How to Install llama.cpp on Windows

Setting up llama.cpp on Windows is straightforward. You can either grab a ready-to-go version or build it yourself for better speed if you have a powerful graphics card.

Option 1: The Fast Way (Pre-built)

This is perfect if you want to skip the technical setup and just start using AI.

1. Head over to the llama.cpp releases page.
2. Download the win-cuda-x64.zip if you have an NVIDIA card, or win-avx2-x64.zip for standard PC use.
3. Unzip the folder, and you’re ready to run llama-cli.exe from your terminal.

Option 2: The Pro Way (Build from Source)

If you want the best performance, building it specifically for your hardware is the way to go. You’ll need Visual Studio 2022 (with C++ Desktop development checked), CMake, and Git installed.

# 1. Open your terminal and clone the project
git clone https://github.com
cd llama.cpp

# 2. Set up the build folder
cmake -B build

# 3. Compile the software
# For standard use:
cmake --build build --config Release

# For NVIDIA users (much faster):
cmake -B build -DGGML_CUDA=ON
cmake --build build --config Release

How to start chatting

Once you have a model file (in .GGUF format), you can start it up with a simple command:

./llama-cli -m your-model-name.gguf -p "Hi there!"

Tip: If this feels like too much typing, check out Ollama—it’s a one-click installer that does all of this for you behind the scenes.