard
For owners & controllers
The whole business at a glance across five tabs — finance, customers, operations, products & vendors, people & payroll — with big-number KPIs, trends, and tables rendered live from the data layer. Each role sees only the tabs its grants allow.
Open the dashboard →
🔌
Live
MCP Connector
For Claude, ChatGPT & any MCP client
Add the business to your own AI assistant as a remote MCP connector and ask it anything — "what's my open A/R over 60 days," "which techs are running overtime." The same role scoping applies: the assistant can only reach what your role is allowed to see.
Set up the connector →
📱
Coming soon
Field App
For technicians on a phone
A thin mobile answer pane for the field: spec and clearance lookups, part identification, and pre-job checks, each answer cited to the manufacturer's manual or the code — and honest when a spec isn't in the corpus, rather than guessing.
In development
Why you can trust it
The same rules govern the dashboard and the AI01 Role-scoped at the source
Access is decided in the data layer, not the screen. A tool or a dashboard tab only returns what the caller's role and row-scope allow — so a field tech sees their own pay, never the company's, and the boundary holds no matter how the question is phrased.
02 Grounded in one warehouse
Every figure traces to the same governed gold layer, built from the systems a shop already runs. The dashboard and the AI answer from that one source, so a number on a chart matches the number the assistant gives — no second version of the truth.
03 Honest about its limits
Where a figure is modeled rather than measured, it says so. Where the data can't answer, it says that too, instead of inventing a number. The value is a dependable answer, not a confident guess.