Pull a model
Browse a catalog of open models — Llama, Mistral, Gemma, Qwen, DeepSeek — and download with a single command. Models live on your disk, not on someone else's server.
llama3.2
mistral
gemma3
ollama · run open models locally
Run large language models on your own machine.
Pull an open model, run it offline, and chat with it from your terminal or your app. No accounts, no API keys, no tokens billed by the thousand — your machine is the server.
$
ollama run llama3.2
what ollama does
One command. Any model. On your hardware.
Talk to it
Chat from the terminal, or call the OpenAI-compatible local API from any language. Streaming, tool calls, and structured output work the same as the hosted runtimes your app already targets.
http://localhost:11434
Plug it in
Drop Ollama behind the agent framework, IDE plugin, or chat client you already use — LangChain, OpenWebUI, Continue, your homegrown CLI. The runtime stays out of the way.
langchain
opencode
continue
stay close to the work
Get a note when a new model lands.
Roughly one email a month — release notes for new models, changelog highlights, and the occasional benchmark. No marketing, no upsells, unsubscribe in one click.
© Ollama · run open models locally
v0.5.0 · changelog