You can run an AI model on your own computer, without sending anything over the internet, with a program such as Ollama or LM Studio. You download the program, download an open model, and the conversation happens on your machine. It works well for experimenting, for confidential texts and for learning. It does not match the biggest cloud models, and the speed depends on your computer’s memory.
This is different from running Ollama on a server, which is covered in running models on your VPS. Here you need no server, no domain and no outside connection.
Ollama or LM Studio
|
Ollama |
LM Studio |
| How you use it |
From the command line, and through a local API for other programs |
A windowed application: search for models, download and chat |
| Who prefers it |
Those at ease with the terminal or wanting to connect other tools |
Those who want to experiment without commands |
| Where to get it |
The project’s official website |
The project’s official website |
| Connecting other programs |
By default it listens only on the computer itself (the port the documentation gives is 11434) |
Has an optional local server; see the documentation |
First steps, in short
| 1 |
Download the program only from the official site. There are fake copies with the same name.
|
|
| 2 |
Pick a small model to begin with. Each model states on its page the memory it needs; read it before downloading. The files are large: check your disk space.
|
|
| 3 |
In Ollama, the basic commands are ollama pull model-name to download, ollama run model-name to chat and ollama list to see what you have. The exact name comes from the project’s model list.
|
|
| 4 |
In LM Studio, search for the model inside the application, download it and open a chat.
|
|
| 5 |
Measure. See how long it takes to answer and whether the result is enough for the task. Only then move to a bigger model, if memory allows.
|
|
What to expect
| Aspect |
Reality |
| Speed |
Depends mostly on memory and on whether there is a graphics card. Without one, the bigger models get slow. |
| Quality |
Small models make more mistakes and follow complicated instructions less well. |
| Privacy |
The text does not leave the machine, as long as the program is not opened to the network. |
| Model licence |
Each model has its own. Some restrict commercial use; read it before using the model for business. |
| Errors |
They invent like the cloud ones do. See how to check an answer. |
|
Do not open the program to the network without knowing what you are doing. If a local service becomes reachable from other computers, anyone on the network can use your equipment. Keep it on localhost.
|
|
If the goal is to protect data, start by asking whether the task really needs a big model. Summarising, classifying and rewording short texts often goes well on small models.
|
|
Need more memory than your computer has? Have a look at the VPS plans.
See VPS plans
|
RECOMMENDED PRODUCT Web hosting with cPanel Domain and SSL included, daily backups and the panel you already know. from KSh858.00/mo (3-year plan, with coupon) See plans |
Angola
Brasil
Kenya
Moçambique
Nigeria
Portugal
South Africa
United Kingdom
United States
International
