Skip to content
ratliff.ioSmall Business CTO

Example setup

Office AI, private by design

A Mac mini in the office routes the work, a local model server handles it, and cloud services are called only when a task needs them. Everything is backed up to a NAS and again, encrypted, offsite.

Hardware
OFFICE LAN · TAILSCALE MESHCLOUDTRIAGED?FILED · SENTiPhonehow staff reach itMac mini M4Hermes gateway, always onChief of Staff agentiMessage bridge · cron · skillsMac Studiogateway + models, one boxM5 Ultra, up to 512 GBno separate model serverHermes gateway, always onChief of Staff · iMessage bridgeAMD Ryzen AI Halolocal models, the default176B MoE · ~35-40 tok/s w/ MTP~36 tok/s at ~147K ctxno vision · 261K contextRyzen AI Max+ 395 · 128 GBreachable on LAN onlyNASgit remote · SMB targetrepo history · watchdog stateSystems of recordemail · calendar · docs · CRMagents act via scoped OAuthFrontier modelsvision fallback · hard reasoningused sparingly, local firstOffsite backupencrypted before it leaveskeys stay in the officeiMessagelocal APIlocal APIscoped OAuthon demand onlygit push · SMBencrypted, then sent offsite
Messages Local inference Cloud, on demand Backup & replication

How it works

Three ideas holding it together

Local first

A local AI machine runs the everyday model —the Halo. Routine tasks — triage, drafting, filing — never leave the office network. The cloud is used only when a task earns the trip.

One office hub

A small server runs the assistant gateway around the clock: the Chief of Staff routes email, calendar, filing, and messages, and runs scheduled reports on a schedule.

Backed up twice

Git remotes and SMB replication land on the office NAS, a watchdog checks the backups themselves, and the offsite copy is encrypted end to end — keys stay in the building.

Sizing it

Built around your office

Hardware and services vary by office — the shape stays the same. Want this sized to your office: hardware, models, and what should stay local?

One conversation to scope it.

Start a conversation