Setting Up Your Local Agentic AI Environment

In this tutorial series, examples default to Python so the material can stay focused and go deeper.

We strongly recommend running everything locally with Ollama.
Cloud APIs (even free tiers like Gemini) quickly hit rate limits when you experiment, debug, or run many iterations. Ollama gives you unlimited local inference, zero cost, and full privacy — ideal for learning agentic AI by trial and error.

Note: Local Ollama will be significantly slower than cloud APIs, especially on older hardware. If you need more speed, you can switch to a cloud API (like Gemini) at any time — just be mindful of rate limits and costs.

1. Programming Language Setup

We use Python as the default language throughout the tutorials.

Official install links:

Python 3.12+

2. Install Ollama (Local LLM Runtime)

Ollama lets you run powerful local models (like Qwen 3.5) directly on your laptop.

macOS (Recommended: Homebrew)

# Install Ollama via Homebrew (easiest way on macOS)
brew install ollama

# Start the Ollama service
brew services start ollama

Windows / Linux / Other macOS methods

Visit the official download page → https://ollama.com/download

Best multi-platform video (Mac + Windows + Linux) – “Install Ollama on Mac, Windows & Linux (Step-by-Step)” (Dec 2025)

Additional excellent videos:

Installing Ollama is EASY Everywhere (macOS, Windows, Linux)
How to Install Ollama on macOS (Apple Silicon) – great for M-series Macs
Ollama on Windows in 5 minutes

3. Pull the Tested Local Baseline

# Pull the tested local baseline used by this series
ollama pull qwen3.5:9b

This series uses qwen3.5:9b as a tested local baseline: small enough for practical local experimentation, strong enough for agent and tool examples, and stable enough that the code remains reproducible across the curriculum.

It is not meant to be a permanent claim about the single best local model. Local model rankings change quickly, so treat this as the model the examples are tested against.

Need	Suggested setup	Why
Reproduce the tutorials	`qwen3.5:9b` with Ollama	Tested baseline for the series
Lower-memory machine	Smaller Qwen/Gemma-class local model	Useful when the 9B model is too slow
Stronger reasoning or multimodal work	Hosted model or current multimodal local model	Better speed, vision, and long-context behavior

This model runs well on many recent laptops (8GB+ RAM recommended), but performance depends heavily on CPU/GPU, memory bandwidth, quantization, and background load.

4. Quick Test – Is Ollama Working?

Run the interactive chat:

ollama run qwen3.5:9b

Then type:

Explain how a ReAct agent works in one paragraph.

You should get a coherent answer instantly.

5. Verifying Your Setup (Ollama + Python)

Install the Python packages used by the examples:

python3 -m pip install ollama pydantic

Python

from ollama import chat
from pydantic import BaseModel

response = chat(
    model="qwen3.5:9b",
    messages=[{"role": "user", "content": "Say hello from Ollama!"}]
)

print(response.message.content)

When to Use What

Use Case	Recommended Setup	Reason
Learning / experimentation	Ollama (local)	Unlimited trials, no cost
Production / high-scale	OpenAI / Gemini	Higher speed & scale
Offline / privacy-first	Ollama	Runs 100% locally

Optional: Cloud Model Providers

This series uses Ollama by default. If you prefer a hosted model, you can use any provider that supports chat/completion APIs and tool calling.

Gemini: https://aistudio.google.com/
OpenAI: https://platform.openai.com/
Anthropic: https://console.anthropic.com/

Cloud providers are useful for speed and stronger frontier models, but they introduce API keys, costs, rate limits, and data-sharing considerations.

You’re now fully set up!

→ Next: What is an Agent?