Install Ollama and a Local LLM on macOS and Windows

A local LLM is an AI model that runs on your own computer instead of on a company's server. Ollama is a free app that downloads these models and runs them for you on macOS and Windows. After the first download, the model works without an internet connection, costs nothing per request, and your notes and chart screenshots never leave your computer.

This guide shows how to install Ollama, choose and download a model, check that it works, and connect it to Red Candle.

What you need

	macOS	Windows
System	macOS 14 Sonoma or newer	Windows 10 22H2 or newer, Home or Pro
Processor	Apple Silicon (M1 or newer) uses the GPU. Intel Macs run on the CPU only, which is slower.	Any recent PC. An NVIDIA or AMD graphics card makes replies faster.
Memory (RAM)	8 GB for a small model, 16 GB or more for a larger one	8 GB for a small model, 16 GB or more for a larger one
Disk space	Ollama plus 3 to 6 GB per model	About 4 GB for Ollama plus 3 to 6 GB per model

On Windows with an NVIDIA card, update the driver to version 551.61 or newer. AMD cards work with a current Radeon driver.

Choose a model

For a trading journal you want a vision model, one that can read chart screenshots as well as text. Red Candle's guides use Qwen2.5-VL, which Ollama offers in several sizes:

Model	Download	Suits	What to expect
qwen2.5vl:3b	3.2 GB	8 GB RAM	Fine for checklist rules and short explanations. It can misread charts and mix up levels.
qwen2.5vl:7b	6.0 GB	16 GB RAM or more	Reads and explains charts noticeably better, and replies more slowly.

Start with the 3B model if you are not sure. You can download the 7B model later and switch in a minute.

Install Ollama on macOS

1. Go to ollama.com/download and download the macOS version.
2. Open the downloaded Ollama.dmg and drag Ollama into Applications.
3. Open Ollama from Applications. The first time, it may ask for permission to install the ollama command line tool. Allow it, because you will use it in the next steps.
4. The Ollama icon appears in the menu bar at the top of the screen. That means it is running.

To check the installation, open Terminal (press ⌘ Space and type "Terminal") and run:

ollama --version

You should see a version number. Qwen2.5-VL needs Ollama 0.7.0 or newer. If yours is older, download Ollama again to update it.

Install Ollama on Windows

1. Go to ollama.com/download and download the Windows version, OllamaSetup.exe.
2. Run it and follow the installer. It installs into your user folder and doesn't need administrator rights.
3. Ollama starts by itself. Its icon appears in the system tray, at the bottom right next to the clock.

To check the installation, open PowerShell (press the Windows key and type "PowerShell") and run:

ollama --version

If PowerShell says ollama is not recognized, close it and open a new PowerShell window. A window opened before the installation doesn't know the new command yet.

Download the model

The commands are the same on macOS and Windows. In Terminal or PowerShell, run one of these and wait until it says success:

ollama pull qwen2.5vl:3b

ollama pull qwen2.5vl:7b

Then list your installed models:

ollama list

The name in the list, for example qwen2.5vl:3b, is exactly what you will type into Red Candle later.

Test the model in the terminal

Before you connect anything, check that the model answers:

ollama run qwen2.5vl:3b

Type a short question, for example "Explain what a higher low is in one sentence.", and press Enter. The first answer can take a minute while the model loads into memory. Answers after that are faster. Type /bye to leave.

You can also open http://localhost:11434 in your browser. If it shows "Ollama is running", the local server that other apps connect to is ready.

Connect the model to Red Candle

1. Open Settings → AI Provider.
2. Provider: choose Local (Ollama / LM Studio).
3. Server URL: http://localhost:11434/v1. This is Ollama's default and is already filled in.
4. Model: the name from ollama list, for example qwen2.5vl:3b.
5. Click Test Connection.

The first test loads the model and can take 1 to 2 minutes. While it waits you see "Server found · waiting for the model". When it shows "✓ Connected · responded in …s", you're ready. No API key is needed. Leave AI depth on Auto, which uses Basic for local models.

Now you can use Explain with AI in AI Analysis and Similar Patterns, and the other AI features, with everything staying on your computer. The full settings are described in Connect an AI provider.

Where models are stored

Models are large files, so it helps to know where they go:

• macOS: in the .ollama folder in your home folder.
• Windows: in the .ollama folder in your user folder. To store models on another drive, set the environment variable OLLAMA_MODELS for your user account to a folder on that drive, then quit Ollama from the tray and open it again.

To free up space, remove a model you no longer use:

ollama rm qwen2.5vl:7b

Fix common problems

• Red Candle shows "Not connected". Ollama isn't running. Open it again from Applications or the Start menu, then click Test Connection.
• The model isn't found. The name in Model doesn't match ollama list. Copy the name exactly, including the part after the colon.
• Replies are very slow. The model shares memory with your other apps. Close heavy apps such as browsers with many tabs, or use the 3B model.
• The download stops. Run the same ollama pull command again. It continues where it stopped.
• Explanations mix up numbers. Small local models can do that. Compare the numbers with the panels in the app, try the 7B model, or use a cloud provider with your own API key for better explanations.

More fixes are in AI isn't connected or is slow.

Key takeaways

• Ollama runs AI models on your own computer on macOS 14 or newer and Windows 10 22H2 or newer.
• For chart screenshots, use a vision model such as qwen2.5vl:3b (8 GB RAM) or qwen2.5vl:7b (16 GB RAM).
• Install Ollama, run ollama pull, check with ollama list, and test with ollama run.
• In Red Candle, choose Local (Ollama / LM Studio), enter the model name and click Test Connection.
• A local model is free and private, but small models make more mistakes, so always check the numbers.