Setups
Three ways to set up your agents
Same agents, same jobs, same contract language about who owns what. The choice is where the machine lives and where the model work happens: from a rented cloud computer you can start on this month, to a personal data center that never lets your data leave the building.
Not sure where you land? Take the fit check eight questions, honest answer, nothing leaves your browser.
Setup 01 · Fully cloud · hosted by Orgo
The Outpost
Lowest upfront cost, fastest start.
Your agent stack on a dedicated cloud computer built for AI agents. Own the stack, rent the metal.
How it runs
- Your agents, their memory, and the audit log run on a dedicated Orgo virtual machine: a full computer purpose-built for AI agents, not a shared chatbot service.
- All model work goes to frontier cloud models (Claude and its peers) through metered API accounts opened in your name. Business terms, retention off, no training on your data.
- Everything is built portable by design. The same stack we install on a Start9 can be packed up and moved onto hardware you own, later, without a rebuild.
The move-home path is part of the deal. When you’re ready, we install a Start9 (Frontier) or a full personal data center (Homestead), migrate the stack, and turn the Outpost off. It’s a hardware install plus a migration day — not a rebuild.
- Models
- Frontier cloud models only, through accounts you own
- Data at rest
- On your dedicated Orgo VM (rented, single-tenant, yours to export any day)
- Hardware
- None to buy. The machine is rented.
- Build (draft pricing)
- $4,500–$8,000 build · hosting passed through at cost
- Monthly shape
- Care plan + ~$50–$250/mo hosting + metered model APIs
Stated plainly
- Your data sits on rented infrastructure: contractual protection, not physical possession.
- A monthly hosting bill exists as long as you stay in the cloud.
- If the internet is down, so is everything (there is no local machine).
Best for: Offices that want agents working this month with the smallest upfront spend, handle mostly non-regulated data, and like the idea of moving home later.
Setup 02 · Hybrid · Start9 server + cloud models
The Frontier
Owned server, cloud brains when useful.
A Start9 server in your building houses the agents. Frontier cloud models do the heavy reasoning. The models your server can handle run at home.
How it runs
- A Start9 server in your building is home base. Agents, memory, customer records, and the audit log live on it, sandboxed, encrypted at rest, behind your locks.
- Frontier cloud models handle the demanding reasoning through metered accounts you own. The workloads your server’s CPU (or a modest GPU) can carry, like triage, sorting, and templated drafting, run locally.
- The blueprint pins any workflow too sensitive for the wire to a local model, workflow by workflow, in writing.
Adding a GPU machine later upgrades a Frontier to a Homestead without touching the hub. The agents don’t move; the brains do.
- Models
- Cloud models for heavy reasoning + local models for what the box can carry
- Data at rest
- In your building, on hardware you own
- Hardware
- Start9 server (or Apple Mac mini, if you prefer that ecosystem)
- Build (draft pricing)
- $9,000–$18,000 fixed price · hardware included
- Monthly shape
- Care plan + smaller metered API usage + ~$20/mo electricity
Stated plainly
- Cloud-bound requests still cross the internet (under business terms).
- Local model capacity is real but limited. The heavy thinking stays metered.
Best for: The middle most businesses land on: records stay home, the strongest models stay available, and the monthly bill stays modest.
Setup 03 · Fully local · personal server + AI GPU hardware
The Homestead
Maximum local control.
The complete personal data center. Your server plus dedicated GPU hardware runs full local AI, and cloud models remain available as a tool when you want them.
How it runs
- Everything the Frontier has, plus a dedicated inference machine: an NVIDIA GPU box or Apple Silicon (Mac Studio–class) running serious open-weight models.
- Day-to-day agent work (reading, writing, talking, document analysis) happens entirely inside your walls. Overnight jobs run free on hardware you’ve already paid for.
- Cloud models stay on the menu as a metered tool for the rare task that wants the frontier. Your choice, workflow by workflow, off by default for sensitive data.
This is the destination tier. From here the only moves are bigger hardware or newer models, both of which slot in without a rebuild.
- Models
- Full local AI on your own GPU · cloud models optional, as a tool
- Data at rest
- In your building, on hardware you own. Nothing needs to leave.
- Hardware
- Start9 or Apple hub + NVIDIA GPU box or Apple Silicon (Mac Studio–class)
- Build (draft pricing)
- $14,000–$28,000 fixed price · all hardware included
- Monthly shape
- Care plan + ~$40/mo electricity · $0 mandatory cloud spend
Stated plainly
- The largest upfront spend of the three.
- Open models on your own hardware are superb at the daily jobs. For the deepest reasoning tasks the frontier is still stronger, which is why it stays available as a tool.
Best for: Clinics, law offices, accountants. Anyone with regulated or privileged data, high volume that would be expensive by the token, or the conviction that the machine should be theirs.
