Installing and Running Ollama as a Non-Root User on Ubuntu 25.04

This comprehensive guide provides detailed instructions for installing and running Ollama, an open-source tool for running large language models locally, on Ubuntu 25.04 as a non-root user. The goal is to enable individual users to set up and manage their own isolated Ollama instances, ensuring that each user’s configuration, models, and runtime environment are independent and stored within their home directory. This approach avoids the need for root privileges and prevents conflicts between multiple users on the same system.

Overview

Ollama is a lightweight, extensible platform for running large language models (LLMs) locally, such as LLaMA, Mistral, or Gemma. Unlike system-wide installations that require root access, this guide focuses on a per-user installation, allowing each user on an Ubuntu 25.04 system to run their own instance of Ollama. This setup ensures:

Isolation: Each user’s models and configurations are stored in their home directory (~/.ollama).
No Root Privileges: The installation avoids sudo or system-wide changes.
Flexibility: Users can manage their own models, update Ollama independently, and customize settings like ports.

This guide is tailored for Ubuntu 25.04, a hypothetical future release, assuming compatibility with the latest Ollama binaries and standard Linux tools available in Ubuntu’s ecosystem as of July 2025.

Prerequisites

Before starting, ensure the following:

Operating System: Ubuntu 25.04 (64-bit, AMD64 or ARM64 architecture).
Disk Space: At least 10GB of free space in the user’s home directory for models (e.g., the llama3 8B model requires ~4.7GB).
RAM: Minimum 8GB, with 16GB or more recommended for larger models.
Internet Access: Required to download the Ollama binary and models.
Tools: curl for downloading the binary and systemd (optional) for running Ollama as a user service.
Permissions: The user must have write access to their home directory (~).

To install curl, run:

sudo apt update
sudo apt install -y curl

Step-by-Step Installation

Step 1: Log in as the Target User

Each user must perform the installation in their own account to ensure isolation. Log in to the user’s account via a terminal or SSH:

whoami  # Verify the current user

All subsequent steps are executed as this non-root user.

Step 2: Download the Ollama Binary

Ollama provides a precompiled binary for Linux, which can be downloaded and used without root access.

Create a directory in the user’s home directory to store the Ollama binary: mkdir -p ~/ollama cd ~/ollama
Download the latest Ollama binary for Linux (AMD64). Adjust the URL for ARM64 if needed: curl -L https://ollama.com/download/ollama-linux-amd64 -o ollama
- Check the Ollama GitHub releases page for the latest binary or alternative architectures (e.g., ollama-linux-arm64 for ARM systems).

Step 3: Make the Binary Executable

Grant execute permissions to the downloaded binary:

chmod +x ~/ollama/ollama

Step 4: Set Up Environment Variables

To make the ollama command accessible from any directory, add the binary’s location to the user’s PATH.

Edit the user’s shell configuration file (e.g., ~/.bashrc for Bash or ~/.zshrc for Zsh): nano ~/.bashrc
Append the following line to add ~/ollama to the PATH: export PATH="$HOME/ollama:$PATH"
Apply the changes: source ~/.bashrc
Verify the binary is accessible: ollama --version This should display the installed Ollama version.

Step 5: Set Up Ollama Data Directory

Ollama stores models and configuration in the ~/.ollama directory by default, which is perfect for per-user setups. Create this directory if it doesn’t exist:

mkdir -p ~/.ollama

No further configuration is needed, as Ollama automatically uses this directory for model storage and logs.

Step 6: Run Ollama Manually

Start the Ollama server in the background to handle model requests:

ollama serve &

The & runs the server in the background. To keep it running after closing the terminal, consider using tmux or screen:tmux new -s ollama ollama serve Detach from tmux with Ctrl+B, then D. Reattach later with tmux attach -t ollama.

Step 7: Pull and Run a Model

To test Ollama, download and run a model like llama3:

Pull the model: ollama pull llama3
- Models are stored in ~/.ollama/models, isolated to the user.
- The llama3 8B model requires approximately 4.7GB of disk space.
Run the model interactively: ollama run llama3 This starts a command-line interface where you can interact with the model. Type queries and press Enter to receive responses.

Step 8: Optional – Run Ollama as a User-Level Systemd Service

For persistent operation without keeping a terminal open, set up a user-level systemd service.

Create a systemd service file for the user: mkdir -p ~/.config/systemd/user nano ~/.config/systemd/user/ollama.service
Add the following configuration: [Unit] Description=Ollama Service for %u After=network.target [Service] ExecStart=%h/ollama/ollama serve Restart=always Type=simple [Install] WantedBy=default.target
Enable and start the service: systemctl --user enable ollama systemctl --user start ollama
Check the service status: systemctl --user status ollama The service will start automatically when the user logs in and restart if it crashes.

Step 9: Verify the Installation

Confirm that Ollama is installed and running correctly:

Check the version: ollama --version
Verify the server is running: curl http://localhost:11434 A response like Ollama is running indicates success.
Test model interaction: ollama run llama3 "Hello, world!" This should return a response from the llama3 model.

Additional Configuration

Managing Port Conflicts

Ollama’s default port is 11434. If multiple users run Ollama on the same machine, port conflicts may occur. To assign a unique port for a user:

Set the OLLAMA_HOST environment variable: export OLLAMA_HOST=127.0.0.1:11435
Add it to ~/.bashrc for persistence: echo 'export OLLAMA_HOST=127.0.0.1:11435' >> ~/.bashrc source ~/.bashrc
Restart the Ollama server: ollama serve &

Each user should choose a unique port (e.g., 11435, 11436, etc.).

Updating Ollama

To update Ollama to a newer version:

Stop the running Ollama server or service: systemctl --user stop ollama # If using systemd pkill ollama # If running manually
Download the new binary: cd ~/ollama curl -L https://ollama.com/download/ollama-linux-amd64 -o ollama chmod +x ollama
Restart the server or service: systemctl --user start ollama # If using systemd ollama serve & # If running manually

Storage Considerations

Model Size: Models vary in size (e.g., llama3 8B is ~~4.7GB, while larger models like llama3 70B require ~38GB). Ensure sufficient disk space in `~~/.ollama/models`.
Multiple Models: Users can pull multiple models, but this increases storage needs.
Backup: To save space or migrate, back up ~/.ollama/models and restore it later.

Troubleshooting

Port Conflict: If curl http://localhost:11434 fails with a connection error, check for port conflicts: netstat -tuln | grep 11434 Assign a new port as described in Managing Port Conflicts.
Permission Issues: Ensure the ollama binary and ~/.ollama are owned by the user: chown -R $USER:$USER ~/ollama ~/.ollama
Service Fails to Start: Check logs for user-level systemd services: journalctl --user -u ollama
Model Download Issues: Verify internet connectivity and sufficient disk space: df -h ~
Binary Incompatibility: Ensure the correct binary (AMD64 or ARM64) is downloaded for your system’s architecture: uname -m

For additional help, consult the Ollama GitHub issues page or search for community discussions on platforms like X.

References

This guide was last updated on July 29, 2025, based on the latest available information for Ubuntu 25.04 and Ollama.

etcinitd.web.id