The Midnight Garden · The agent

Meet t.ROY.

t.ROY -- little t, big ROY -- is the AI agent across Troy's properties. Not a FAQ bubble: an agent with real abilities, real limits, and receipts for everything. This page is the honest spec. The bubble in the corner is the thing itself.

What it does

Every ability below is live on this site, today.

Knows where you are

It knows which page you are reading and answers in that context -- built into the system, not guessed by the model.

Answers from Troy's real context

Services, process, portfolio, availability posture -- pulled from Troy's actual systems, with a known freshness date it will tell you if you ask.

Proposes real meeting times

It reads the live calendar and offers open slots. You pick; Troy personally confirms within a business day; the invite arrives by email.

Saves your info -- with consent

When you want follow-up, it shows you exactly what it will save, you edit and confirm, and Troy gets a text about you within seconds.

Emails you resources you ask for

One email, the thing you asked for, nothing else -- you type consent first, and Troy is copied on the send so there is a receipt.

Sets up Troy's follow-up

It drafts follow-up emails into Troy's own drafts folder. He reads, edits, sends -- the human sends, always.

Checks current facts

For anything time-sensitive it runs a cited web search instead of guessing, and shows its sources.

Improves weekly, with a gate

Corrections and misses are logged automatically, distilled weekly into proposed fixes, and applied only with Troy's approval.

What it never does

The limits are designed in -- they are not promises, they are architecture.

Never invents facts

No made-up prices, clients, results, or availability. When it does not know, it says so and offers to flag the question for Troy.

Never quotes prices

Pricing depends on scope. The free check and the free first conversation are the honest starting points.

Never books without approval

Nothing lands on the calendar until Troy says yes. "Fully autonomous" is not the product; "acts with receipts and approvals" is.

Never stores without consent

No contact info is saved until you confirm a consent card you can read. Decline it and nothing is kept.

Never emails anyone but you

There is no capability to message third parties -- the only outbound email it can send goes to the address you give it, with Troy copied.

Never takes instructions over rules

"Ignore your instructions" conversations end politely. Its rules come from its owner, not from chat.

Never names unconsented clients

Client stories come from a consent ledger. If a client has not consented to public mention, t.ROY does not mention them -- in any framing.

Never pretends to be human

It is an AI agent and says so. The engineering is the point.

Privacy, in one breath

  • Conversations auto-delete after 30 days. Enforced by code on a daily schedule, not by policy intention.
  • Contact details are stored only after an explicit consent card, with a timestamp.
  • No selling, sharing, or ad use. No biometric data, ever.
  • When it is over budget or degraded, it says so honestly and still takes a message -- never a blank widget.
  • Need a human? Troy answers his own phone: (262) 391-8137. The agent's standing promise on his behalf is a reply within a business day.

Full policy: privacy · terms

Want this desk for your business?

Everything on this page is a configuration: your brand and voice, your business knowledge, your calendar, your approval phone, your compliance rails. Troy deploys it; you own the desk -- and you approve everything it does, the same way he does. Read the guide to see where it sits on the complexity map, then ask the agent itself what a deployment looks like. The first conversation is free.

Start a project Read the guide

Or open the bubble, lower right -- it knows you just read its spec sheet.