---
title: "install ollama"
description: "Understand what Ollama is, install it, smoke-test the CLI, then pull a model that fits your RAM."
---

# 01 — Install Ollama

## Objective

Install Ollama, prove the CLI works, pull a model sized for **your** RAM, and learn to watch Memory Pressure.

Tutor pacing: concept → check → install → verify → pull model → verify list.

---

## Concept: What Ollama is doing

Ollama runs large language models **on your machine**:

- A background service (menu bar icon on Mac)  
- A CLI (`ollama pull`, `ollama run`, …)  
- Models as downloadable weights on disk  

You are not “installing ChatGPT.” You are installing a **local runtime** plus one model file big enough to be useful and small enough to keep Memory Pressure green.

---

## Concept check

**Q1.** In one sentence, what does Ollama provide that a website chatbot does not?

**Q2.** Why might a “smarter” larger model be the wrong pull for an 8GB machine?

<details>
<summary>Concept answer key — attempt first</summary>

## Answer key (concept)

### Q1

**Model answer:** Local inference — models run on your hardware/offline, weights on disk.

**Pass criteria:** local / on-device (or offline)

### Q2

**Model answer:** It may not fit in RAM; thrashing / red memory pressure makes it unusable.

**Pass criteria:** RAM / fit / performance

</details>

---

## Gate: Install Ollama

1. Go to [ollama.com](https://ollama.com) → **Download**  
2. Open the app; confirm the llama icon in the menu bar  

**Verify:**

```bash
ollama --version
```

You should see something like `ollama version is 0.x.x`. **Stop here if this fails.**

---

## Gate: Know the CLI

```bash
ollama pull <model>     # Download a model
ollama list             # Show models you've downloaded
ollama run <model>      # Start an interactive chat with a model
ollama rm <model>       # Delete a model to free space
```

---

## Gate: Pull a model that fits

| Your RAM | Recommended Model | Pull Command                  |
| -------- | ----------------- | ----------------------------- |
| 8GB      | `nemotron-3-nano` | `ollama pull nemotron-3-nano` |
| 16GB     | `gemma4:26b`      | `ollama pull gemma4:26b`      |
| 32GB+    | `qwen3:30b-a3b`   | `ollama pull qwen3:30b-a3b`   |

Course examples use `gemma4:26b` when it fits:

```bash
ollama pull gemma4:26b
ollama list
```

**Pass:** model appears in `ollama list`.

---

## Gate: Memory Pressure

Open **Activity Monitor → Memory**. Watch **Memory Pressure** during a short `ollama run`:

- **Green** — headroom  
- **Yellow** — tight  
- **Red** — too big; pull a smaller model  

> Goal: smartest model that stays green during chat.

---

## Check your understanding

**Q3.** What command proves Ollama is installed before you spend time on a multi-GB pull?

**Q4.** After a pull, how do you confirm the model is on disk?

<details>
<summary>Answer key — attempt every question first</summary>

## Answer key

### Q3

**Model answer:** `ollama --version` (and/or menu bar icon).

**Pass criteria:** version or service evidence

### Q4

**Model answer:** `ollama list` shows the model name.

**Pass criteria:** ollama list

</details>

---

**Next: [02 — Your First Prompt →](/run-ai-locally/01-getting-started/02-first-prompt/)**