- Install Ollama on Windows, macOS, or Linux.
- Choose a model based on memory, speed, and quality.
- Test the local API so you can use it later with your apps.
Realistic minimum requirements
- 8 GB of RAM: small 1B to 4B models for testing, summaries, and light chat.
- 16 GB of RAM: 7B to 8B models, the best balance for learning.
- 32 GB or more: 14B models and smoother workflows with long documents.
- GPU: helps a lot, but is not required to get started.
Installation
Go to the official Ollama website, install the version for your system, and open a new terminal. On Linux you can also use the terminal installer:
curl -fsSL https://ollama.com/install.sh | sh
Verify that the command exists:
ollama --version
Your first model
To start, use a small, fast model. If your hardware handles it well, you can move up in size later.
ollama run qwen3:4b
Recommended models for learning
- qwen3:4b: a good first choice for modest hardware.
- llama3.1:8b: the classic balance if you have 16 GB of RAM or more.
- mistral: fast and practical for general testing.
- codellama: useful for code examples, though it does not replace Claude Code.
Test the local API
Ollama listens at http://localhost:11434. Your applications will talk to that address.
curl http://localhost:11434/api/generate -d '{
"model": "qwen3:4b",
"prompt": "Summarize in one sentence what local AI is.",
"stream": false
}'Commands you'll use every week
ollama list ollama run qwen3:4b ollama pull llama3.1:8b ollama rm model:tag ollama ps