Install Ollama and Qwen (local AI)

Ollama runs AI models on your own computer. With the Qwen2.5-VL vision model it can read your chart screenshots — free, and nothing leaves this computer. The only time you need the internet is to download Ollama and the model.

Before you start

• Memory (RAM): at least 8 GB for the small model, 16 GB or more for the larger one.
• Disk space: about 3.2 GB (small model) or 6 GB (larger model), plus Ollama itself.
• Speed: a Mac with Apple Silicon (M1 or newer) or a PC with a recent graphics card replies fastest. On an Intel Mac or a PC without a graphics card it still works, only slower.

1. Install Ollama

macOS

1. Download Ollama.dmg from ollama.com/download.
2. Open it and drag Ollama into Applications.
3. Open Ollama from Applications. The Ollama icon appears in the menu bar — that means it is running.

Windows

1. Download OllamaSetup.exe from ollama.com/download.
2. Run it and follow the installer.
3. Ollama starts by itself; its icon appears in the system tray (bottom right, next to the clock).

To check, open Terminal (macOS) or PowerShell (Windows) and type:

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.

2. Download the Qwen model

Choose the size that suits your computer:

Model	Download	Good for
qwen2.5vl:3b	3.2 GB	8 GB RAM. Fine for checklist rules. Can misread charts and mix up levels: for better analysis, use a cloud provider.
qwen2.5vl:7b	6.0 GB	16 GB RAM or more. Reads and explains charts noticeably better.

In Terminal or PowerShell, type one of these and wait until it says success:

ollama pull qwen2.5vl:3b

ollama pull qwen2.5vl:7b

Check that it's installed:

ollama list

The model name in the list (for example qwen2.5vl:3b) is exactly what you type into Red Candle.

3. Connect it in Red Candle

1. Open Settings → AI Provider.
2. Provider: choose Local (Ollama / LM Studio).
3. Server URL: http://localhost:11434/v1 (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 into memory and can take 1–2 minutes — you'll see Server found · waiting for the model. When it shows ✓ Connected · responded in …s, you're ready. After that, replies take seconds.

No API key is needed. Leave AI depth on Auto (Basic for local models). Try Full only with the 7B model or larger, and only if the results look right.

Everyday use

• Keep Ollama running while you use the AI features — on macOS keep it in the menu bar, on Windows in the system tray. Ollama can start automatically when you log in.
• Close heavy apps if replies are slow; the model shares memory with everything else.
• Switch model: pull another one with ollama pull …, then change Model in Red Candle and click Test Connection.
• Remove a model to free up disk space: ollama rm qwen2.5vl:3b.

Problems

• Not connected — Ollama isn't running. Open it again and click Test Connection.
• …but the model "…" isn't installed — the name in Model doesn't match ollama list. Pull the model or correct the name.
• ollama is not recognized (Windows) — close PowerShell and open it again after installing.
• More fixes: AI isn't connected or is slow.

Other providers (LM Studio, ChatGPT, Gemini, Anthropic) are in Connect an AI provider.