Lakehouse Software
Home / Local AI Lab / 03

First conversation

Run your first local model

Download a small model, chat with it, and verify that inference continues after you disconnect the network.

Chapter 03/10Time 15–30 minCost $0 software

Download a first model

For a typical 16 GB machine, this guide starts with Qwen 3.5 2B. If your machine is memory-constrained, use the 0.8B variant. If you have more headroom, you can try 4B or 9B later.

ollama run qwen3.5:2b

The first run downloads the model. When the prompt appears, type:

Explain in two sentences what a local language model is.

Prove the model is local

  1. Finish the download first.Do not disconnect the network while model files are still being retrieved.
  2. Disconnect Wi-Fi or Ethernet.You are testing runtime behavior, not the download process.
  3. Start the model again.Run the same ollama run command and ask a normal question.
What this proves

The inference process and model files are on your machine. It does not prove the model's answer is current or correct, and it does not make every future feature offline automatically.

Try another size

ollama run qwen3.5:0.8b
# or
ollama run qwen3.5:4b

Choose the smallest model that feels responsive enough to keep going. The controller we build later does not care which model size you choose.

Checkpoint

You can chat with the model while the computer is disconnected from the Internet.

Finished this chapter?
Your progress is stored only in this browser.