How to Run DeepSeek With Ollama, Step by Step

The short version

Ollama runs DeepSeek-R1 on your computer with one command. It’s free, it works offline after the download, and your prompts stay on your machine. Just make sure you pick a local model, not a cloud one.

Step by step

1. Install Ollama

Go to ollama.com/download and install it for macOS, Windows or Linux. The page also gives a one-line install command for each.

2. Run DeepSeek

Open Terminal (Mac) or PowerShell (Windows) and type:

ollama run deepseek-r1

The first run downloads the default model, DeepSeek-R1-0528-Qwen3-8B (5.2 GB). When it’s done, you can type messages straight into the terminal.

3. Pick a different size if you need one

Start with the default. If replies are slow, go smaller; if you have plenty of memory, try 14b. See run DeepSeek locally for how to choose.

4. Use it from other apps

Ollama runs a local server at http://localhost:11434. Apps that support Ollama can use your DeepSeek model through it, and Ollama’s library page shows how to call it from Python or JavaScript.

5. Leave and come back

Type /bye to end the chat. Run the same command again to start a new one; the model is already downloaded.

Prompts to copy

DeepSeek’s R1 model card recommends putting all instructions in your message, not a system prompt:

Read the notes below and turn them into a numbered to-do list, most urgent first.
Add a one-line reason for each item's position.
Notes: [paste]

Summarize this in five bullet points for someone who hasn't read it. Then list
anything that sounds like a deadline or a cost.
[paste text]

Example

You run: ollama run deepseek-r1:14b on a desktop with 32 GB of memory, then paste a two-page meeting transcript.

What you get: a block of visible reasoning (DeepSeek-R1 “thinks” first), then the summary. It’s slower than the app, and everything stays on your computer.

Mistakes to avoid

• Running a cloud model by accident. Ollama also lists deepseek-v4.1-flash and deepseek-v4-pro, but only as cloud models. They run on Ollama’s servers, and you have to sign in to Ollama to use them.
• Choosing 671b. The full DeepSeek-R1 is a 404 GB download.
• Expecting current facts. A local model can’t browse the web.
• Forgetting the base model license. The 70b version is based on Llama 3.3 and the 1.5b to 32b versions on Qwen. See the DeepSeek open-weight license.

When not to use it

If you need the strongest DeepSeek model, web search or file uploads with no setup, the app or the API is the better choice. Use Ollama when privacy or offline access matters more than top quality.

Related: run DeepSeek in LM Studio and run DeepSeek locally.