AI Toolkit for Local Use

TL;DR.

This post is a guide to setting up a local AI toolkit on a Debian-based Linux distribution, specifically Ubuntu. It covers the installation and management of various AI tools and frameworks, including:

Ollama,
CrewAI,
Crawl4AI,
AI Extensions for VS Code,
OpenRouter.ai,
Cursor,
Windsurf,
Open WebUI,
LM Studio,
Pinokio, and
Ngrok.

This guide includes step-by-step instructions for installing, updating, and uninstalling these tools, as well as promoting virtual environments (specifically Conda) for best practices. Also, this post occasionally offers insights into these tools and how they are utilised during AI development.

Attributions:

various ↗.

An Introduction.

This post highlights the tools I find useful as an AI enthusiast. I want to share my insights into these tools and explain why they are beneficial to AI developers:

The purpose of this post is to describe the tools that I, as an AI enthusiast, find useful.

The Big Picture.

November 30th 2022 saw the launch of AI’s first killer app: ChatGPT (Generative Pre-trained Transformer). Three months later, in February 2023, someone “leaked“ the first, open LLM (Large Language Model) which was built by Meta, the corporation that owns Facebook. Since then, many open models have dropped and Hugging Face has become the de facto centre for these AI machines.

NOTE: From here on, I will refer to LLMs as models.

It is now early 2025 and, over the last two and-a-bit years, there has been an explosion of tools and workflows hitting the Internet. The continued growth of commercial, frontier models (e.g. OpenAI models such as GPT-4o, GPT-4o mini, and the GPT-4.1 series) has easily matched, and usually surpassed, the power of my favourite open models (DeepSeek-R1, Phi4, Qwen3, etc.) However, thanks to the rise of MoE (Mixture of Experts) models, reasoning models, agents, agentic tasks and tools, RAG (Retrieval-Augmented Generation) processes, and AI-specific scraping tools, AI enthusiasts can easily develop vibe coding skills on local PCs using open models, frontier models, or a combination of both types of systems.

Prerequisites.

A Debian-based Linux distribution (I use Ubuntu),
Python 3.11+.

Updating my Base System.

From the (base) terminal, I update my (base) system:

sudo apt clean && \
sudo apt update && \
sudo apt dist-upgrade -y && \
sudo apt --fix-broken install && \
sudo apt autoclean && \
sudo apt autoremove -y

What is Ollama?

Ollama is a local, large language model manager that facilitates the installation, management, and operation of AI models on personal computers.

Installing Ollama.

From the terminal, I install Ollama:

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

I run Ollama as a background service:

ollama serve &

NOTE: An error typically displays (see below) because Ollama, by default, already runs as a background service. Also, by default, Ollama runs on port 11434.

Updating Ollama.

I update Ollama:

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

NOTE: The update command is the same as the install command.

Pulling Models from Ollama.

I pull the following models from Ollama (it will take time for these downloads to complete):

ollama pull nomic-embed-text:v1.5 &&
ollama pull gemma3:12b &&
ollama pull deepcoder:14b &&
ollama pull codellama:13b &&
ollama pull granite3.2-vision:2b &&
ollama pull granite3.3:8b &&
ollama pull llama3.1:8b &&
ollama pull llama3.2-vision:11b &&
ollama pull llama3.2:3b &&
ollama pull qwen2.5-coder:14b &&
ollama pull qwen3:14b &&
ollama pull phi4:14b &&
ollama pull phi4-reasoning:14b
ollama pull deepseek-coder-v2:16b &&
ollama pull deepseek-r1:14b

NOTE: Theses models are perfect for running on my GTX3060 GPU with 12GB VRAM. Also, this list is continually updated. If a newer version of a model is released, I remove the old listing above, replace it with the new listing, run the pull commands, list the installed models (ollama ls), and remove (ollama rm <model-name>) the older model.

I list the models downloaded by Ollama:

ollama list

Uninstalling Ollama.

I stop the Ollama service:

sudo systemctl stop ollama

I disable the Ollama service:

sudo systemctl disable ollama

I remove the Ollama service:

sudo rm /etc/systemd/system/ollama.service

I remove the Ollama binary from my bin directory:

sudo rm $(which ollama)

I remove the models downloaded by Ollama:

sudo rm -r /usr/share/ollama

I delete the Ollama service user:

sudo userdel ollama

I delete the Ollama service group:

sudo groupdel ollama

What is CrewAI?

