Ollama installs with one command and runs open AI models on your own computer. Aim for about 8 GB of VRAM, or unified memory on a Mac, and your first model is two commands away.
Ollama’s GitHub page says it helps you “get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.” It serves a local REST API at http://localhost:11434. This guide uses the official pages checked on 19 October 2026. We did not run it ourselves.
The short answer
- What it is: a command-line tool plus a local API for downloading and running open models.
- Install: one command on macOS, Linux and Windows (below), or a downloadable installer.
- Hardware: about 8 GB of VRAM or Mac unified memory. More memory allows a larger context window.
- Less memory: Ollama can use system RAM, “but responses may be slower.”
- Prefer clicking to typing: try LM Studio, a desktop app with a graphical interface.
How do I install Ollama?
On macOS or Linux, run this in a terminal: curl -fsSL https://ollama.com/install.sh | sh
On Windows, run this in PowerShell: irm https://ollama.com/install.ps1 | iex
Installers are also offered as downloads: Ollama.dmg for macOS and OllamaSetup.exe for Windows. Piping a script from the internet into a shell runs whatever it contains. Read it first, or use the downloadable installer if you prefer.
Windows requirements
The docs list Windows 10 22H2 or newer. NVIDIA GPUs need drivers 551.61 or newer. AMD cards need a ROCm v7 driver stack or a Vulkan-capable Radeon driver. The install does not require Administrator.
Disk space
The install needs at least 4 GB of disk space. Models can be “tens to hundreds of GB,” so check your drive before you pull several. Set the OLLAMA_MODELS environment variable to move model storage to a bigger disk.
How do I run my first model with Ollama?
The quickstart uses two commands: ollama pull gemma4:e2b, then ollama run gemma4:e2b. Type /bye to exit. The example model is about 7.2 GB to download, so start it on a good connection.
On Linux, ollama serve starts the server if it is not already running. Model names change quickly, so browse ollama.com for what is current.
What hardware do I need to run Ollama?
- “We recommend 8 GB of available VRAM, or unified memory on a Mac.”
- “Larger context windows need more memory.”
- “With less VRAM, Ollama can use system RAM, but responses may be slower.”
The GitHub README gives no RAM table. To learn what one model needs, test a small one first and watch memory use.
LM Studio or Ollama: which fits you?
| Ollama | LM Studio | |
|---|---|---|
| Style | Command line first, local API | Desktop app with chat and model management |
| Local server | REST API on localhost:11434 | “Serve local models on OpenAI-like endpoints” |
| Headless option | Runs as a server | “llmster is the headless version of LM Studio” |
| Platforms | macOS, Windows, Linux | macOS 14+ (Apple Silicon only), Windows x64 (AVX2) or ARM, Linux Ubuntu 20.04+ |
| Memory guidance | 8 GB VRAM or unified memory recommended | 16 GB+ RAM recommended on Mac and Windows; 4 GB+ dedicated VRAM recommended on Windows |
Choose Ollama if you work in a terminal or want an API for your own scripts. Choose LM Studio if you want a chat window and model manager. On an Intel Mac, note that LM Studio needs Apple Silicon. Each side lists its own requirements, since neither vendor publishes a side-by-side comparison. LM Studio’s homepage promotes a newer product called Bionic, but its documentation still describes the desktop app for all three systems.
Why run a model locally?
- Your prompts stay on your machine, which may matter for private material. Check each app’s own network and telemetry settings.
- There is no per-token fee, but you pay in hardware and time.
- Open models differ from hosted frontier models. Test them on your own tasks with a method like our one-afternoon evaluation.
Mistakes to avoid
- Pulling a model larger than your memory, then blaming Ollama for slow replies.
- Filling the system drive with models. Move storage with OLLAMA_MODELS first.
- Running an install script you have not read.
Ollama: frequently asked questions
What is Ollama?
A tool for running open AI models on your own computer. It is a command-line tool plus a local REST API at http://localhost:11434.
How much disk space does Ollama need?
At least 4 GB for the install. Models can be “tens to hundreds of GB.” Set the OLLAMA_MODELS environment variable to move model storage.
What are the Ollama requirements on Windows?
Windows 10 22H2 or newer, NVIDIA drivers 551.61 or newer for NVIDIA GPUs, a ROCm v7 driver stack or a Vulkan-capable Radeon driver for AMD. Administrator rights are not required.
Is it safe to pipe the Ollama install script into a shell?
Any piped script runs whatever it contains. Read it first, or use the downloadable installer.
Can I run Ollama without a GPU?
Yes. With less VRAM, Ollama can use system RAM, but responses may be slower.
Ollama or LM Studio?
Pick by workflow: command line and API, or a desktop app with a graphical interface. Both offer a local server.
Sources
- Ollama on GitHub
- Ollama docs: quickstart
- Ollama docs: Windows
- LM Studio docs: app
- LM Studio docs: system requirements
Facts last checked on 19 October 2026; not tested by us. Model names and requirements change.
