🦀 agentic-cookbook

A barebones implementation of agentic architectures dedicated to demystifying how recursive agent loops work. Inspired by Claude Code and Open Claw.

🎯 Objectives

• Educational: Understand tool-calling, memory, and feedback loops without heavy abstractions.
• Prototyping: Minimalist playground for testing agentic concepts.
• Documentation: A curated guide to modern agentic architectures.

🏗️ Core Architecture

• Agent Loop: Sequential ReAct-style loop.
• Tool Registry: Dynamic schema generation from Python functions.
• Model Abstraction: Multi-provider support via LiteLLM.
• Memory: Simple list-based message history.

🚀 Getting Started

Prerequisites

• Python 3.10+
• OpenAI or Anthropic API Key

Installation

Option 1: Using uv (Recommended)

uv is an extremely fast Python package manager.

# Install dependencies
uv pip install -e .

# Or run directly if using uv project management (if applicable)
# uv run some_example.py

Setup

1. Copy .env.example to .env:

   cp .env.example .env

2. Add your API keys to the .env file.

📚 Documentation

Check out Architectures Guide for a deep dive into ReAct, Plan-and-Execute, and Reflexion patterns.

🛠️ Usage

Check the examples/ directory for working implementation examples.

python examples/01_react_basic.py