Ollama: one command to run an AI model on your own computer

My take: Ollama is how I run models on my own machine: one command starts a model, and its local API answers like OpenAI's.
What to know
- Ollama downloads open models, such as Gemma, Qwen, DeepSeek or gpt-oss from its library, and runs them on your computer: macOS, Windows, Linux or Docker. It is open source under the MIT license: 187,200 stars on GitHub on November 16, 2026, and version 0.40.16 on November 9, 2026.
- On the Mac, the app has had a chat window since July 30, 2025, where you can drop text, PDFs or images. It needs macOS 14 Sonoma or later; Apple chips use their GPU, Intel Macs only their processor. Models live in
~/.ollama, and the docs warn they can take tens to hundreds of gigabytes. - Ollama listens on
localhost:11434, and its/v1endpoints follow OpenAI's API (chat, embeddings, Responses): code written for OpenAI works by changing its base URL. The client wants an API key; Ollama ignores it. - The limit: models tagged
cloud, such asgemma4:cloud, run on Ollama's servers and need an account. The others stay on your machine.
ollama run gemma4
What changes
For you if you want a local model inside your scripts and tools, not only in a chat window. The site already covers LM Studio: pick LM Studio for a polished app to browse models and chat, Ollama for an open-source engine you drive from the terminal and from your code. Start with Gemma 4 above, then point one of your tools at http://localhost:11434/v1.
Source: Ollama, “Ollama v0.40.16”, November 9, 2026
Drafted with Claude, Anthropic's AI, from my take. Reviewed and owned by me.