#Ollama Tutorial

Welcome to the Ollama tutorial. It follows the official Ollama docs and the ollama/ollama project so you can run open models on your machine (or Ollama Cloud) and wire them into apps through the CLI, REST, and an OpenAI-compatible API.

Ollama packages download, local inference, and an HTTP server into one workflow: after install, ollama run gemma4 starts a chat, and http://localhost:11434/api accepts requests. Code, notes, and customer text can stay on-device—you do not need a cloud vendor account just to try a prompt.

---

#Table of Contents

#Basics

1. Introduction — Local LLM runtime, vs vLLM / cloud APIs, privacy
2. Installation — macOS / Windows / Linux, ollama --version, GPU and VRAM
3. Quick Start — Interactive menu, pull / run / /bye, ls / ps / stop

#Models and APIs

4. Models and quantization — Library, tags, quantization, cloud models
5. REST and OpenAI-compatible API — /api/generate, /api/chat, /api/embed, /v1
6. Python and JavaScript — Official ollama libraries, streaming, Client

#Customization and retrieval

7. Modelfile — FROM, SYSTEM, PARAMETER, ollama create
8. Embeddings and a RAG teaser — Embed models, vector API, LangChain RAG

#Integration and practice

9. Integrations — Dify, LangChain, Hermes / OpenClaw, Cursor, Docker networking
10. Practical examples — Local Q&A CLI, OpenAI-compatible client, coding Modelfile
11. Resources — Official docs, model library, GitHub

---

#Learning path

#Prerequisites

• Comfort with a terminal (macOS / Linux bash, or Windows PowerShell / curl.exe)
• The ideas of model, prompt, and token
• Disk space: models are often several GB to tens of GB
• Optional: an NVIDIA GPU (CUDA) or Apple Silicon (Metal). CPU-only works, just slower

#Conventions in this tutorial

• Official quickstart currently uses gemma4. Library tags change—trust ollama.com/library. For Chinese-heavy work try qwen2.5; Llama-family tags are common general alternatives.
• Local API base: http://localhost:11434/api. OpenAI-compatible base: http://localhost:11434/v1.
• Cloud uses the same API shape at https://ollama.com/api (account / API key).
• Commands and model names in docs.ollama.com win when this page and the site disagree.

#What you can do after this course

• Install on macOS / Windows / Linux and confirm ollama --version
• Finish a local chat with gemma4 (or qwen2.5 / Llama) and leave with /bye
• Land one request each on http://localhost:11434/api and http://localhost:11434/v1 (PowerShell or curl.exe on Windows)
• Write a Modelfile and ollama create a reusable assistant name
• Point Dify, LangChain, Cursor, or a Dockerized app at Ollama

#Related courses

---

#Next steps

• Introduction
• Installation
• Quick Start