Consulting
Private AI Workstation
A workstation that runs a 27-billion-parameter AI with vision, offline, on your own documents - for about the price of a good laptop in hardware. You own the machine, the model and the data.
Who this is for. Firms of 10 to 100 people whose staff already paste contracts, customer records or financials into public AI tools, and whose owner has finally asked where that data goes. Also any business where a per-seat cloud subscription or a data-residency clause rules the public tools out.
What you get. A parts list and build, either two mid-range GPUs (the reference build runs Qwen 3.8 27B at three-bit with vision on two RTX 4070-class cards, about 24 GB combined, roughly USD 1,100 in cards, at 50 to 60 tokens a second) or a 128 GB Ryzen AI Max+ mini PC for bigger models. A tuned model served through an OpenAI-compatible API so your existing tools can point at it. Chat with your own files, with source citations in every answer. Named users, no internet exposure by default, backups with a documented restore. A runbook, a 60-minute hand-over, and 30 days of fixes.
Our honest position. A local 27B model is excellent at summarising, drafting, extraction and question-answering over your own documents. It is not frontier-model quality on hard reasoning, and we say so on the first call rather than after the build. It is an appliance, but someone on your side has to own it.
How an engagement works. A 20-minute call on what you want it to do, how many users and what data. A fixed-fee proposal within two working days with the hardware itemised at cost; we do not resell hardware or software. Build and test in about ten working days from parts arriving, hand-over on site in Doha or remote.
Other ways we help