Ollama is a Go-based, MIT-licensed runtime that downloads and runs open-source large language models such as DeepSeek, Qwen, Gemma, GLM, MiniMax and gpt-oss locally on a user's own machine, and it is aimed at developers and teams that want model inference without routing prompts through a hosted API.

What it is

Ollama is a command-line tool and local server that manages open-weight language models on the machine it runs on. It ships a CLI, a REST API on port 11434, and client libraries for Python and JavaScript, so the same model can be driven from a terminal, a script, or an existing agent framework. Model execution is handled by the llama.cpp backend, and the project distributes installers for macOS, Windows and Linux alongside an official ollama/ollama Docker image.

The concrete problem it solves is the gap between an open model file and a working local endpoint. Without a runtime of this kind, a developer has to source weights, build an inference engine, wire up a server, and keep everything compatible. Ollama replaces that work with ollama run and a stable HTTP interface, and it positions itself as an open alternative to the OpenAI API — the paid hosted service that many applications are already written against.

Key capabilities

• Runs models directly from the terminal, for example ollama run gemma4, and prompts for a model or an integration on first launch of the bare ollama command.
• Exposes a REST API for running and managing models, including POST http://localhost:11434/api/chat with a model, messages array, and stream flag.
• Provides official client libraries, ollama-python installed with pip install ollama and ollama-js installed with npm i ollama.
• Supports launching coding and assistant integrations through subcommands such as ollama launch claude and ollama launch openclaw, covering Claude Code, Codex, Copilot CLI, DeepSeek Harness, Droid, and OpenCode.
• Defines models through a Modelfile format and documents importing existing models, so local weights can be packaged and reused.
• Publishes a browsable model library at ollama.com/library covering families such as DeepSeek, Gemma, GLM, llama3, MiniMax, and gpt-oss.
• Runs on macOS, Windows, and Linux, and is available as the official ollama/ollama Docker image on Docker Hub.

Who uses it and how

• Individual developers who want a local chat and coding model on a laptop, installed with the shell script or the .dmg and OllamaSetup.exe downloads.
• Engineering teams that need an OpenAI-compatible local endpoint for applications already written against a chat completions style API, pointed at localhost:11434.
• Self-hosters who pair Ollama with community chat interfaces such as Open WebUI, LibreChat, Lobe Chat, NextChat, big-AGI, or Perplexica.
• Python and JavaScript application developers who call models through ollama-python or ollama-js rather than issuing raw HTTP requests.
• Users who want a personal assistant across WhatsApp, Telegram, Slack, and Discord by launching OpenClaw against the local runtime.

Getting started

Install with curl -fsSL https://ollama.com/install.sh | sh on macOS or Linux, irm https://ollama.com/install.ps1 | iex on Windows, or pull the ollama/ollama Docker image. Then run ollama run gemma4, or consult the quickstart guide at docs.ollama.com for the first steps.

How it compares

The project is positioned as an open alternative to the OpenAI API, so the contrast is mainly one of licence and control: Ollama is MIT-licensed and runs on hardware the user owns, while the OpenAI API is a paid hosted service. Because inference happens locally, prompts and model outputs do not leave the machine, which changes the data-ownership story for teams handling sensitive input. The cost model shifts from per-token billing to the hardware the self-hoster already operates.

When to use it — and when not to

Choose Ollama when local execution, an MIT licence, and a self-hosted endpoint matter more than peak model quality or managed uptime, and when the operator is prepared to supply and maintain the hardware that runs inference. Avoid it if the workload depends on the largest frontier hosted models, or if nobody is available to run and update the local server. The project currently carries 3,957 open issues, which is worth weighing before adopting it as a critical dependency.

curl -fsSL https://ollama.com/install.sh | sh

irm https://ollama.com/install.ps1 | iex

curl -fsSL https://ollama.com/install.sh | sh

ollama

ollama launch claude

ollama launch openclaw

ollama run gemma4

curl http://localhost:11434/api/chat -d '{
  "model": "gemma4",
  "messages": [{
    "role": "user",
    "content": "Why is the sky blue?"
  }],
  "stream": false
}'

pip install ollama

from ollama import chat

response = chat(model='gemma4', messages=[
  {
    'role': 'user',
    'content': 'Why is the sky blue?',
  },
])
print(response.message.content)

npm i ollama

import ollama from "ollama";

const response = await ollama.chat({
  model: "gemma4",
  messages: [{ role: "user", content: "Why is the sky blue?" }],
});
console.log(response.message.content);