private AI infrastructure

Why we built
our own.

Most AI consultants run your business through someone else's public cloud — your data passing through models you can't see, on terms you don't set. plainworks runs sensitive workflows on its own secure, private AI inference and serving infrastructure, built and managed by Nathan. This is the part most studios can't offer, and won't talk about.

the promise, stated honestly
Sensitive workflows can run on plainworks-controlled infrastructure. When an external API is the right tool, you'll know why, what data is sent, and what the fallback is.

No black boxes, and no claims we can't back. That second sentence is the part you won't hear from most vendors — because most of them route everything through a public API and hope you don't ask. We'd rather tell you exactly how your work runs.

where your workflows run

Three places to run it.
Your call, not ours.

We match the deployment to your sensitivity, budget, and control needs — and we tell you the trade-offs of each in plain terms.

plainworks private data center

Our own secure inference and serving stack. The most private option — your sensitive workflows never touch a shared public cloud, and the hardware is operated in-house.

Private cloud

A dedicated, access-controlled cloud environment when you need elastic scale with stronger isolation than a shared public endpoint.

On-premise servers

The model runs on hardware in your building, behind your network. For workflows where data simply cannot leave the premises.

Most engagements use local-first routing: we keep the work on plainworks-controlled infrastructure by default, and reach for an external model only when it's the right tool — disclosed and approved, never silent.

for the technically minded

The stack,
in full.

plainworks · production cluster
private data center · plainworks office
operational
01

Inference & fine-tuning

ComputeNvidia Spark GB10 cluster
Interconnect200 Gbps ConnectX-7 switch fabric
02

Servers

PrimaryAMD Ryzen 9 5900XT · 16C / 32T · 32 GB
BackupIntel Core i9-10850K · 10C / 20T @ 3.6 GHz · 32 GB
03

Storage & network

Storage20 TB RAID hybrid SSD / HDD
InternetComcast Business 1.25 Gbps
WAN failoverVerizon 5G backup
04

Applications

BuildCustom, modular apps on thoroughly tested open-source bases
Served fromplainworks private data center, in-office
05

Triple redundancy

BackupsDocker containers snapshotted 2× daily → local backup server + AWS
Failover10-min UPS bridges to tertiary AWS cloud on extended outage
06

Monitoring

Uptime watch24 / 7 / 365 cloud-AI monitored & auto-maintained
EscalationImmediate human intervention on any fallback or failure

Specs evolve as the cluster grows — this reflects the current production environment.

what this means for you

Control you can
actually point to.

01

Your data stays yours

It isn't used to train anyone else's model. Your workflows, metrics, and customer details stay private.

02

Private by architecture

Sensitive work runs on plainworks-controlled infrastructure — not a shared public API by default.

03

Transparent routing

Every external model call is disclosed: what's sent, why, and the fallback. You approve the boundary.

04

Run by one accountable person

The stack is built and operated by Nathan — not a reseller badge, not an outsourced ops team you'll never meet.

trust is more than hardware

Private hardware is
necessary, not sufficient.

Owning the infrastructure is the easy headline. Real trust comes from managing risk across the whole lifecycle of a system — the way mature AI risk frameworks describe it. We design for that, not just for a server in a rack.

design

Scoping what data a workflow actually needs — and excluding what it doesn't — before a line of it is built.

evaluation

Testing outputs against real cases, with a human in the loop where judgment matters.

privacy

Clear boundaries on storage, retention, and who can see what — written down, not assumed.

operation

Monitoring in production, with the ability to roll back, retrain, or pull a model if something drifts.

honest about limits

A readiness roadmap,
not a compliance badge.

For regulated industries, we'll talk openly about a readiness roadmap — the controls, contracts, and scope it would take to operate there. What we won't do is slap a compliance acronym on a slide before it's legally and operationally true.

If a vendor leads with a certification logo on day one, ask them to show you the controls behind it. We'd rather earn the claim than borrow it.

our standard
  • We name what we can do today, plainly.
  • We name what we can't yet — and what it would take.
  • We disclose every external model boundary.
  • We put data handling in writing before we build.
  • We don't make claims we can't substantiate.
why it's built this way

Technology in service
of people.

Empower, don't replace

The goal isn't to cut your people. It's to take the operational drag off them so they spend time on the work only humans can do.

Give time back

More room for the relationships, the craft, and the specialty you actually got into business to do — without scaling the hours to match the growth.

Sustainable by intent

Right-sized models on efficient infrastructure — we don't burn a data center's worth of compute to send a follow-up email.

Value, not theater

Every system has to earn its place by closing real profit leakage. If it doesn't pay for itself, we don't build it.

your data, your call

Have a workflow that
can't leave the building?

Tell the consultant what you're working with. We'll talk through where it should run, what it would take, and where the honest limits are — before anyone signs anything.