X-OS. Switch the token meter off.
AI agents, an fine tuned model fine tuned to your business, and the hardware appliance to run it.
One box on your premises. Starting from USD 1,799 per month*. No token meter.
The problem
There’s a token meter running under enterprise AI.
Every document it reads. Every question it answers. Every agent you switch on. It all counts, and the bill arrives after the fact.
Token meter, monthly invoice
Up 41% on last month
Output tokens, billed at the higher rate
2,860,000,000
Input tokens
1,240,000,000
Embeddings and retrieval
388,000,000
Agent runs (new team onboarded)
1,341,000,000
Amount due
Metered
ILLUSTRATIVE. THE PATTERN IS THE POINT, NOT THE NUMBERS.
Your data leaves the company, and the country, to think.
Every prompt carries pricing logic, contracts and customer records to shared infrastructure outside your company, often outside the UAE. For regulated and sovereign environments, that is no longer acceptable.
Paying more for AI the harder it works. That’s backwards.
AI should behave like every other piece of infrastructure you own: a known cost, doing more work every month.
AI should behave like every other piece of infrastructure you own: a known cost, doing more work every month.
Introducing X-OS
Three things you used to buy separately. One box.
Agents, model and hardware appliance, engineered as one system and priced as one line. Installed in your building, run by Amantra.
1
Your agents
Choose 5 to 10 from a marketplace of 1,000+ prebuilt, pretested, governed agents. Add more as priorities change.
2
Your model
An open source LLM, built for your business context and fine tuned on your documents, rules and language. Swap it when a better one arrives.
3
Hardware appliance
Up to one petaflop of AI performance, 128 GB unified memory and self encrypting storage in a 150 mm box. Installed and supported by Amantra.
Hardware appliance
Data centre class AI. On a desk. Inside your building.
Up to one petaflop of AI performance in a 150 mm box that weighs 1.2 kg and runs from a 240 W external power supply. No rack, no server room. Installed and supported by Amantra.
Sized to your volumes at scoping. Installed, updated and supported by Amantra, inside your building.
1
PFLOP
Runs large models locally
Up to one petaflop of AI performance at FP4 precision with sparsity, on a GPU with fifth generation Tensor Cores.
128
GB
Model and agents share one memory pool
128 GB of coherent unified system memory at 273 GB/s, shared by CPU and GPU.
4
TB
Your data, encrypted at rest
Up to 4 TB of NVMe storage with self encryption, staying on the appliance inside your walls.
200
Gbps
Enterprise network, enterprise speed
ConnectX-7 smart NIC with dual QSFP ports, plus 10 GbE, Wi-Fi 7 and Bluetooth 5.4.
20
cores
Orchestration alongside inference
A 20 core Arm CPU handles orchestration and workflow logic while the GPU runs the model.
240
W
Plugs in like a laptop, not a server
External 240 W power supply, 150mm x 150mm x 50.5mm, 1.2 kg.
Technical specification
CPU
20 core Arm: 10 Cortex-X925 and 10 Cortex-A725
GPU
5th Gen Tensor Cores and 4th Gen RT Cores
AI performance
Up to 1 PFLOP, FP4 precision with sparsity
System memory
128 GB LPDDR5x coherent unified system memory, 256 bit, 273 GB/s bandwidth
Storage
Up to 4 TB NVMe M.2 SSD with self encryption
Networking
ConnectX-7 Smart NIC up to 200 Gbps, dual QSFP ports. Wi-Fi 7. Bluetooth 5.4. 10 GbE RJ-45
Ports
4x USB Type-C, 1x HDMI 2.1a, HDMI multichannel audio
Power supply
240 W external power supply, 140 W chip TDP
Dimensions
150 mm L x 150 mm W x 50.5 mm H, 1.2 kg (2.6 lbs)
See it running at a preview
Book a preview
Agent marketplace
Pick 5 to 10 agents. Start with the work that costs you most.
A marketplace of 1,000+ prebuilt agents across every enterprise function. Every agent sits on top of your existing systems and executes inside them. Search, filter and shortlist below, then bring the list to your preview.
Models
Open source. Fine tuned to your business. Never locked in.
X-OS runs open source language models, built for your business context and fine tuned on your premises. Your agents don’t change when the model does.
Model families available at launch
Llama
Mistral
Qwen
DeepSeek
Sovereignty
Nothing leaves the company. Nothing leaves the country.
Agents, model and hardware appliance live on your premises. Confidential data never leaves your company, and never leaves the UAE. Sovereignty by architecture, not by policy.
On your premises
Installed in your building, inside your network.
In country
Processed and stored in the UAE. No external calls required to run agents.
Governed autonomy
Every agent action is logged, auditable and subject to human approval where you set the threshold.
Pricing
One line item. Known on day one.
No tokens. No per seat fees. No integration project before the first agent goes live.
TOKEN METERED CLOUD AI
Metered
per token, per call, per seat
Rises with every user and use case
Bill rises when a new team is onboarded
36 month total: unknown
Data processed outside the company, often outside the UAE
Locked to one model provider
X-OS
From USD 1,799
per month*
*On a 3 year contract. Final pricing confirmed at scoping.
Agents, model and hardware appliance, one monthly amount
36 month total from USD 64,764, known on day one
Zero token spend. Same amount when a new team joins
No large upfront commitment
Book a private preview
How it works
1
Choose
Pick 5 to 10 agents from the marketplace, matched to the work that costs you most today.
2
Scope
We size the appliance to your volumes and confirm the systems the agents will work inside.
3
Train
The model learns your business context inside your walls. Your data never travels to do it.
4
Run
The appliance goes in. The agents go live. The bill stays flat, month after month.
Private preview
See your use cases running on X-OS, with no meter underneath.
Twenty private previews before general availability. Bring your priority use cases. We bring the appliance.
Format
60 minutes, on site or online
Commercial model
From USD 1,799 per month*
Availability
20 slots