DT

Written and reviewed by DeviceterraDeviceterra editorial team · Updated August 2026

KEY TAKEAWAY

A safe installation records the hardware, exact model file, engine, settings, and test result so the setup can be checked and repeated.

TRY IT YOURSELF

Install and verify your first model

A good installation guide should leave you with a working model and a record of what you installed. These steps use Ollama because it offers a simple first path.

  1. 1

    Record your computer

    Write the operating system, CPU, RAM, GPU, VRAM, and free storage.

  2. 2

    Download Ollama

    Visit ollama.com/download. Choose Windows or macOS and run the installer. On Linux, open Terminal and use the official command below.

    curl -fsSL https://ollama.com/install.sh | sh
  3. 3

    Open a command window

    Windows: Start, type PowerShell, select it. Mac: Command and Space, type Terminal, press Enter. Linux: press Ctrl, Alt, and T.

  4. 4

    Check Ollama

    Run this command. It should print the installed version.

    ollama -v
  5. 5

    Start a small model

    Run the command and wait for the download.

    ollama run gemma3:1b
  6. 6

    Finish the test

    Ask one safe question. Type /bye when finished.

How to know it worked
  • ollama -v prints a version.
  • The model downloads and answers.
  • You recorded the exact model and engine version.
If something goes wrong
  • If the command is not found, close and reopen PowerShell or Terminal.
  • If the model stops because of memory, choose a smaller model and close heavy apps.
WATCH THE EXPLANATION

Learn Ollama in 15 Minutes

Tech With Tim gives a concise walkthrough for installing Ollama and getting a local model running. Use it with the checklist below so you also record hardware, versions, safety settings, and real test results.

UNDERSTAND THE DETAILS

Use the explanations below when you want to know why each step matters.

01

Record the computer first

  • Operating system and version.
  • CPU model and core class.
  • Total RAM.
  • Exact GPU name, type, and dedicated VRAM when present.
  • Free storage.
  • The number of people who will use the model.
02

Confirm the model and engine

Choose a model for the task, then confirm that the engine supports its architecture and file format. A model may appear in several formats, and the setup steps are not always the same.

Download the engine from its official website. Use an exact model tag or trusted file. Save the names and versions so the test can be repeated.

03

Protect the first test

  • Use non-sensitive information.
  • Keep the service on the local computer or private network.
  • Do not expose an unauthenticated model port to the internet.
  • Close heavy apps if memory is limited.
  • Start with a short context and one user.
  • Keep a way to stop the service.
04

Measure the result

Run five to twenty prompts from the real job. Check whether the first word arrives soon enough, whether the full answer is fast enough, and whether memory stays stable.

Score correctness, instruction following, made-up facts, and unsafe answers. A setup is not complete just because the model opens.

05

Save a working baseline

  • Record the exact model and quantization.
  • Record the engine and driver versions.
  • Save important prompts and settings.
  • Write down the pass mark and results.
  • Back up configuration files that do not contain secrets.
  • Test again after every important change.
RESEARCH SOURCES

Official facts and real user evidence

Official documentation supports product and model facts. Community discussions show real setups, failures, and questions. A community result is supporting evidence, not a promise that another computer will perform the same way.

MAKE IT PRACTICAL

Find a model your computer can run.

MamiLens checks your hardware and shows a careful starting point.

Run the free compatibility check →

This guide is educational. Model software, licenses, and hardware support can change. Check official sources before an important deployment.