How to Use Ollama: The Beginner's Guide (Install, First Model, First Chat)

CodexFuse Team · 2026-10-29T10:28:02.226Z · 15 min read min read

Learn how to use Ollama: install it on Mac, Windows or Linux, run your first local AI model, size RAM before you download, and keep it private — free forever, offline.

Learning how to use Ollama is the fastest way to get a real AI running on your own computer instead of someone else's server. Ollama is a free, open-source tool that downloads open AI models like Llama, Gemma, Mistral, and Qwen and runs them locally, so your prompts never leave your machine. This beginner's guide walks you from zero to your first private AI chat in about fifteen minutes: what Ollama is, what your computer needs, how to install it on Mac, Windows, and Linux, how to pick your first model by the RAM you actually have, the daily commands, the local API, and the security rules that matter. If you outgrow your laptop later, you can self-host Ollama on a VPS for bigger models — but start on the computer in front of you.

How do you use Ollama?

Using Ollama means installing the Ollama program, pulling one AI model with a single command, and chatting with it in your terminal — all free, all offline after the download. There is no account, no subscription, and no cloud involved for local models.

1. Install Ollama from the official download page for your operating system (Mac, Windows, or Linux).
2. Check it works by opening a terminal and typing ollama -v — you should see a version number.
3. Run your first model with ollama run llama3.2 — Ollama downloads it the first time, then opens a chat at the >>> prompt.
4. Chat normally — type a question, press Enter, and the answer is generated entirely on your machine.
5. Leave the chat by typing /bye, and manage models later with ollama ls, ollama ps, ollama stop, and ollama rm.

What is Ollama, exactly?

Ollama has two parts that work together. A background server does the heavy lifting of loading a model into memory and running it efficiently, and a command-line tool called ollama talks to that server. On macOS and Windows there is also a desktop app with its own chat window, so you are not forced to live in the terminal.

Everything about local use is free. The project's code is published on GitHub under the MIT license, and the official pricing page lists running models locally on a $0 Free plan — you only ever pay if you choose Ollama's optional cloud models, which run on their servers instead of your computer (Coursiv's Ollama tutorial, updated October 2026 and checked against Ollama's own documentation).

The models themselves are open-weight versions of the kind of AI that powers chatbots: families like Meta's Llama, Google's Gemma, Mistral, Microsoft's Phi, and Alibaba's Qwen. They are free to download and run. They are not quite as powerful as the biggest paid cloud models, but they are more than capable for everyday writing, summarizing, answering questions, and coding help — and they keep working with your internet switched off entirely.

What does your computer need before you install it?

The model you pick is a RAM decision first, not a taste decision. Models are big files, and the biggest beginner mistake is downloading a model that does not fit your machine. Check two things before anything else: your free disk space and your memory.

For disk, plan room for more than one model: Ollama's own documentation warns that models can be tens to hundreds of gigabytes in total, and the Windows installer alone needs at least 4 GB before any models. For memory, the honest rule is that a 4-bit quantized model needs its file size plus room for the conversation and your operating system. The TechCommune Ollama cheat sheet gives this rough guide:

The practical takeaway: 8 GB of RAM is the minimum for a comfortable experience and 16 GB is where it gets pleasant, per both the TechCommune cheat sheet and Udemy's beginner guide. A GPU (NVIDIA, AMD, or Apple Silicon) speeds things up substantially — without one, models run on the CPU, only slower. To find your numbers: on a Mac, Apple menu then About This Mac; on Windows, Task Manager then Performance.

How do you install Ollama on Mac, Windows, and Linux?

Ollama installs like any normal application on each operating system, and the steps below are checked against Ollama's official documentation as of October 2026.

macOS

You need macOS Sonoma (version 14) or newer. Apple M-series Macs get both CPU and GPU support; older Intel Macs run on the CPU only, which is slower. Download the app from the official Ollama download page, open the .dmg file, and drag Ollama into your Applications folder. Start the app once: on first launch it checks that the ollama command is available and asks permission to create a link in /usr/local/bin if it is missing.

Windows

You need Windows 10 version 22H2 or newer, Home or Pro — and if you have an NVIDIA card, driver 551.61 or newer. Download OllamaSetup.exe from the download page and run it: it installs into your user folder, needs no administrator rights, and then runs in the background with a tray icon. Open a new PowerShell or Command Prompt window and type ollama -v to confirm.

Linux

The download page offers a one-line install script:

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

That downloads a script and runs it, so if you prefer, save and read it first. The script installs the program and creates a systemd service that starts at boot. If the background service did not start on its own, launch it with ollama serve from a second terminal, then confirm with ollama -v.

How do you run your first model?

With Ollama installed, one command is enough to go from nothing to a working chat:

ollama run llama3.2

If the model is not on your computer yet, Ollama pulls it first and shows a progress bar — this is the download, so it takes a few minutes the first time and needs internet. When you see the >>> prompt, type a question and press Enter. The answer is generated entirely on your own machine. To paste text that spans several lines, wrap it in triple quotes ("""), as the official CLI reference shows. Type /bye to end the chat.

