Ollama desktop app: install and run AI models locally

Ollama is a local model runner that downloads, manages, and serves compatible language and vision models on your computer. It provides a command-line interface and local API, while supported desktop releases add a graphical way to interact with models and files.

Running locally can keep prompts and model processing on the machine, but privacy is not automatic. Connected tools, remote model features, imported files, or a server exposed to the network can still transmit or expose data. Review each model's license and source before using it in a product.

What Ollama provides

• Local inference: run supported models without sending each prompt to a hosted model API.
• Model management: download, list, update, copy, and remove model packages.
• Local API: connect scripts and applications to the Ollama service.
• Hardware acceleration: use supported GPU or platform acceleration when available.
• Model configuration: set a system prompt and selected runtime parameters in a reusable model definition.
• Vision workflows: compatible multimodal models can analyze supplied images.

The available desktop interface, supported operating systems, hardware backends, and minimum versions can change. Use the official Ollama download page for the current installer and platform requirements.

Plan storage and memory before downloading a model

The application itself is much smaller than the models it runs. Model files can consume several gigabytes or considerably more, and a model also needs enough usable memory while running. The requirement depends on model size, quantization, context length, and hardware.

Start with a smaller model whose published requirements fit your system. Leave additional disk space for updates and more than one model version. A larger context setting can increase memory use even when the model file is unchanged.

Install Ollama on Windows

1. Download the Windows installer from Ollama's official site.
2. Run the installer and start Ollama from the Start menu if it does not start automatically.
3. Open a new PowerShell window and verify the command is available:

ollama --version

To download and start a model, use its exact current name from Ollama's model library:

ollama run MODEL_NAME

Replace MODEL_NAME with a model that fits your computer. The first run downloads its files; later runs use the local copy.

Change the model storage folder on Windows

Create or edit the user environment variable OLLAMA_MODELS and set it to the full folder where model data should be stored. Then fully quit Ollama and start it again. Confirm that the destination drive has enough free space and appropriate permissions before moving or downloading models.

1. Search Windows for “environment variables.”
2. Open Edit environment variables for your account.
3. Create OLLAMA_MODELS with the desired absolute path.
4. Apply the change and restart Ollama.

Install Ollama on macOS

1. Download the current macOS package from the official site.
2. Open the disk image and place Ollama in Applications.
3. Launch the application and follow any prompt needed to make the ollama command available.
4. Open a new Terminal window and run:

ollama --version
ollama run MODEL_NAME

Apple silicon and Intel systems have different performance characteristics and may not support the same practical model sizes. Select a model based on the machine's usable memory rather than assuming that every downloadable model will run well.

Install Ollama on Linux

The current official installation page provides an install script and downloadable packages for supported architectures. Review any script before running it with elevated privileges, especially if it has been copied from another site.

A commonly documented installation command is:

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

Then verify the installation:

ollama --version

If the installer configured a systemd service, inspect its status:

systemctl status ollama

Otherwise, start the server in a terminal:

ollama serve

In another terminal, run a model:

ollama run MODEL_NAME

Do not combine unrelated commands onto one line. Distribution packaging, GPU drivers, service paths, and supported accelerators change; follow the current Ollama and hardware-vendor instructions for your exact distribution and GPU.

Manage the Linux service

sudo systemctl start ollama
sudo systemctl enable ollama
sudo systemctl status ollama

View recent service logs with:

journalctl -e -u ollama

To add supported environment settings without replacing the packaged service file:

sudo systemctl edit ollama

After saving an override, reload systemd and restart the service:

sudo systemctl daemon-reload
sudo systemctl restart ollama

Use the local Ollama API

Ollama commonly serves an HTTP API on the local machine. Confirm the address in your current installation rather than exposing it to the network unnecessarily. A PowerShell request can be structured as follows:

$body = @{
  model  = "MODEL_NAME"
  prompt = "Explain why the sky appears blue."
  stream = $false
} | ConvertTo-Json

Invoke-RestMethod -Method Post -Uri "http://localhost:11434/api/generate" -ContentType "application/json" -Body $body

Replace the model name with one already installed. Application code should set timeouts, handle non-success responses, limit prompt size, and avoid logging sensitive content.

Useful model-management commands

ollama list
ollama ps
ollama show MODEL_NAME
ollama stop MODEL_NAME
ollama rm MODEL_NAME

ollama list shows downloaded models, while ollama ps shows models currently loaded. The remove command deletes the named local model, so verify the exact name before using it.

Security and privacy checklist

• Keep the API local unless necessary. Binding the service to a network interface can allow other devices to submit prompts or consume resources.
• Add protection outside Ollama. If remote access is required, use an authenticated, access-controlled proxy and firewall rules appropriate to the environment.
• Trust model sources deliberately. Review the model card, license, publisher, and any custom templates or adapters.
• Protect imported data. Local processing does not remove the need for filesystem permissions, encryption, backups, and retention rules.
• Review connected applications. A local model can still call a tool or application that sends information elsewhere.
• Update carefully. Read release notes, test important workflows, and keep a rollback plan for production use.

Troubleshooting

Symptom	Checks
Command not found	Open a new terminal, confirm installation, and check whether the CLI location is in PATH
Model is very slow	Check model size, available memory, context length, accelerator support, and other running workloads
Download fails	Check disk space, network access, proxy settings, and the exact model name
Local API does not respond	Confirm the Ollama process or service is running and inspect its logs
Models use the wrong drive	Verify OLLAMA_MODELS and restart every Ollama process after changing it

Ollama makes local model experimentation more accessible, but the hardware, model license, and surrounding application still determine whether a setup is private, reliable, and suitable for production.