Custom builds
Towers built in Nashville, for people who know what they want
Our business work is installing private AI appliances for offices. These are the machines we build ourselves: six tested configurations for developers, researchers and power users, and fully custom builds on request. Priced at parts cost plus 20%. Windows 11 Pro, assembled and tested here, local pickup available.
Buying for an office? Start with the levels and the install packages instead.
Six tested towers
Configurations we have built and benchmarked ourselves. Each ships assembled, tested, with Ollama and Open WebUI ready on Windows 11 Pro.
Starter
Your first local AI machine
Capabilities
- 7-20B parameter models (gpt-oss-20b, Qwen3 8B)
- A private assistant for one person
- Local code completion & chat
Pro
The sweet spot for power users
Capabilities
- 13-30B parameter models (Gemma 3 27B, Qwen3-Coder 30B)
- Document search over your own files, one user
- Multi-model inference & RAG pipelines
Ultra
No compromises
Capabilities
- 32B models with partial offload (Qwen3 32B, DeepSeek-R1 32B)
- Two models loaded at once (chat + code)
- Fine-tuning & model training
Premium Apex Builds
Next-generation hardware for serious AI work. Run 32B models at full speed, or step up to the 96 GB Blackwell card for 70B-class models, and deploy production AI infrastructure from your desk.
Apex
Next-gen single card powerhouse
Capabilities
- 1,792 GB/s memory bandwidth — fastest consumer card
- ~213 tok/s on 8B models, ~61 tok/s on 32B models
- 32B models fully resident in VRAM — no offload needed
Apex Dual
Two brains, zero compromise
Capabilities
- Run two models simultaneously — coding + chat, both hot in VRAM
- A dedicated 32B model per card — both at full speed, no splitting
- One card per model: a document model and a coding model, both at full speed
- H100-class performance at a fraction of the cost
Apex Pro
96GB single card — no compromises, no splitting
Capabilities
- 70B models at full FP16 — zero quantization needed
- 120B+ parameter models at Q4 on a single card
- ECC VRAM, workstation drivers, built for 24/7 operation
Build Your Perfect AI Machine
Have specific requirements? Tell us what you need and we'll design a custom configuration.