Large language models can be run locally on a personal computer rather than through cloud-based services such as ChatGPT or Claude. This can provide offline access, stronger privacy and freedom from subscriptions or usage limits, although local models are generally less capable and convenient than leading paid services and require users to manage installations and updates themselves.

Local LLMs work on Windows, macOS and Linux, with Macs often favoured because Apple Silicon combines processing and memory resources effectively. At least 8 GB of RAM is needed, while 16 GB is preferable and 32 GB or more supports larger models; a dedicated GPU with over 8 GB of VRAM, particularly Nvidia hardware on Windows, can significantly improve performance. Users need both an application, such as LM Studio Bionic, Ollama or GPT4All, and a model, which can be found through the software or repositories including Hugging Face.