Fixed prices, published in public.

Across our published fixed-scope packages the weighted average engagement is $10,000. Individual work starts at $4,000 for a private AI assistant on a single node; business platform builds run $12,800–$27,500; multi-region and regulated estates start at $60,000. Those are list prices you can hold us to, not an average of past invoices.

No. Our fees cover design, build, testing, documentation and handover. Hardware, GPU rental, cloud consumption and software licences are billed to you directly by the vendor, so you keep ownership and there is no markup hidden in our invoice. We will size and source it for you at cost plus 10% if you want us to handle procurement.

50% to schedule the work, 50% on acceptance. Engagements over $20,000 can be split into three milestone payments. We accept ACH, wire and card; net-30 terms are available for established businesses.

Both, and the engineering standard is the same either way. Our individual packages suit researchers, creators and small practices building private AI or a home cloud; our business packages suit companies from five to several hundred people. We expect most engagements to be individual work, because that is where the unmet need is largest.

Any open-weight family — Llama, Qwen, Mistral, DeepSeek, Gemma and the Whisper/embedding/image models around them. We benchmark two or three candidates against your actual prompts and hardware before recommending one, and the stack lets you swap models later without re-architecting.

A single 24GB GPU comfortably runs a quantised 24–32B model for one to five concurrent users — roughly $2,000–$9,000 of hardware. Team-scale serving with 70B-class models typically means 2–8 professional GPUs, or rented GPU capacity while you validate demand. We size this in discovery, before you buy anything.

Yes. We start with a $2,400 audit that documents what exists, flags the risks and prices the fixes. Many clients stop there, apply the roadmap themselves, and come back later for the parts they would rather not own.

No. Everything is standard open tooling — Terraform/OpenTofu, Docker, Ansible, Prometheus — in your repositories and your accounts, with runbooks written for someone who is not us. Care plans are month to month with 30 days notice.

We are a Wyoming LLC based in Sheridan and work remotely with clients across the United States, with on-site visits available for colocation and hardware build-outs. Most delivery is remote; discovery calls are scheduled in your timezone.