The "run a model locally" space has split into layers, and comparing a server with a desktop app causes most of the confusion. Sorting the projects tracked here by what they are:
Servers. Ollama (MIT): a daemon plus CLI, pulls models by name, exposes its own API and an OpenAI-compatible /v1. LocalAI (MIT): a drop-in OpenAI API replacement that also covers images, audio and embeddings, heavier to set up and broader.
Web front-ends. Open WebUI: a full chat UI with users, RAG over uploaded documents, tool calling, connecting to Ollama or any OpenAI-style endpoint. Runs as a container; it has had its own licence with a branding clause since 2025. LibreChat and NextChat are the lighter alternatives: /compare/librechat-vs-open-webui, /compare/nextchat-vs-open-webui.
Desktop apps. Jan (AGPL-3.0): a desktop app that bundles a llama.cpp engine and a model library, and can also expose a local OpenAI-compatible server. GPT4All (MIT): similar idea, longer history, simpler. Both are "install and chat" with no terminal.
Closed but popular. LM Studio is the polished desktop app many people start with. It is not open source, so it is not tracked here, but it is the fair comparison for Jan and GPT4All.
A reasonable path: Jan or GPT4All to find out whether a local model is useful to you at all; Ollama plus Open WebUI when you want the model shared on a network or used from other tools. /compare/jan-vs-ollama, /compare/gpt4all-vs-jan, /compare/localai-vs-ollama and the /collections/local-first-ai list cover the pairs.