Model hosting
Your model.
An API on the cloud
you pick.
We serve the weights behind HTTPS. The server sits in that provider’s region. Kenya is a separate choice, and a cheap GPU abroad is still a cheap GPU abroad.
Model hosting is for a file you already have: an open-weight model, an adapter from fine-tuning, or a task model from a build we did with you. We containerise it, put TLS in front, and hand you an endpoint.
The region follows the cloud. You name the host and the region, or we pick a cheap SKU and show you the region before anything is switched on. A typical public GPU region sits outside Kenya. Nairobi colo and your own rack are the paths when the data must stay here. We write that split on the quote. We never describe a foreign region as a Kenya deployment.
What you get
A URL for the model, with the region written down.
Weights in
Your file, an open-weight checkpoint, or a model we trained. A closed vendor API stays on that vendor.
Host you choose
Your cloud account, or one we operate with the bill passed through. CPU or GPU, disk, backups, keys you can rotate.
Batch or always-on
A queue can start the GPU and stop it when the window ends. A live API stays up, and the host bills those hours.
Same box, new address
The container can move to another cloud, or onto Kenya hardware and a Nairobi GPU API, if counsel later requires it.
$ corvux host up --model your-weights --cloud client-choice REGION written on the quote · Kenya only when that host has it ENDPOINT https://model.your-domain/v1 KEYS rotated · prompts logged in the tenant you approve IDLE batch GPU stops when the window ends ✓ relocate path: same image → Nairobi colo
When this is the wrong cloud
Customer files pick the address.
Bank customer files, employee records, and patient images are a KDPA Section 48 transfer question on a foreign region, and sometimes a Section 50 Kenya-server question. For those, we quote on-premise or Nairobi colo, and we leave the cheap foreign GPU off the proposal. Model hosting on a public cloud is for workloads you and your counsel accept outside Kenya: demos, non-sensitive ops, and models whose prompts can travel.
The wider deploy, apps included, is cloud AI. The residency split is on on-prem vs cloud. Building the model first is custom models.
FAQ
Region, account, and what we will serve.
Will the server be in Kenya?
Kenya is the location when you choose Nairobi colo, your own rack, or a cloud region inside Kenya. A typical public GPU region sits abroad. The quote names the region before go-live.
Whose cloud account is it?
Yours if you have one. A Corvux-operated account if you want us to hold the keys, with the host bill passed through.
Which models can you host?
Open-weight language models, LoRA adapters, and smaller task models. A closed vendor API stays on that vendor.
What happens when the GPU is idle?
Batch jobs start the GPU for the window and stop it when the queue is empty. An always-on API stays up, and the host bills those hours.
Send the model and the region you can accept
Weights or a link, plus whether prompts can leave Kenya. We will name a host, or tell you to stay on Nairobi hardware.
Request model hosting