Run capable models locally
Evaluate model behavior, context handling, memory pressure, and application fit on owned hardware.
In-house flagship AI data center machine · Active engineering platform
PROJECT
KEYSTONE
Our in-house flagship AI data center machine is designed, assembled, and configured to run and evaluate AI models locally—with enough GPU memory and system capacity to make the machine itself part of the engineering strategy.
ProcessorAMD Ryzen Threadripper 9960X
CPU capacity24 cores / 48 threads
GPUNVIDIA RTX PRO 6000 Blackwell Workstation
GPU memory96 GB GDDR7
System memory128 GB DDR5-5600
Local storage2 × Samsung 990 EVO Plus 4 TB
Why it exists
AI capability is different when the compute is close at hand.
Project Keystone is Delaney Solutions' in-house flagship AI data center machine, built around a simple idea: the right AI workload can benefit from dedicated infrastructure under our direct control. Local and cloud AI each have strengths; Keystone creates a practical local option for model evaluation, application development, document-connected workflows, and automation experiments.
The verified build uses a 24-core, 48-thread Threadripper 9960X, an RTX PRO 6000 Blackwell Workstation Edition with 96 GB of GDDR7 memory, 128 GB of DDR5-5600 system memory, and two 4 TB Samsung 990 EVO Plus drives. A local Qwen Flash Next model is loaded for current experimentation.
Current status: Keystone is our active in-house flagship AI data center machine and test platform. This page documents the actual build and its role; it does not publish performance benchmarks or present experimental workflows as finished commercial products.
What it enables
A local lab for the systems behind the systems.
The value is not the parts list by itself. It is the ability to test, compare, build, and iterate with local compute under direct control.
Build AI-enabled applications
Prototype local assistants, retrieval systems, automation layers, and workflows that can combine local and cloud models as needed.
Keep local-first options open
For workloads where local processing matters, Keystone provides a path that does not require cloud inference by default.
Test before committing
Compare models, quantization levels, tooling, and integration approaches before choosing what belongs in a production system.
Engineering philosophy
The workstation is a tool.
The system design is the product.
Keystone supports Delaney Solutions' broader approach: use AI where it improves capability, verify the result, and keep the human owner in control of what actually ships.