Bring your own GPU (BYO): how your node and inference are secured
Last updated:
On this page
With bring your own GPU (BYO), you run inference on your own hardware, in your own environment. Your prompts and responses are processed on your side, not on shared infrastructure. staik's role is to securely route your organization's requests to your node and give you the same OpenAI-compatible API as the rest of the platform.
The protection is built in layers. No single layer carries the whole load, and if one layer were to fail, the others still stand.
The core promise
Your data stays on your hardware. staik is a router, not a data store. We keep usage counters for billing, never the content of your requests.
The layers of protection
1. Data stays with you (data residency)
Inference runs on your GPU. Prompts and responses never leave your machine to be processed elsewhere. What reaches us is billing metrics, not content.
2. Encrypted transport
All traffic between staik and your node runs over an encrypted private network (WireGuard-class encryption), and requests from your client to the API are protected by TLS. Inference never travels in plaintext across the network.
3. Isolated, private network
Your node is never exposed to the open internet. It is reachable only through a single, trusted path from staik's gateway, and the node's own firewall rejects everything else. Even if something in the outer layer were to give way, there is no open door into the node.
4. Only your organization (tenant isolation)
Your node serves exclusively your own organization's requests. It cannot end up in the shared pool and never receives another customer's traffic. This is the hardest guarantee in the model, and it is built into how requests are routed, not something that depends on after-the-fact configuration.
5. Mutual authentication
The node and the gateway verify each other. Unauthenticated attempts to reach the node are rejected, and staik never exposes internal secrets to your node. Traffic in both directions is tied to an identity, not just to an address.
6. Minimal surface
The node runs only what is needed to serve your model, and operations are locked down. The less that is exposed, the less there is to attack.
What this means for you
Sensitive data never leaves your control. You retain physical and operational control of the hardware, and staik gives you a simple, encrypted and isolated way to make your own GPU part of the platform, without anyone, including us, getting a usable view of your inference.
Want to know more?
Is your organization considering its own GPUs for sensitive workloads? Get in touch and we'll walk through the setup and the security model in detail for your specific case.