Home › Guides

How to run DeepSeek and Qwen privately on your own computer

By AI On Hand · Updated 14 November 2026 · 6 min read

Why run them locally?

DeepSeek (from DeepSeek) and Qwen (from Alibaba) are among the strongest open-weight AI models — free to download and run yourself. The catch is the hosted apps: the DeepSeek app and website, for example, store data on servers in China. Running the model on your own computer avoids that entirely. It's just a file on your machine; it doesn't phone home.

What hardware do you need?

Your computer's memory	Model size that runs well	Good for
8 GB	1.5B – 4B	Quick drafts, simple Q&A
16 GB	7B – 8B	Everyday writing, summaries, coding help
32 GB	14B – 32B	Stronger reasoning, longer documents
64 GB+ or a 24 GB GPU	32B – 70B	Near cloud-level quality for many tasks

Macs with Apple Silicon (M1 or later) are excellent because the graphics chip shares the main memory. On Windows and Linux, an NVIDIA graphics card makes a big speed difference.

Step 1 — Install Ollama

Download it free from ollama.com for Mac or Windows and install it like any app. On Linux:

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

Step 2 — Download and run a model

Open a terminal and run one of these. The first run downloads the model (several gigabytes); after that it starts in seconds.

ollama run qwen3:8b

ollama run deepseek-r1:8b

Then just type your question. Swap 8b for a smaller or larger size to match your computer. New versions appear often, so check the Ollama model library for the latest Qwen and DeepSeek releases.

Step 3 — Prove it's private

Turn off Wi-Fi and ask another question. It still works — because everything happens on your machine. By default Ollama only listens on your own computer (localhost), not the internet.

Step 4 — Prefer a normal app window?

• LM Studio — a free desktop app: search for a model, click download, start chatting. It can also chat with your documents.
• Open WebUI — a ChatGPT-style interface on top of Ollama, with user logins and document chat. Ideal for sharing with a team inside your office network.

Honest limitations

• Smaller than the cloud giants. Local models are very good at writing, summarising, document Q&A and coding help, but the largest cloud models are still stronger on the hardest problems.
• Speed depends on your hardware. Bigger models on modest machines can be slow.
• Built-in viewpoints. Chinese models may refuse or give filtered answers on politically sensitive topics related to China. For most business tasks this doesn't come up.
• Check the licence. Many DeepSeek and Qwen releases use permissive licences, but read the licence on the model page before commercial use.

Using it in a business

• Keep Ollama bound to localhost. Never expose its port to the internet.
• To share with colleagues, run it on one office machine behind a VPN or a login (Open WebUI handles this).
• Write down which tasks go to private AI and which can use cloud AI — and tell your team.

Frequently asked questions

Is running DeepSeek locally really private?

Yes. When you run the model with Ollama or LM Studio on your own computer, your prompts and files are processed on your machine and not sent anywhere. The privacy concerns apply to DeepSeek's hosted app and website, not the downloadable model.

Is it legal to use DeepSeek or Qwen for business?

Generally yes — many releases are open-licensed for commercial use. Check the licence for the exact model, and any rules in your sector or contracts about which software you may use.

Qwen or DeepSeek — which is better?

Both are strong. Qwen offers many sizes and is a great all-rounder; DeepSeek's R1 models are known for step-by-step reasoning. Try both on your real tasks — switching takes one command.

Do I need a graphics card?

No. Apple Silicon Macs run them well without one, and smaller models run on ordinary PCs. A modern NVIDIA GPU makes larger models much faster on Windows and Linux.

Related