Inferencing

Your models, served in Nepal.

Put models behind an endpoint with NVIDIA NIM microservices and KServe, on GPUs in the Kathmandu Valley. Prompts and answers are processed in Nepal.

Coming soonEnterprise early access is open

How it works

Three steps.

01
Pick a model
An NVIDIA NIM microservice for a popular open model, or your own.
02
Deploy an endpoint
With NVIDIA NIM or KServe.
03
Call it from your app
From inside Nepal, with low latency.

What you get

NVIDIA NIM microservices
Optimised containers for popular open models.
Your own models too
Serve fine-tuned or custom models with KServe.
Familiar APIs
NIM language-model endpoints follow the OpenAI API format, so existing client code carries over.
Right-sized
Smaller models can run on a MIG slice instead of a whole H200 NVL.
In-country by design
Requests are processed on GPUs inside the Kathmandu Valley, under Nepal's law.

Under the hood

Billing

Commitment: none, or a reserved endpoint. Rates are being finalised; every request gets a firm quote in rupees.

Questions

Which models can we serve?

Open models available as NVIDIA NIM microservices, and your own models through KServe. Tell us what you want to run and we confirm it fits.

Do prompts leave Nepal?

No. Endpoints run on our GPUs in the Kathmandu Valley, so prompts and answers are processed there.

Tell us what you want to run. It takes about a minute.