This repository provides a hands-on implementation of four core agentic patterns—Reflection, Tool Use, Planning (ReAct), and Multi-Agent—from scratch. It targets developers and researchers seeking a deep understanding of agentic AI without relying on higher-level frameworks like LangChain or CrewAI. The project's benefit lies in its pure API call approach to Groq, offering clarity on fundamental agent mechanics.

How It Works

The project implements each pattern using direct API calls to Groq's LLM. The Reflection pattern enhances LLM output by allowing self-critique and refinement. The Tool Use pattern integrates external functions (e.g., API calls, calculations) to augment LLM knowledge. The Planning pattern, exemplified by ReAct, enables LLMs to break down complex tasks into sequential steps, using tools and reasoning. The Multi-Agent pattern orchestrates multiple agents with defined roles and dependencies, similar to CrewAI but implemented from the ground up.

Quick Start & Requirements

• Install via pip: pip install -U agentic-patterns or using Poetry: poetry install.
• Requires a Groq API Key, configured via a .env file.
• Official notebooks and YouTube videos are available for guided learning.

Highlighted Details

• Implements four distinct agentic patterns from scratch.
• Focuses on pure Groq API calls, avoiding intermediate frameworks.
• Demonstrates ReAct for planning and a custom Crew abstraction for multi-agent workflows.
• Includes examples for custom tool creation and integration.

Maintenance & Community

The project appears to be an educational initiative by neural-maze, focused on teaching agentic patterns. Community engagement channels are not explicitly mentioned in the README.

Licensing & Compatibility

The README does not specify a license. Compatibility for commercial use or closed-source linking is not detailed.

Limitations & Caveats

This project is presented as an educational tool rather than a production-ready framework. It relies exclusively on Groq as the LLM provider, limiting flexibility. The multi-agent implementation is a custom abstraction and may not cover all features of established frameworks.