A standard application library your employees open on day one, a governance layer your IT office controls, and managed operations under one SLA. The operational machinery stays on our side, and it keeps working because we keep it working.
Hardware, model serving, and the engineer-facing management plane are solved problems, so we build on NVIDIA AI Enterprise and certified enterprise hardware as they ship. Our engineering goes to the layer the substrate leaves open: the finished applications and the governance experience on top.
Three things, delivered together and priced as a configured deployment. Customer differences are expressed as configuration over a fixed catalog, which is what keeps a Freehold deployment fast to stand up and predictable to run.
Enterprise chat and content creation, retrieval over your own corpus, and a private coding assistant in the IDE, plus add-on applications from a growing catalog. Every app arrives finished.
Model assignment per app and per group, access and roles, per-person budgets, egress posture, and a content-free audit trail, all operated from a console built for IT generalists.
We operate the platform, gate every upgrade, watch capacity, and hold on-call. Hardware field service is delivered by the vendor as our subcontractor, so your recourse is always us.
The models on the menu and the models loaded in GPU memory, ready to answer instantly, are two different numbers, and VRAM sets the ceiling. The menu is tiered around that fact, with each tier's response behavior disclosed before anyone selects it, and every hosted model shows its capacity with provenance: an industry-benchmark estimate at first, then a number measured on your own hardware.
| Tier | What it is | What you're promised |
|---|---|---|
| Hot | Pinned in GPU memory, occupying its footprint continuously. The tier for everyday chat, retrieval, and coding models. | Answers instantly, always. |
| Available | Loaded on demand from the local cache, freeing its capacity while idle. The tier for the specialized tail. | A stated first-call wait, disclosed in the console before anyone selects the model. |
| External, under your keys | Frontier and hosted open models reached through your gateway, with per-person allowances. | Capacity taken from the provider's own rate limit, with the data boundary labelled and every call in the audit trail. |
Bring your own keys. Claude, Gemini, and other frontier providers plug in under your own accounts. Per-person allowances cap spend, and when an allowance is spent the app falls back automatically to a local model. Cross-vendor optionality keeps your deployment portable: each provider is one entry on your menu, under your keys.
You bought "no ML team, never touch the console," so operating what's underneath is our job, permanently:
The accountability chain is deliberately simple: you hold us, and we hold our subcontractors. Dell, HPE, or Lenovo field engineers replace a failed GPU in your building; Red Hat and NVIDIA fix defects in their layers; none of that is a second vendor relationship you manage.
The one thing we never hand off is the control plane: the surface where models are deployed, access is set, egress is enforced, and your audit trail lives. It runs inside your perimeter, it holds your keys, and its custody is precisely what you're paying us to own.
Why the control plane stays inside →The catalog stays coherent because every application, whether first-party, adopted open source, or partner-supplied, enters under the same contract:
Real product names, real versions, a labelled publisher. You're trusted to know what you're running.
Access, egress, and model assignment enforced by the same gateway and policy layer as every other app.
Usage measured the same way everywhere, feeding budgets, capacity planning, and your posture decisions.
Updated across the fleet through the signed release path. An app you add today is an app that stays maintained tomorrow.
Tell us your posture requirements and the applications your teams need. We'll walk you through a configured deployment, priced from the catalog.