Virtual Environments Guide

Why Virtual Environments?

Every AI project needs different package versions. Project A might need TensorFlow 2.10, while Project B needs TensorFlow 2.15. Without virtual environments, these conflict and break everything. Virtual environments create isolated spaces where each project has its own dependencies.

The Problem Without Virtual Environments

❌ Global Python (BAD):

All projects share the same packages
Upgrading one package breaks other projects
Can't test different versions
Dependency hell is inevitable

✅ Virtual Environments (GOOD):

Each project has its own isolated packages
Upgrade freely without breaking other projects
Test different versions side-by-side
Clean, reproducible setups

Option 1: venv (Built-in, Lightweight)

venv comes with Python 3.3+. It's simple, fast, and perfect for most AI projects.

Create a Virtual Environment

cd ~/projects/my-ai-project
python3 -m venv venv

This creates a venv/ folder in your project directory with isolated Python and packages.

Activate the Environment

source venv/bin/activate

You'll see (venv) in your terminal prompt. Now all pip install commands install into this environment only.

Install Packages

pip install langchain openai transformers

These packages are now isolated to this project only.

Deactivate When Done

deactivate

Returns you to the global Python environment.

Option 2: Conda (For Data Science & ML)

Conda is a more powerful environment manager. It handles non-Python dependencies (like CUDA for GPU) and is popular in data science.

Install Miniconda (Lightweight Conda)

brew install --cask miniconda

Create a Conda Environment

conda create -n my-ai-env python=3.11
conda activate my-ai-env

Install Packages with Conda

conda install numpy pandas scikit-learn
pip install langchain openai  # Can use pip too!

Deactivate

conda deactivate

venv vs conda: Which to Use?

Feature	venv	conda
Built-in	✅ Yes (Python 3.3+)	❌ Install separately
Speed	✅ Fast, lightweight	⚠️ Slower, heavier
Non-Python packages	❌ Python only	✅ Yes (CUDA, R, etc.)
Best for	General AI projects	Data science, ML with GPU

Recommendation: Start with venv. Switch to conda if you need GPU support or specific data science packages.

Best Practices

1. One Environment Per Project

Never share environments between projects. Create a new one for each.

2. Use requirements.txt

Save your dependencies so you (or others) can recreate the environment:

pip freeze > requirements.txt  # Save dependencies
pip install -r requirements.txt  # Restore later

3. Add venv to .gitignore

Never commit virtual environments to Git—they're large and platform-specific:

echo "venv/" >> .gitignore
echo ".env/" >> .gitignore
echo "__pycache__/" >> .gitignore

4. Name Environments Clearly

Use descriptive names: langchain-chatbot, image-gen-sd, etc.

Quick Start Example

Here's a complete workflow for a new AI project:

# Create project folder
mkdir ~/projects/ai-chatbot
cd ~/projects/ai-chatbot

# Create virtual environment
python3 -m venv venv

# Activate it
source venv/bin/activate

# Install AI packages
pip install langchain openai python-dotenv

# Create requirements file
pip freeze > requirements.txt

# Work on your project...

# When done, deactivate
deactivate