How it works

Your data stays on our hardware.

Umbreonics will answer your team's questions and do its routine work on GPUs we own in Florence, South Carolina, from January 2027. Here is what goes in, where it is processed, and what comes back.

The data flow

  1. In

    From your business

    • Files, email and CRM records
    • Staff questions and requests
    • Routine tasks to automate
  2. Umbreonics private AI stack

    Where the work happens

    • A search index over your documents
    • Open-weight models, fine-tuned where useful
    • A workflow engine for repeat tasks
  3. Out

    Back to your team

    • Answers that cite their sources
    • Drafts of emails, quotes and procedures
    • Tasks done on a schedule

Two promises

What we will never do with your data.

  • No outside AI provider sees it.

    Every request is handled on our own GPUs in Florence. Nothing is passed to a public AI service.

  • It never trains anyone else's model.

    A model fine-tuned on your data is used for your business only.

The hardware

Four Blackwell GPUs, owned, not rented.

Owning the hardware is what lets us charge one flat price per business and keep every customer's data on machines we control.

GPUs
4 × NVIDIA RTX PRO 6000 Blackwell
GPU memory
384 GB in total
Where
Florence, South Carolina
Status
Arriving for the January 2027 launch

The software

Open models, kept up to date.

Models
Open-weight language models, such as gpt-oss-120b, which fits on a single one of the four GPUs.
Serving
vLLM and TensorRT-LLM, open-source software that runs the models on the GPUs.
Your documents
A search index over each customer's own documents, so answers can cite the file they came from.
Fine-tuning
Fine-tuning on a customer's own data, where it makes the work better.
Upgrades
When better open-weight models are released, customers move to them without changing tools.

Tell us what your team spends time on.

A short conversation is enough to know whether Umbreonics fits your business. Email us with the kind of work you do, roughly how many people, and what you would like AI to help with.