FAQ
Questions, answered straight.
No jargon, no pressure. Here is how private, on-premise AI actually works, what it costs, and how to start.
No jargon, no pressure. Here is how private, on-premise AI actually works, what it costs, and how to start.
We do two things, both privately. First, we stand up private AI infrastructure on hardware you own, the servers, GPUs, models, and the open stack that runs them. Second, we build the custom AI agents and workflows that run on top of it. Everything operates inside your building.
Nowhere. The model, the documents it reads, and the answers it produces all stay on your premises. There is no third-party API in the path of a query, so your prompts and files are never sent to OpenAI, Anthropic, Google, or anyone else.
For most business work, no. Around 80% of typical tasks, summarizing documents, answering questions over your knowledge base, drafting, classification, and defined workflows, run fine on open models you can host yourself. A smaller model fine-tuned on your data often beats a generic frontier model on your actual work.
Plans run from a Starter tier at $499/mo to a Sovereign tier at $5,999/mo, covering the build, management, and support. Hardware passes through close to cost, roughly $20k to $180k depending on tier, and you own it outright. See the pricing page for the full breakdown.
Yes. The servers and GPUs are yours. We size them, source them close to cost, install and harden the stack, and then manage the system, but the asset sits on your books and in your building.
The architecture is built for it. Because regulated data never leaves your control, the hardest parts of a HIPAA or CMMC review become straightforward. We handle the security posture, access control, audit logging, and air-gapping as part of the build, and map it to the framework you answer to.
A typical build runs a few weeks from spec to a running system: sizing the hardware and model, installing the stack on-premise, fine-tuning on your documents, locking it down, and handing your team a working interface. Larger or air-gapped deployments take longer.
That is exactly the risk of renting a frontier model, and it is why we run open weights you hold yourself. When a hosted model is restricted or retired, nothing on your side changes. You own the weights and the system keeps running. We wrote about this after the Fable 5 shutdown.
Yes. For the most sensitive deployments the system runs with no route to the public internet at all. It does not need the cloud to function, because the model and everything around it are local.
We do, under contract. We monitor performance, patch the stack, and apply deliberate, tested updates so the system stays fast, current, and secure. You are not left to operate it yourself.
From a small practice running a single GPU for private chat and document work, up to a firm running a large model company-wide. The tiers exist so you can start small and grow into it.
We are onboarding our first New Jersey clients now. Sign up to learn more, tell us what you need to keep private, and we will come back with a straight read on fit, tier, and rough numbers.