Run a model on your own computer

Open models (Llama, Qwen, Mistral, DeepSeek) can be downloaded and run on your own computer or server. That’s especially useful when:

• the data is confidential — banks, healthcare, government;
• the internet is slow or expensive — the model works offline.

What computer you need

Model size	Memory needed (RAM or GPU)	Quality
1–4B	4–8 GB	simple tasks, fast
7–9B	8–16 GB	good for daily work
30B+	24 GB+	strong, but needs costly hardware

Apple M-series MacBooks and recent NVIDIA graphics cards work best.

Option 1: LM Studio — no coding

1. Download the app from lmstudio.ai.
2. Search for “qwen” or “llama” and download a model that fits your computer.
3. Chat in the window — it works even with the internet off.

Option 2: Ollama — for developers

Install Ollama and run in a terminal:

ollama run qwen3

The model downloads and a chat starts. Ollama also serves an API on localhost:11434 that your own app can call:

curl http://localhost:11434/api/generate -d '{
  "model": "qwen3",
  "prompt": "What is artificial intelligence? One sentence."
}'

About Uzbek

Small open models write Uzbek less well than large cloud models; Russian results are usually better. Always check important text and compare models.

Next step

To connect your own documents to a local model, read about RAG (in the glossary) and try Open WebUI — a ChatGPT-style interface that runs entirely on your server.