What it is for
Ollama lets you run open models on your own computer or in the cloud, with full privacy locally. The README shows downloads for macOS, Windows and Linux, a Docker image and libraries for Python and JavaScript. The ollama command starts an interactive choice, and ollama launch connects Ollama with agents such as Claude Code or Codex. A REST API is also available at localhost:11434.
How to start
Install with the script from the README or the installer, then run a model with the ollama run command and a model name from the library. When working with agents and tools set a longer context: the documentation by default picks the length according to graphics memory and advises at least 64 thousand tokens for tasks with agents and coding. The ollama ps command shows whether a model runs on the GPU or on the processor.
curl -fsSL https://ollama.com/install.sh | sh
ollama run gemma4What to watch out for
Check the licence of every model you download, because the licence of the Ollama program (MIT) does not cover models. A larger context increases the memory requirement, and moving a model to the processor slows work down noticeably. A server listening on the network needs securing: do not expose it without authentication.
Further reading
More in the article Ollama, LM Studio and llama.cpp compared and in the Local AI hub. You will find the other repositories on the repository list. Star and release data come from the freshest live data, and the description from the README as of 20 October 2026.