Ollama
Download and run large language models on your own machine with one command.
- Stars
- 187k
- Forks
- 19k
- Licence
- MIT
- Language
- Go
- Updated
At a glance
- One command to pull and run a model
- Local HTTP API, including an OpenAI-compatible endpoint
- Uses Apple Silicon, NVIDIA and AMD GPUs automatically
- Modelfiles package a model, prompt and parameters together
Get started
Install on Linux
curl -fsSL https://ollama.com/install.sh | shRun a model
ollama run llama3.2What it is
Ollama makes running a language model locally feel like running a container. ollama run downloads the weights, loads them onto whatever GPU is available and opens a chat prompt. The same process serves an HTTP API on localhost:11434 for applications to call.
Why it stands out
It removes the setup. Quantised models are published in a registry, hardware acceleration is detected automatically, and a Modelfile describes a model, its system prompt and parameters in one place. An OpenAI-compatible endpoint means much existing tooling works by changing a base URL.
Who it is for
Developers prototyping against local models, and anyone who wants prompts and data to stay on their own computer.
