Apps that run models on your own computer: supported systems, memory needs, GPU support and a download from our catalogue.
Updated
Runs open-weight models on your own computer from the command line or a small desktop app, and exposes them through a local, OpenAI-compatible API. Open source under MIT.
Desktop app for finding, downloading and chatting with local models, with a built-in API server for developers. Free to use, though not open source.
Open-source chat app in the ChatGPT style that runs models offline and can switch to cloud APIs in the same window. It also starts a local OpenAI-compatible server.
Nomic's open-source desktop app for local models on CPU or GPU, known for LocalDocs, which answers questions about your own files. New releases have been rare since February 2025.
All-in-one open-source workspace, on the desktop or in Docker, where documents, agents and local or cloud models come together. Suited to private work with your own files.
Self-hosted web interface for Ollama and OpenAI-compatible APIs, installed with Docker or pip. Several people can share one server through their browsers.
Open-source C/C++ engine that runs GGUF models on CPU and GPU, with command-line tools and an OpenAI-compatible server. Many local AI apps are built on top of it.
Single-file program based on llama.cpp, with a web interface for chat and story writing and KoboldAI- and OpenAI-compatible APIs, favoured for fiction and role-play.
A local LLM app downloads an open-weight model and runs it on your PC or laptop. Your prompts stay on the machine, there is no per-token bill, and it works offline once the model is downloaded. This page lists such apps: desktop programs (Ollama, LM Studio, Jan, GPT4All, AnythingLLM), a web interface for a home server (Open WebUI) and engines for people who want full control (llama.cpp, KoboldCpp).
The app is free; the model is a separate download with its own size and licence. A model has to fit into RAM, or into the GPU's memory if you want the graphics card to do the work, so a bigger model needs a bigger machine. Without a suitable GPU, models run on the processor, which works but is slower. The cards list what the vendors state; we do not promise any particular speed or answer quality, because both depend on the model and the hardware.
The strongest models are not published as open weights and run only in the cloud. If you need them, look at the API providers or the AI apps; several local apps, Jan and AnythingLLM among them, can also connect to a cloud API alongside local models.