MCMLV1 LLC
Private AI Systems & Software

Private AI Systems

AI infrastructure and software engineering tools designed around the security boundary your organization chooses.

Verified Engineering Result ยท November 2026

GPT-OSS-120B on a Single Workstation

An independently developed inference runtime using two NVIDIA GPUs from different generations, a 10-core CPU and 128 GB RAM. Across six tested prompt lengths, it led tuned llama.cpp and ik_llama.cpp in all 18 prefill, decode and full-response measurements, while matching 2,113 of 2,113 reference next-token choices.

The public report includes hardware, speed charts, full-response times, numerical fidelity and measurement methodology. Results apply to the tested configuration, not every device or model.

View benchmark and slides
Flagship Product

CodeLoop

CodeLoop is MCMLV1’s private AI software-engineering system, currently in active development. It is designed to inspect a project, reason about the work, modify files in a private workspace, build and test the result, recover from errors, and present a reviewable diff before changes are applied.

Local First

Private by Architecture

CodeLoop is designed to work with local and private inference rather than requiring source code to be sent to a public cloud service.

Engineering Loop

Build, Test, Repair

The model proposes work, while compilers, tests, structured tooling, and human review provide the verification loop.

Human Control

Review Before Apply

Changes are prepared for review rather than silently becoming production code. The professional engineer remains responsible for acceptance and deployment.

Choose the Security Boundary

MCMLV1 can design an inference setup around the customer’s hardware, network, security policy, and performance requirements.

◆ Single Workstation

CodeLoop and inference on one Mac or PC for the simplest local deployment.

◆ Private LAN

Developer workstations use a stronger inference workstation or server available only on the organization’s private network.

◆ On-Premises

Dedicated customer-owned inference hardware deployed inside the organization’s facility or data environment.

◆ Isolated Environment

Offline or physically isolated architectures can be designed when the organization requires stronger separation from external networks.

◆ Hybrid

Keep routine or sensitive work local while routing approved workloads to stronger private or cloud models when policy permits.

◆ Cloud

Cloud inference can also be deployed when scale, convenience, or an organization’s existing cloud architecture makes it the right fit.

Engagement

From Assessment to Working System

A deployment can include hardware and model evaluation, inference serving, network architecture, CodeLoop configuration, benchmark validation, offline testing, and documentation for the environment.

01

Assess

Workload, users, hardware, privacy constraints, model requirements, and expected performance.

02

Deploy

Configure the selected local, private-network, on-premises, isolated, hybrid, or cloud architecture.

03

Validate

Benchmark the system, test failure/recovery paths, and document exactly what was deployed.

Private AI

Discuss an AI Deployment

Tell us what needs to stay private, what hardware you already have, and what you want the system to accomplish.

Contact MCMLV1