Side by side
| The Outpost | The Frontier | The Homestead | |
|---|---|---|---|
| Where your agents live | Dedicated Orgo VM (rented) | Start9 in your building | Your personal data center |
| Model work | Frontier cloud APIs | Cloud + what the box carries | Full local AI · cloud optional |
| Data at rest | On the rented VM | In your building | In your building |
| Upfront (draft) | $4,500–$8,000 | $9,000–$18,000 | $14,000–$28,000 |
| Monthly shape | Hosting + APIs + care | Small APIs + power + care | Power + care · cloud optional |
| Internet down? | Everything pauses | Local work continues | Local work continues |
| Regulated data | Case-by-case, with counsel | Good: pin it local | The clean answer |
Build prices are published drafts; your blueprint sets the fixed price. Hosting and API costs pass through at cost, through accounts in your name.
The hardware menu
On the Frontier and the Homestead, the machines are yours and the choice is yours. We spec, procure, burn in, and install. You own the receipts.
Start9 server
The hub, StartOS
Our default home base: purpose-built for running your own services, every app sandboxed, data encrypted at rest, an interface an owner can actually use. The most sovereign option per dollar.
Apple hardware
Hub and/or inference
A Mac mini can serve as the hub for shops that live in the Apple ecosystem. A Mac Studio’s unified memory runs surprisingly large local models quietly and efficiently, which makes it a strong single-box Homestead.
NVIDIA GPU box
Serious local inference
A dedicated GPU machine for the heaviest local AI: bigger models, faster responses, room for overnight batch work. Paired with a Start9 or Apple hub in full Homestead builds.
Every combination ships with the same operations binder, the same signed statement of ownership, and the same care-plan coverage.
The price sheet
| Hardware | Price | As of |
|---|---|---|
| Server One - 6600H2TB / 16GB | $899 | live from the Start9 store · 2026-09-18 |
| Server One - 6600H2TB / 32GB | $1,099 | live from the Start9 store · 2026-09-18 |
| Server One - 6600H4TB / 16GB | $1,199 | live from the Start9 store · 2026-09-18 |
| Server One - 6600H4TB / 32GB | $1,399 | live from the Start9 store · 2026-09-18 |
| Mac mini (M4)16 GB / 512 GB: hub duty, light local models | $799 | Apple published price · checked 2026-07-14 |
| Mac mini (M4 Pro)24 GB base, configurable to 48 GB | $1,599 | Apple published price · checked 2026-07-14 |
| Mac Studio (M3 Ultra)96 GB unified memory, the largest-memory Mac sold new | $5,299 | Apple published price · checked 2026-07-14 |
| RTX 5090-class inference boxcomplete prebuilt system | $3,600–$4,700 | typical range · checked 2026-07-14 |
| Workstation-class local-LLM buildhigher-memory custom build | $5,000–$8,000 | typical range · checked 2026-07-14 |
Hardware prices moved 15–50% in the ninety days before these were checked. Start9 prices refresh from the Start9 store at every site deploy; Apple prices change only at Apple announcements; GPU systems are quoted as ranges because street prices swing weekly. Your blueprint locks exact hardware at signing.
Why Orgo for the cloud tier
Orgo (orgo.ai) builds cloud computers specifically for AI agents: full, dedicated virtual machines an agent can operate like a real computer, not a shared chatbot backend. Your Outpost is a single-tenant machine in your name. We build on it, you hold the keys, and the hosting bill passes through at cost. If Orgo ever stops fitting, the stack is portable. That’s the point of owning it.
Full pricing detail is on pricing.
The architecture under all three
A personal data center is the hardware and software you own and run yourself. Instead of renting compute from someone else’s cloud, the equipment sits in your building. You hold the keys, the data, and the off switch.
The Start9 hubWhere your agents live. Always.
A Start9 server running StartOS is the heart of every installation: the quiet box on the shelf in your back office. Your agents run on it. Your customer records, call transcripts, agent memory, and audit log live on it, encrypted at rest.
The local GPUOwned intelligence, for the sensitive and the steady.
The second layer is a dedicated inference machine on the same shelf, an NVIDIA GPU box or Apple Silicon (Mac Studio–class) if you prefer that ecosystem, running open-weight models. It handles the reading, writing, and talking your agents do all day, plus the work you’d never send outside: analysis of patient files, client matters, and the books.
Cloud APIsA tool, not a landlord.
The third layer is the frontier: cloud models (Claude and its peers) reached over APIs, for customers who want the strongest available reasoning behind their agents.
The always-on caretaker
A personal data center includes something a pile of subscriptions never will: an agent whose job is the stack itself. It lives on your server around the clock, watching that services are up, that backups ran and verify, that disks are healthy, that certificates renew, and it writes up what it sees.
The deeper technical version of this page lives at personaldatacenter.xyz, the founder’s field manual for the same stack we install commercially.
The blueprint prices every setup that fits.
The free assessment finds your fit; the blueprint maps your workflows, sorts your data obligations, and fixes a price for each viable path, in writing.