Simple
Two cables and a login. No Kubernetes, no DevOps team, no cloud onboarding project.
On-prem AI hardware
Your company's AI. In one box.
Kaldryn One is a plug-in AI appliance the size of an office locker. Give it power, ethernet and your documents, and your whole team runs private AI from your own floor, with nothing leaving the building.
kaldryn-one · first boot
$ kaldryn-one boot
From box to production, in real time.
Performance
Small footprint. Serious metal.
2×
data-centre GPUs
200+
curated models
180
tok/s peak streaming
0
bytes to the internet
Compute2× data-centre GPUs · 96 GB VRAM total
Memory128 GB ECC
Storage8 TB NVMe · RAID 1
Networking2× 2.5 GbE
Acoustics< 38 dB under full load
PowerStandard wall outlet · 0.9 kW typical
Form factor62 × 28 × 30 cm · 19 kg
Models200+ curated · offline USB updates
APIOpenAI-compatible endpoint
Time to production< 10 minutes
Model tiers on the box
Model tier
First token
Streaming
Concurrent users
Swift8B class
First token0.3 s
Streaming180 tok/s
Concurrent users60+
Balanced32B class
First token0.6 s
Streaming95 tok/s
Concurrent users25
Deep70B class
First token0.9 s
Streaming52 tok/s
Concurrent users12
Typical figures for the 2× GPU configuration on an internal-documents workload; throughput varies with model choice and document mix.
No round-trips
The fastest network request is the one you never make.
On your floor
< 0.4 s
to first token, straight over your LAN
Via a cloud region
+ every round-trip
internet latency, on every single prompt
Setup
Production in four steps.
01
It arrives pre-configured
Models, license and security are baked in before we ship. Unbox it and stand it next to the printer.
02
Plug in power and ethernet
A standard wall outlet and your LAN, no server room, no rack, no cooling project.
03
It indexes your documents
Files land on the box, get indexed on the box, and stay on the box.
04
Your team serves AI from your floor
Chat, API and automations for everyone, unlimited usage at a flat cost.
Pre-installed platform
The whole Kaldryn platform, ready on first boot.
- A polished chat portal for every employee
- An admin console for models, users and usage
- An OpenAI-compatible API for your own tools
Three surfaces, one hardware-bound license, upgrades light up new capabilities instantly.
Kaldryn One
kld-prod-01 · on-prem
Models3 running
GLM-5.2
355B · general
Llama-4 70B
70B · fast
Qwen3-32B
32B · coding
DeepSeek-V3
671B · reasoning
Tokens this month
48.2M
▲ 12% vs last month
Token & credit fees
€0.00
only electricity
Daily throughput7d
This month
- Tokens processed48.2M
- Models available200+
- Token & credit fees€0.00
- Cloud-AI equivalent~€4,820
You pay for power, not per token.
Why Kaldryn One
Simple. Sovereign. Flat.
Sovereign
Your prompts, documents and answers never leave the building, enforced at the host firewall.
Flat
No tokens, no per-seat fees. Electricity is the only meter that runs.
Built for
Teams that can't ship their data to someone else's cloud.
Private by construction
Unplug the internet. Keep working.
Kaldryn One is built to run with the cable out, no third-party API, no phone-home license, no cloud dependency anywhere in the path.
0 bytes outbound
Data-isolation mode seals outbound networking, enforced at the host firewall.
DLP on every message
12 built-in rules catch PII and secrets before they reach a prompt or an answer.
Immutable audit
Every question and answer is logged, tamper-evident and SIEM-ready.
Respects your permissions
The box only answers from documents each user is allowed to open, SSO and roles included.
Cloud AI was renting.This is owning.
One box, on your floor, that your whole company runs on, and that nobody can meter, throttle or read.
On your hardware. Under your signature.
Bring the AI in-house.
Talk to our team about a Kaldryn deployment, or a pre-built appliance shipped ready to seal.
Talk to Sales
or email info@kaldryn.com