Try a real first task rather than small talk: paste a long email you wrote yourself and ask "Summarize this in three bullet points and list any dates mentioned." Then follow up with "Rewrite the second bullet for a busy manager." Summaries and rewrites of your own text are where small local models shine. They can still misread dates or invent details, so check anything that matters against the original.

Which model should you run first?

Pick by memory first and quality second. A fast small model teaches you more than a large one that crawls. Here is the decision framework, built from the model pages on the official Ollama library (checked live on October 29, 2026) and the download sizes Ollama publishes:

Two truths no short tutorial leads with. First, the model names move fast: Gemma 4 (with vision, tools, thinking, and audio variants) is current as of late October 2026, alongside Qwen 3.5, GPT-OSS (OpenAI's open-weight models), and Gemma 3n built for everyday devices — always browse the official library for what exists today rather than copying a model name from an old guide. Second, check the license on the model's library page before you build anything for work: Ollama itself is MIT-licensed, but each model carries its own — Gemma 4 E2B ships under Apache 2.0, while Llama 3.2 ships under the Llama 3.2 Community License Agreement with an acceptable use policy, and licenses can limit commercial use. The DeepSeek beginner guide covers the hosted-model lane if you would rather not run anything locally.

What are the everyday Ollama commands?

After the first chat, a handful of commands covers almost all daily use. They come from the official CLI reference and the TechCommune cheat sheet:

A tidy habit: run ollama ls once a month and remove models you no longer use. Models live in ~/.ollama/models on Mac and Linux, and C:\Users\%username%\.ollama\models on Windows — and if a drive fills up, you can point them at another disk with the OLLAMA_MODELS environment variable.

How do you chat better inside Ollama?

The chat prompt has its own little command set, and learning four of them makes the tool dramatically more useful. Type /? inside any chat to see the full list.

• /bye — exit the chat (or press Ctrl+D).
• /set parameter temperature 0.2 — change a setting for this chat on the fly; lower is more focused, higher is more creative.
• /set parameter num_ctx 8192 — raise the context window, how much text the model keeps in view; a bigger window uses more memory.
• /set system "You are a Linux expert. Be brief." — set the system prompt that steers the model's behavior.
• /save mychat and /load mychat — save a session as a model and reload it later.
• /clear — clear the conversation context without leaving.

Practical tricks from the cheat sheet: feed a file in with ollama run llama3.2 "Summarize this file" < notes.txt, pipe logs straight at it with tail -200 /var/log/syslog | ollama run llama3.2 "Summarize any problems. Be brief.", and stop a long answer mid-stream with Ctrl+C.

How do you get a friendly chat window instead of the terminal?

Ollama is only the engine — the friendly chat UI comes separately, and that confuses most beginners. You have two routes. On macOS and Windows, the official Ollama desktop app has its own chat window (announced July 30, 2025), including dropping in text files or PDFs, and its settings include a context-length slider. On the newest app, open Ollama and select Apps to connect Claude or ChatGPT's desktop app to your local models — local brains inside a polished window.

The other route is a community web interface like Open WebUI, which connects to Ollama over the local API. The standard install is one Docker command:

docker run -d --name open-webui -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:/app/backend/data ghcr.io/open-webui/open-webui:main

Then open http://localhost:3000, create the first (admin) account, and pick a model. If it cannot find Ollama, set the Ollama URL to http://host.docker.internal:11434 in its settings. These are third-party tools, not made by Ollama, so check who maintains them and what permissions they ask for before you install — and never expose Open WebUI to the internet without accounts, HTTPS, and a firewall.

How do you use Ollama from your own programs?

Once a model runs in the terminal, other programs can reach it through the local API at http://localhost:11434 — the same API that chat apps, note tools, and coding plugins use. A raw request from a Mac or Linux terminal looks like this:

curl http://localhost:11434/api/chat -d '{
  "model": "llama3.2",
  "messages": [{"role": "user", "content": "Give me three tips for a tidy desk."}],
  "stream": false
}'

The answer comes back as JSON. From Python, install the official library with pip install ollama and call ollama.chat(model="llama3.2", messages=[...]). Ollama also speaks an OpenAI-compatible endpoint at http://localhost:11434/v1/chat/completions, so most existing AI tools can point at it unchanged — use that base URL and any dummy API key, like "ollama", in tools that ask for an OpenAI-style setup.

Can you save your favorite settings as a reusable model?

Yes, with a Modelfile — a small text file that bundles a base model with your favorite settings and instructions. Here is a complete example for a meeting-notes summarizer:

FROM llama3.2
PARAMETER temperature 0.3
PARAMETER num_ctx 8192
SYSTEM """You are a concise meeting-notes assistant. Use short bullet points and list action items with owners."""

Save it as Modelfile, then run ollama create notes-helper -f Modelfile, followed by ollama run notes-helper. The lower temperature makes answers steadier, which suits summaries. You can also inspect any model you downloaded with ollama show --modelfile <model> to see the settings it was built with.

