VYROX AI projects are custom built, so each one is scoped and quoted on its own. This page explains what shapes a project, what it includes, and what we need from you to prepare a written quotation.
Five steps take you from the first conversation to a working system.
In a free 45-minute session we go through your use cases, data, systems and security rules.
We recommend the models, hardware, integrations and support package that fit, and set out what is in and out of scope.
You receive one written quotation with an inclusion list and an exclusion list side by side.
We source, assemble, burn-in test, install and tune the system, then connect the agreed documents and systems.
Your team is trained, the documentation is handed over, and your chosen support package begins.
Two businesses of the same headcount can need very different systems. These are the variables that decide it.
Swipe to See All Columns
| Variable | Expands the Scope | Keeps the Scope Smaller |
|---|---|---|
| Concurrent Users | Many people generating at the same moment | Intermittent office use, where roughly one person in five is actively generating |
| Model Size | A 70B or larger model for the hardest reasoning work | A 27B to 32B class model, which handles most business tasks |
| Context Length | Long documents held in memory for every session | Shorter prompts, or retrieving the right passages instead of loading whole files |
| Quantization | FP16 or Q8 weights when maximum fidelity is required | Q4_K_M, the common default, which cuts the memory requirement substantially |
| Agent Workload | Automations running around the clock | Human-initiated use during working hours only |
| Redundancy | A second unit so there is no single point of failure | A single machine with an accepted manual fallback |
| Integration Scope | Several systems connected, each needing its own mapping and testing | Starting with one document store and adding systems later |
| Platform Choice | Multi-GPU NVIDIA builds for batching throughput | Apple Silicon unified memory, which reaches large models at lower throughput |
| Support Package | Immediate 24/7 support for business-critical workloads | A lighter package or self-managed operation with your own IT staff |
Every project is delivered as a commissioned, working system, not a box of parts.
We spec the GPU, CPU, memory and storage for your model and team, source them, then assemble and burn-in test before delivery.
Ollama, vLLM or LM Studio installed and tuned, with your chosen open model downloaded, quantized and load-tested at your real context length.
Connection to your documents and agreed systems, such as accounting, case files, EMR or POS, via MCP (Model Context Protocol).
On-site or remote training for the people who use and administer the system, plus written documentation.
Hardware and setup are covered for 12 months from commissioning.
Support is set by the package you choose, up to immediate 24/7 support.


These items depend on your site and systems, so they are only included when the quotation says so.
Ask any vendor, including VYROX, to put the exclusion list in writing next to the inclusion list.
Use this framing to turn a specification into a request your finance lead or board can sign.
One line for the one-time build, one line for the annual running cost.
Put your current AI subscription spend directly beside the new lines.
Hardware is normally handled differently from a monthly service fee, but the treatment depends on your entity and your auditor.
For clinics, firms and agencies, documents staying in the building is often the point that closes the approval.
State who administers the machine, who decides which documents it reads, and who reports back at 90 days.
Book a free 45-minute Local-AI Audit. We review your use cases, recommend the right system and prepare a written custom quotation, with no obligation.
No sales deck. Just engineers scoping your project.
Free Local-AI Audit