People who want private experiments with small open models.
GLWS · Use cases
AI on your own server.
Private models, honest limits.
Our plans have no GPU, so models run on the CPU. That is slow but private, and it suits small models, personal use, agents and gateways that call hosted models.
Who it is for.
Typical reasons people choose it.
Developers running agents, MCP servers or an API gateway that call hosted models.
Anyone who wants a private chat interface or document search on their own data.
Choose a size.
What fits where. These are guides, so measure your own workload.
2 GB
Agents, gateways and tools that call hosted models.
4 GB
A 3-billion-parameter model quantised, with care.
8 GB
A 7 to 8-billion-parameter model, or several services beside a small model.
16 GB
A larger model or a fuller stack. Still CPU speed, so measure first.
GLWS plans that fit.
The lowest-priced GLWS VPS plans with enough memory.
Get started with GLWS.
Three steps.
- Choose the size from the model you want, using the sizing guide, and measure real speed on your plan.
- Keep every AI service on localhost and reach it through a tunnel or an authenticated proxy.
- Use hosted APIs when you need quality or speed that a CPU cannot give, and keep keys out of code.
Guides for this setup.
Step-by-step instructions from the knowledge base.
Choose a VPS size for a local AI model
Work out how much memory, disk and CPU a quantised model needs, then test on a small plan and measure before you upgrade.
Read guide →Run a local language model with Ollama on a CPU-only VPS
Install Ollama, pull a small quantised model, measure its real speed on your plan and keep the API private to the server.
Read guide →Self-host Open WebUI with Ollama using Docker Compose
Run a private ChatGPT-style interface and Ollama in one Compose project, publish it on localhost only and lock down sign-ups.
Read guide →Secure Ollama, Open WebUI and other AI services on a public server
Keep AI services off the open internet: bind to localhost, avoid the Docker firewall bypass, add authentication and verify from outside.
Read guide →Create an AI API gateway with LiteLLM
Run the LiteLLM proxy with Docker Compose, keep your provider keys in one place, issue limited virtual keys and put the gateway behind HTTPS.
Read guide →Run OpenClaw on your VPS
Install OpenClaw as an unprivileged user, run its gateway as a service, keep the control interface private and limit who and what the assistant can reach.
Read guide →FAQ
Questions.
Quick answers.
Can I run a large language model on a VPS?
Small ones, on the CPU, yes. They are slower than hosted services, and our plans have no GPU. The sizing guide explains memory and speed.
Which plan for Ollama?
Around 4 GB for a small model, 8 GB for a 7B-class model. Test speed on a small plan before you move up.
Is it private?
Local models keep prompts on your server. If you call hosted models, your prompts go to that provider.