Eigna Box · Enterprise
Eigna Box Enterprise
Racks on your own servers. One Figure in every unit.
The plate reads
Keller’s AI
An example owner
- Runs
- Eigna Local
- Size
- 482.6 × 634 × 88.9 mm
- Status
- Planned for Q1 2027
- In a standard rack, 482.6 mm wide, as designed
- 2U
- The Figure, the unit’s only light
- 26 mm
- Hot-swap power supplies at the rear
- 2
- Planned. Dates after 2026 are targets.
- Q1 2027
Who it is for
For a company’s own data centre: one unit or a full rack, delivered, set up and run by us. A dedicated GPU in your region is the other way.
- Owner
- Keller’s AI, an example
- Runs
- Eigna Local, on your own servers
What runs on it
- Every expert, on siteThe company’s experts and our everyday model, next to the data they learn from.
- The release recordEvery version, its test result and who approved it, read in the Console.
- A status lightReleased, learning, held or rolled back, each with its own light. Only the Enterprise has one.
Keller’s AI, unit after unit, rack after rack. The Figure is each unit’s only light.
Materials and the plate
Graphite, silver, and light through two plates.
A silver faceplate with machined handles and captive thumbscrews, sleds either side, and a graphite field in the middle with the Figure as its only light. Light shows only where the cuts cross; nothing else on the device lights up.
Specs
Design facts, and what waits for the first build.
Design dimensions from the series geometry, width × depth × height as it stands. Compute, power and throughput are set with the first build and measured before we claim them.
- Width, depth, height
- 482.6 × 634 × 88.9 mm, 2U; chassis 440 × 600 mm
- The Figure
- 26 mm in the centre module, the unit’s only light
- Materials
- Silver faceplate and chassis, graphite field, diamond cut pocket
- Rear
- A vent field, four network cages, two management ports, two hot-swap power supplies
- Mounts
- Machined handles, captive thumbscrews
- Status light
- Released, learning, held, rolled back
- Run mode
- Eigna Local, on your own servers
- Compute and memory
- Standard parts: Apple silicon or an NVIDIA desk machine, with memory tiers from 36 to 256 GB, sized to the model. The tier for this device is set with the first build.
- Power and throughput
- Set with the first build, and measured before any capacity claim.
- Packed size
- 535 × 752 × 150 mm
In the box
The unit, the Key and the Model Deed.
Plug it in and the Figure lights: the owner’s model is home. The Key signs a release and rolls one back.
The KeySigns a release, and rolls one back.
The Model DeedTwo leaves: the base model and the owner’s work.
Status
Planned for Q1 2027.
The first box is built by us, from standard parts in our own case. Dates after 2026 are targets.
Start
Find out which box fits your work.
Thirty minutes with a founder: your task, where the data must stay, and which box fits.