🤖 AI Agents 101

Build AI Agents from Scratch with Python

No frameworks. No LangChain. Just Python and the Claude API.

Read the series on Medium →

---

Companion code for my AI Agents 101 blog series. Each folder is a self-contained project that maps to one blog post — clone the repo, pick a folder, and run it.

📚 The Series

	Post	Folder	What You'll Build
01	What Is an AI Agent, Really?	01-basic-agent/	A minimal agent with the agentic loop pattern
02	ReAct in 50 Lines of Python	02-react-agent/	An agent that thinks before it acts
03	Real Tools, Real APIs	03-real-tools/	Web search, calculator & file reader
04	Give Your Agent Memory	04-agent-memory/	Persistent memory across sessions
05	Agents vs. RAG Pipelines	05-agents-vs-rag/	Side-by-side RAG pipeline vs agent
06	Multi-Agent Systems	06-multi-agent/	Researcher → Writer → Editor pipeline

🚀 Quick Start

git clone https://github.com/RittikaJ/ai-agent-tutorial.git
cd ai-agent-tutorial
pip install -r requirements.txt

cp .env.template .env   # paste your Anthropic API key

# run any post:
python 01-basic-agent/agent.py
python 02-react-agent/react_agent.py

🗂 Repo Structure

ai-agent-tutorial/
│
├── 01-basic-agent/
│   └── agent.py                  # Agentic loop — weather + time tools
│
├── 02-react-agent/
│   └── react_agent.py            # Think → Act → Observe → Repeat
│
├── 03-real-tools/
│   └── agent_with_tools.py       # DuckDuckGo, safe eval calculator, file reader
│
├── 04-agent-memory/
│   └── memory_agent.py           # Sliding window + JSON persistent store
│
├── 05-agents-vs-rag/
│   ├── rag_pipeline.py           # Retrieve → Augment → Generate
│   └── agent_approach.py         # Same knowledge base, but with actions
│
├── 06-multi-agent/
│   └── multi_agent.py            # 3-agent pipeline with revision loop
│
├── .env.template
├── requirements.txt
└── README.md

Every script is standalone — no shared imports, no hidden dependencies between folders. Read top to bottom and it makes sense.

🧠 How the Series Builds

Post 1: The agentic loop          → LLM + tools + while loop = agent
Post 2: ReAct reasoning           → Think before you act (debuggable agents)
Post 3: Real tools                → Web search, math, file I/O
Post 4: Memory                    → Remember across sessions
Post 5: Agents vs RAG             → Know when to use which
Post 6: Multi-agent               → Specialized agents working together

By the end you've built from scratch everything that frameworks like LangChain and CrewAI abstract away.

📋 Requirements

• Python 3.8+
• An Anthropic API key (free tier works)

📝 License

MIT — use the code however you want.

---

Built by Rittika · Follow along on Medium for new posts