CrewAI is a framework that is used to build AI crews, agents, tasks, and tools.

Installing CrewAI.

NOTE: Best practice involves installing this app in a virtual environment, using either venv or Conda.

From the terminal, I use the pip command to install CrewAI and its’ tools:

pip install crewai 'crewai[tools]'

What is Crawl4AI?

Crawl4AI is a web-scraping tool that can conform website data into a format that can be understood by AI models, or the data can be prepared for uploading to a vector database (pending).

Installing Crawl4AI.

NOTE: Best practice involves installing this app in a virtual environment, using either venv or Conda.

I use the pip command to install the updated version of Crawl4AI:

pip install -U crawl4ai

I setup Crawl4AI:

crawl4ai-setup

I verify the Crawl4AI installation:

crawl4ai-doctor

What are AI Extensions for VS Code?

VS Code is a FREE, open-source IDE (integrated development environment) from Microsoft. Extensions are add-on that are installed within VS Code. These extensions provide extra functionality without having to make changes to the VS Code source code. AI extensions for VS Code are tools and integrations that connect to local, AI models

Some Popular AI Extensions for VS Code.

Some of my favourite AI Extensions for VS Code include:

Twinny,
Roo Code,
Cline, and
Continue.

What is OpenRouter.ai?

OpenRouter is a platform that provides a unified API for accessing various models from multiple providers.

Open Models vs. OpenRouter.

I prefer downloading open models to my local system, and using then using these models on my personal computer. That way, I can ensure my data does not leak, and my constructs are not used to train frontier models. Sometimes, however, I need to run complex inference. In those cases, I access better models with more parameters that run on powerful hardware. That is why I occasionally switch to using OpenRouter.

Choosing OpenRouter over open models depends on my requirements at each stage of a project.

What is Cursor?

Cursor is a VS Code fork that provides an improved AI development experience.

NOTE: The following process can be use on any AppImage app.

Installing Cursor.

ATTRIBUTION: https://forum.cursor.com/t/tutorial-install-cursor-permanently-when-appimage-install-didnt-work-on-linux/7712

From a browser, I download the AppImage file from the cursor.com website.
From the file manager, I move the Cursor app to the NAS (network attached storage) server.
From the terminal, I create the Apps/Cursor directory:

mkdir ~/Apps/Cursor

I copy the Cursor download to the the Apps/Cursor directory:

cp ~/Downloads/Ubuntu/Cursor/Cursor*.AppImage ~/Apps/Cursor/Cursor.AppImage

NOTE: This copy command changes the download name to Cursor.AppImage.

OPTIONAL:
OPTIONAL: I copy the 128px by 128px logo to the ~/Apps/Cursor directory:

cp /media/brian/Drawings/logos/cursor/Images/Cursor-icon.png ~/Apps/Cursor/Cursor-icon.png

NOTE: I downloaded this image from the Internet.

I make the AppImage executable:

chmod +x ~/Apps/Cursor/Cursor.AppImage

I use the Nano text editor to create a desktop entry:

nano ~/.local/share/applications/cursor.desktop

I paste (CTRL + SHIFT + V) the following into the desktop entry:

[Desktop Entry]
Name=Cursor
Exec=~/Apps/Cursor/Cursor.AppImage
Icon=~/Apps/Cursor/Cursor-icon.png
Type=Application
Categories=Utility;Development;

I save (CTRL + S) the changes, and exit (CTRL + X) the Nano text editor.
I create a symlink to start Cursor from the terminal:

sudo ln -s ~/Apps/Cursor/Cursor.AppImage /usr/local/bin/cursor

I restart my system.
From the terminal, I run the Cursor IDE:

cursor

What is Windsurf?

Windsurf is an AI-powered code editor designed to enhance the coding experience with features like automatic context analysis, AI-driven autocompletion, and an intuitive user interface.

Installing Windsurf.

I add the Windsurf repo to my local system:

curl -fsSL "https://windsurf-stable.codeiumdata.com/wVxQEIWkwPUEAGf3/windsurf.gpg" | sudo gpg --dearmor -o /usr/share/keyrings/windsurf-stable-archive-keyring.gpg
echo "deb [signed-by=/usr/share/keyrings/windsurf-stable-archive-keyring.gpg arch=amd64] https://windsurf-stable.codeiumdata.com/wVxQEIWkwPUEAGf3/apt stable main" | sudo tee /etc/apt/sources.list.d/windsurf.list > /dev/null

I update my system:

sudo apt update

I upgrade Windsurf:

sudo apt upgrade windsurf

What is Open WebUI?

