services
I install private AI on your hardware, then keep it healthy.
Two offerings, deliberately simple: a one-time setup with a fixed scope, and an optional monthly caretaker plan you can cancel any time. Hardware is yours, bought at cost — I don’t mark it up.
one-time setup
Essentials
from $2,500
Local AI on a machine you already own. One model, one working workflow — drafting, dictation, or document search — plus training until you're comfortable, and a written runbook.
Practice
from $4,500
A dedicated machine specced for your workload. Two or three workflows, an agent with fail-closed guardrails and action receipts, a data-custody write-up you can show whoever audits you.
Firm
custom
Multiple seats, shared local server, integration with your existing tools. Scoped and quoted after the intake call — no number invented before I understand the work.
Every setup ends the same way: we unplug the network cable and run your workflow. If it doesn’t work offline, I’m not done.
caretaker plan
Local AI is an appliance: it needs occasional model updates, health checks, and someone to call. Month to month, cancel any time — the system keeps working without it.
Watch — $150/mo
Updates and model refreshes, health monitoring, backup verification, async email support.
Care — $250/mo
Everything in Watch, plus live support when something breaks and a quarterly tune-up where we review what the system did and adjust.
founding clients
The first three clients get a substantial discount on setup and a locked first-year caretaker rate, in exchange for something I need more than the money: an honest, publishable case study of your installation — what worked and what didn’t — and a reference call. Three slots, stated plainly because that’s the deal.
how an engagement starts
The intake call: one question at a time.
- →What data must never leave your machine — and who you're protecting it from.
- →What the AI should actually do for you on day one. Real tasks from last week, not science fiction.
- →What hardware you have, and what it can honestly run.
- →Your boundaries: by default the agent only drafts — it acts on nothing without your approval, and “it refuses and tells you why” is correct behavior.
You leave with a one-page plan: your data classes, the models I’d use and why, exactly what the agent may and may never do, and a price. It’s yours whether or not you hire me.
Fair warning: I’ll talk you out of this if it’s wrong for you. If you need frontier-model quality on a thin laptop, or want an AI acting autonomously on live client data from day one, local-first isn’t your answer yet — and I’ll say so on the call.
Book an intake call