Running a language model on your own machine used to mean compiling code and reading GitHub issues. In 2026 it is a normal app install. The reasons people do it are practical: nothing you type leaves the computer, there is no per-token bill, it works offline, and you can try open models the week they are released.
This guide compares the four tools most people start with — Ollama, LM Studio, GPT4All and AnythingLLM. All four are free to use. Details below are current as of October 2026; check each project's site before you rely on a specific feature.
The short answer
- Ollama if you are a developer, or want other apps to use a local model through an API.
- LM Studio if you want the most polished point-and-click app for finding and chatting with models.
- GPT4All if you want the simplest offline chat on an ordinary laptop, including chatting with a folder of your documents.
- AnythingLLM if your goal is asking questions about your own files, with workspaces and agents, using either a local or a cloud model.
They are not rivals so much as layers. A common setup is Ollama or LM Studio running the model, with AnythingLLM or another chat app on top.
Ollama: the engine other apps plug into
Ollama is a command-line tool (with a small desktop app) that downloads and runs open models such as Llama, Gemma, Mistral, Qwen and DeepSeek. One command pulls a model, another runs it, and a local REST API lets editors, chat clients and automation tools such as n8n talk to it.
- Licence: MIT, free for personal and commercial use.
- Best at: being a reliable background service for other software.
- Worth knowing: Ollama Inc. also sells optional cloud plans for running larger models on its servers. You do not need them to run models locally.
LM Studio: the friendliest desktop app
LM Studio combines a model browser, a chat window and a local server in one install for Windows, macOS and Linux. It runs GGUF models through llama.cpp and, on Apple Silicon Macs, MLX models, which are usually faster on M-series chips. Its local server speaks the same API format as OpenAI's, so many tools can switch to it by changing one URL.
- Licence: the app is closed source but free, including for work use since mid-2025. Its command-line tool and SDK are open source.
- Best at: browsing models, comparing them, and seeing how much memory each one needs before you download it.
GPT4All: offline chat with your documents
GPT4All, from Nomic, is aimed at ordinary laptops. It runs quantised models on the CPU or GPU you already have and needs no API key. Its LocalDocs feature indexes a folder of your files so answers can draw on them, with nothing leaving the machine.
- Licence: MIT, open source, with a Python package for developers.
- Best at: a private assistant for notes, manuals or study material on a modest computer.
AnythingLLM: a workspace for your files
AnythingLLM, from Mintplex Labs, is built around workspaces. You add PDFs, web pages and documents, then chat with answers grounded in them. It can use a local model (often through Ollama or LM Studio) or a commercial API, and it includes simple agents.
- Licence: the desktop app is free and MIT-licensed; a hosted cloud version is paid.
- Best at: small teams or individuals who want a private document assistant without building one.
What hardware do you need?
Model size matters more than the app. Rough guidance for 4-bit quantised models:
- 7–8 billion parameters: about 5–6 GB of free RAM or VRAM. Usable on most laptops with 16 GB of RAM.
- 12–14 billion: about 9–10 GB. Comfortable with a 12 GB graphics card or a Mac with 24 GB or more of unified memory.
- 30 billion and up: 20 GB or more. Workstations, high-memory Macs, or splitting the model between GPU and system RAM (slower).
A graphics card or Apple Silicon speeds things up a lot, but small models do run on the CPU alone, just more slowly. Start small, check that the answers are good enough for your task, and only then move to a bigger model.
A sensible way to start
- Install LM Studio or Ollama and download one small, recent model.
- Ask it the kind of questions you actually need answered, not benchmark puzzles.
- If you want it to read your files, add AnythingLLM or use GPT4All's LocalDocs.
- If you want other software to use it, turn on the local server and point the app at it.
Local models are not as capable as the largest cloud models on hard reasoning, and very new releases can need an app update before they run. For private notes, drafting, summarising and coding help, they are already good enough for a lot of everyday work.
Related: Ollama · LM Studio · GPT4All · AnythingLLM · Chatbox