Open WebUI is a browser-based interface for using AI models.

Installing Open WebUI.

From the terminal, I use Docker to pull Open WebUI:

sudo docker pull ghcr.io/open-webui/open-webui:main

Now I can run Open WebUI on port 3000 from within the Docker container:

docker run -d -p 3000:8080 -v open-webui:/app/backend/data --name open-webui ghcr.io/open-webui/open-webui:main

Alternatively, I can start Open WebUI from Docker Desktop.

What is LM Studio?

LM Studio is an AI application for locally running AI models.

The following process can be use on any AppImage app.

Installing LM Studio.

I use a browser to download the AppImage file from the https://lmstudio.ai/download website.
From the file manager, I move the LMStudio app to the NAS (network attached storage) server.
From the terminal, I create the Apps/LMStudio directory:

mkdir ~/Apps/LMStudio

I copy the LMStudio download to the the Apps/LMStudio directory:

cp /media/brian/Downloads/Ubuntu/'LM Studio'/LM*.AppImage ~/Apps/LMStudio/LMStudio.AppImage

NOTE: This copy command changes the download name to LMStudio.AppImage.

I copy the 128px by 128px logo from the NAS to the ~/Apps/LMStudio directory:

cp /media/brian/Drawings/logos/lmstudio/Images/LMStudio-icon.png ~/Apps/LMStudio/LMStudio-icon.png

NOTE: I downloaded this image from the Internet.

I make the AppImage executable:

chmod +x ~/Apps/LMStudio/LMStudio.AppImage

I use the Nano text editor to create a desktop entry:

nano ~/.local/share/applications/lmstudio.desktop

I paste (CTRL + SHIFT + V) the following into the desktop entry:

[Desktop Entry]
Name=Cursor
Exec=~/Apps/LMStudio/LMStudio.AppImage
Icon=~/Apps/LMStudio/LMStudio-icon.png
Type=Application
Categories=Utility;Development;

I save (CTRL + S) the changes, and exit (CTRL + X) the Nano text editor.
I create a symbolic link (-s) to start Cursor from the terminal:

sudo ln -s ~/Apps/LMStudio/LMStudio.AppImage /usr/local/bin/lmstudio

I run LMStudio:

lmstudio

What is Pinokio?

Pinokio is a browser that runs applications.

Installing Pinokio.

I use a browser to visit the download page for Pinokio:

https://github.com/pinokiocomputer/pinokio/releases

I download the latest version of Pinokio that runs on Debian (.deb) distributions for AMD64 processors.
From the terminal, I install Pinokio:

sudo apt install -y ~/Downloads/Pinokio*

I run Pinokio from the Apps Drawer.

What is Ngrok?

Ngrok creates secure tunnels to localhost, allowing developers to expose local servers to the Internet.

Installing Ngrok.

In a browser, I visit the https://ngrok.com/ website and create an account.
I copy the authentication token.
From a new terminal tab, I install Ngrok:

sudo snap install ngrok

I add the authentication token:

ngrok config add-authtoken <auth_token>

I point Ngrok to any local port number that is running a server:

ngrok http <local_port_number>

The Results.

Setting up a local AI toolkit significantly improves my chances of improving my AI development skills. By following this guide, I efficiently installed, can manage, and will utilise various AI tools and frameworks. Combining Ollama, CrewAI, Pinokio, Open WebUI, LM Studio, Ngrok, and Cursor provides a robust environment for developing and experimenting with open AI models. This well-equipped, local toolkit will help me develop my vibe coding skills.

In Conclusion.

In this post, I described how to set up a powerful local AI toolkit on Ubuntu with step-by-step instructions. I learned how to install and manage AI tools like Ollama, CrewAI, and more, enhancing my AI development capabilities. This article is perfect for AI enthusiasts and developers looking to harness both open and commercial AI technologies.

Until next time: Be safe, be kind, be awesome.

Hash Tags.

#AI #Artificial-Intelligence #AI-Development #AI-Models #AI-Community #Machine-Learning #Deep-Learning #Linux #Ubuntu #Open-Source #Coding #Tech-Guide #Tech-Tools #Future-Tech #Innovation

Local AI Toolkit for Ubuntu.