Is Ollama really free forever?

For local use, yes — and that is the line to understand clearly. Ollama's software is open source (MIT), and the official pricing page puts running models locally on the $0 Free plan: your only costs are your own hardware and electricity. The paid plans (Pro around $20 per month in the October 2026 tutorial) buy usage credits for cloud models — models whose names end in :cloud and run on Ollama's servers instead of your computer. Sign in to use them.

The distinction matters for privacy. Ollama's FAQ states it does not see your prompts when you run models locally. Cloud models are different: your prompts travel to Ollama's servers for processing. If you want to be sure nothing leaves your machine, the FAQ explains how to turn off cloud features with disable_ollama_cloud in ~/.ollama/server.json. And even locally, keep passwords, ID numbers, and card details out of AI tools — a model cannot keep a secret better than a locked file can. If you would rather let someone else's servers do the work, a hosted chatbot like the one in our ChatGPT beginner guide is the simpler lane.

How do you keep Ollama secure?

Ollama has no login, so anyone who can reach its port can use your GPU and your models — which makes one rule the whole security story: keep the API private. By default Ollama listens only on 127.0.0.1, meaning your own computer, and the local API requires no authentication. Guides online sometimes suggest setting OLLAMA_HOST to 0.0.0.0 so other devices can connect. Do not do that casually: it opens the API to your whole network, and if your router forwards port 11434, to the internet, where strangers could run models on your machine or delete them — and, if you are signed in, burn through your cloud credits.

Safer habits: leave OLLAMA_HOST at its default unless you have a specific need; never forward port 11434 on your router or open it in a cloud server's firewall; if you need remote access, reach it through a private VPN or an SSH tunnel (ssh -L 11434:localhost:11434 user@server) rather than exposing the port. Note that models stay loaded for 5 minutes by default after you finish — use ollama stop <model> to free the memory immediately, or set OLLAMA_KEEP_ALIVE to control the timeout.

When should a VPS run Ollama instead of your laptop?

Your laptop is the right place to learn, but three situations push Ollama onto a server. First, the RAM ceiling: if you want to experiment with a 27B or 70B model and your machine has 16 GB, the math from the sizing table simply does not work — a rented machine with 32–64 GB does. Second, always-on use: a chatbot your team can reach, or a model serving an app's API around the clock, belongs on a machine that never sleeps. Third, GPU access you do not own: a cloud GPU instance can be cheaper than buying a graphics card for a weekend experiment — just stop the instance when you finish so it does not keep costing money.

The security rules get stricter on a server: keep Ollama on 127.0.0.1, firewall the port to trusted addresses only, or put a reverse proxy with authentication and HTTPS in front. Our full walkthrough covers the whole stack — self-host Ollama on a VPS with Open WebUI — for when your laptop runs out of headroom.

Sources

• Coursiv — "Ollama Tutorial: Install and Run AI Models Locally" (updated October 2026; commands and requirements checked against Ollama's own documentation): coursiv.io/blog/ollama-tutorial
• Ollama Library (ollama.com/library — current model list, pull counts, and tags; checked October 29, 2026): ollama.com/library
• Ollama Documentation — Quickstart (updated October 27, 2026; desktop app, Apps connections, coding agents, cloud models): docs.ollama.com/quickstart
• TechCommune — "Ollama Local Setup and Cheat Sheet" (RAM sizing table for 4-bit quantized models, install commands, security notes): techcommune.org/docs/ai/ollama.html
• Udemy Blog — "What is Ollama? A Beginner's Guide To This Platform" (8 GB RAM minimum, 16 GB+ recommended): blog.udemy.com/what-is-ollama

Frequently Asked Questions

Is Ollama free to use?

Yes, for local use. The software is open source under the MIT license, and the official pricing page lists running models locally on the $0 Free plan — your only costs are your own hardware and electricity. Paid plans only add usage credits for Ollama's optional cloud models.

How much RAM do I need to run Ollama?

8 GB is the minimum for a comfortable experience and 16 GB is where it gets pleasant. The model matters more than the program: a 1–3B model needs about 1–2.5 GB, a 7–8B model needs 4.5–6 GB, and a 70B model needs 40 GB or more. Start with a 3B model and step up only if your machine has the memory.

Does Ollama work on Mac, Windows, and Linux?

Yes, all three. macOS needs Sonoma (14) or newer, Windows needs 10 version 22H2 or newer, and Linux installs with a one-line script. Apple Silicon Macs use the GPU automatically; older Intel Macs run on the CPU.

Does Ollama need the internet?

Only to download models. Once a model is on your computer, everything — chats, the API, your custom models — works fully offline, and your prompts never leave your machine.

Is running AI locally actually private?

For local models, yes: Ollama's FAQ states it does not see your prompts when you run models locally. Cloud models (names ending in :cloud) send your prompts to Ollama's servers instead, so keep anything sensitive on local models — and never feed passwords or ID numbers to any AI tool.