Local AI Workstation
There is nothing artificial about our intelligence.
Physical, local AI for your Knowledge. A workstation on your desk that reads your sources, cites them and answers to you.
You are on the list. We will write when your batch opens.
First batch built to order in Asturias, Spain.
How it works
Open it up.Follow a question.
The Local AI Workstation
Click a part · Space opens the caseTap the case to open it · tap a part
A question, end to end
The model reads. Your Knowledge speaks.
Ask the Space on this desk. Every answer is read from its own sources, on the workstation beside it.
Own the engine
Empathy.AI supplies the workstation. You own it.
Govern by design
Every artefact cited to the source, or marked “not found”.
Authored Knowledge
Configured to understand curated inputs: your documents, policies and teaching material.
Sustainable AI
One machine, sized for a desk and the work done at it.
Why physical, why local, why now.
Open-weight models run locally.
A single workstation now holds a capable reader: 27B–32B-class models on 32 GB of video memory.
The model reads. Your Knowledge speaks.
Configured to understand curated inputs and tasks; answers come from your sources, cited.
Owners need governance.
Schools, publishers and creators need answers they can audit, trace and defend.
Specifications
First batch.
Built to order and set up for your Space before it ships.
- Graphics
- NVIDIA RTX PRO 4500 Blackwell, 32 GB video memory
- Models
- Open-weight, 27B–32B class, run locally
- Board
- Mini-ITX with processor, memory and NVMe storage
- Accounts
- Up to 20 per workstation
- Cooling
- 2 × 120 mm fans; low-profile CPU cooler with six heat pipes
- Power
- 850 W power supply
- Case
- 3D-printed shell on a raised stand; front and side dust filters
- Size
- 25.5 × 25.5 × 33 cm1
- Network
- Ethernet, on your own network
- Updates
- Run remotely by Empathy.AI
- Made in
- Asturias, Spain
- Price
- Shared with the waitlist first
Tokens
4,300 M a month, included2
No per-token billing and no usage caps. The ceiling is the card itself.
The same month, billed per token by a provider
1 Width × depth × height.
2 Estimate for a 27B open-weight model running around the clock on the RTX PRO 4500. Provider costs use standard list prices in US dollars on 13 Oct 2026.
Questions
Before you join.
Who is it for?
Schools, publishers, creators and teams who want to work privately with their own Knowledge. One workstation serves up to 20 accounts.
Where does my Knowledge live?
On the workstation, on your network. Your documents, your Spaces and the answers built from them stay on the machine you own.
Who looks after it?
Empathy.AI supplies the workstation and runs updates remotely. You own the hardware and decide who has an account.
When does it ship, and what does it cost?
The first batch is built to order in Asturias, Spain. Dates and price go to the waitlist first.
Waitlist
Relate with Knowledge.Respect the Sources.From your Desk.
You are on the list. We will write when your batch opens.
One email when your batch opens. Nothing else.