Tools to Run LLMs Locally

A clear overview of software to run Large Language Models on your own hardware for maximum privacy, speed, and control.

Running Large Language Models (LLMs) locally offers significant benefits in terms of data privacy, offline availability, and the absence of subscription fees or API limits. Depending on your technical background and hardware configuration (such as GPU VRAM and memory), various specialized tools are available.

Below you will find a curated overview of the most widely used open-source and standalone tools on the market, including specifications per platform, ease of use, and the ideal target audience.

Overview of Local LLM Software

Ollama CLI / Service

A popular runtime based on llama.cpp that allows you to easily start and manage open-source models like Llama 3, Mistral, and Gemma via the command line.

Platforms: macOS, Linux, Windows
Ease of Use: High (after installation via CLI)
Target Audience: Developers & API integrations
LM Studio Desktop App

A user-friendly desktop application with a graphical interface to download GGUF models directly from Hugging Face, test them, and access them via a local OpenAI-compatible API.

Platforms: macOS, Windows, Linux
Ease of Use: Very high (visual)
Target Audience: End users & testers
llama.cpp C/C++ Engine

The foundational C/C++ engine developed by Georgi Gerganov for running LLMs with minimal overhead and maximum performance on a wide range of hardware, including CPU offloading.

Platforms: Cross-platform (C/C++)
Ease of Use: Low (requires compiling/CLI)
Target Audience: Advanced engineers & researchers
Jan Desktop / Open Source

An open-source alternative to ChatGPT that works completely offline on your own computer. Offers full privacy, local storage of conversations, and support for various model formats.

Platforms: Windows, macOS, Linux
Ease of Use: High (out-of-the-box)
Target Audience: Privacy-conscious users

Which tool should you choose?

If you want to get started right away without hassle with configurations or command lines, LM Studio or Jan is the best choice thanks to the visual interface. If you want to run models in the background as a service or integrate them into your own software projects via an API, then Ollama is preferred. For maximum control at a low-level hardware level, llama.cpp remains the gold standard.