Agentic Framework Comparison Repository

This repository provides a comprehensive comparison of different agentic AI orchestration frameworks through working implementations. Each branch contains the same core application built with a different framework, allowing you to explore and compare approaches to building AI agents.

📋 Table of Contents - Framework Implementations

Framework	Branch	Description	Key Features
OpenAI Direct	📁 openai	Direct OpenAI API integration	Simple, minimal setup with raw OpenAI calls
CrewAI	📁 crewai	Multi-agent collaboration framework	Agent crews, role-based workflows, task delegation
LangGraph	📁 langgraph	Graph-based agent workflows	State management, conditional routing, complex workflows
Pydantic AI	📁 pydantic	Type-safe AI agent framework	Built-in validation, structured outputs, type safety
LiteLLM	📁 litellm	Multi-provider LLM gateway	Unified API for 100+ LLM providers, easy provider switching
DSPy	📁 dspy	Declarative language model programming	Signatures, modules, prompt optimization, composability
Groq	📁 groq	High-performance inference API	Lightning-fast responses, open-source models, OpenAI compatibility

🚀 What This Repository Demonstrates

Each implementation provides the same core functionality:

• Interactive chat interface with a web-based demo
• Phoenix observability for tracing and monitoring agent behavior
• Docker containerization for consistent deployment
• REST API for programmatic agent interaction
• Conversation memory and context management

🏗️ Common Infrastructure

All implementations share the same foundational components:

Core Features

• FastAPI server for HTTP endpoints
• Flask demo interface for interactive testing
• Phoenix integration for comprehensive observability and tracing
• Docker containerization with Python 3.12 and uv package management
• LRU caching for conversation state management
• Pydantic schemas for request/response validation

Observability & Monitoring

• Phoenix dashboard at localhost:6006 for trace visualization
• OpenInference instrumentation specific to each framework
• Request/response tracing with conversation context
• Performance metrics and error tracking

Development Tools

• Automatic environment setup with ./bin/bootstrap.sh
• Hot reload for development iterations
• Comprehensive logging for debugging
• Standardized project structure across all implementations

🔧 Quick Start (Any Branch)

1. Choose your framework - Switch to the branch you want to explore
2. Set up environment - Run ./bin/bootstrap.sh (installs Python 3.12 + uv automatically)
3. Configure API keys - Create .env file with your OpenAI API key
4. Launch the stack - Run ./bin/run_agent.sh --build
5. Explore the demo - Visit localhost:8080 for the chat interface
6. Monitor with Phoenix - Visit localhost:6006 for observability

📊 Framework Comparison

Complexity vs. Capability

Framework	Setup Complexity	Learning Curve	Capability	Best For
OpenAI Direct	Low	Low	Basic	Simple chatbots, prototyping
CrewAI	Medium	Medium	High	Multi-agent workflows, team collaboration
LangGraph	High	High	Very High	Complex state machines, conditional logic
Pydantic AI	Low	Low	Medium	Type-safe applications, structured data
LiteLLM	Low	Low	Medium	Multi-provider applications, vendor flexibility
DSPy	Medium	Medium	High	Declarative prompting, optimization, research
Groq	Low	Low	Medium	High-speed inference, open models, performance-critical apps

Key Differences

OpenAI Direct

• Minimal abstraction over OpenAI API
• Direct control over all parameters
• Simplest to understand and debug

CrewAI

• Multi-agent orchestration
• Role-based agent definitions
• Built-in task delegation and collaboration

LangGraph

• Graph-based workflow definition
• Advanced state management
• Conditional routing and complex logic flows

Pydantic AI

• Type-safe agent interactions
• Built-in validation and structured outputs
• Clean, pythonic API design

LiteLLM

• Unified API for 100+ LLM providers
• Easy provider switching without code changes
• Built-in cost tracking and fallback mechanisms

DSPy

• Declarative approach to LM programming
• Automatic prompt optimization
• Modular components (ChainOfThought, ReAct, etc.)
• Research-focused with composability

Groq

• Ultra-fast inference with custom silicon
• OpenAI-compatible API for easy integration
• Access to open-source models (Llama, Mixtral)
• Cost-effective high-performance inference

🛠️ Switching Between Implementations

Each branch is fully self-contained. To explore a different framework:

# Switch to desired framework branch
git checkout <framework-branch>

# Set up and activate the environment (if you want to make changes)
./bin/bootstrap.sh && source .venv-{framework}/bin/activate.sh

# Launch the application
./bin/run_agent.sh --build

📁 Project Structure

All implementations follow this consistent structure:

├── agent/
│   ├── agent.py          # Core agent implementation (framework-specific)
│   ├── server.py         # FastAPI server with observability
│   ├── prompts.py        # Prompt templates and formatting
│   ├── schema.py         # Pydantic models for validation
│   ├── caching.py        # Conversation state management
│   └── demo_code/        # Flask demo interface
├── bin/
│   ├── bootstrap.sh      # Environment setup script
│   └── run_agent.sh      # Docker launch script
├── Dockerfile            # Python 3.12 + uv container
├── docker-compose.yml    # Multi-service orchestration
├── requirements.txt      # Framework-specific dependencies
└── README.md            # Framework-specific documentation

🔍 Observability Features

All implementations include comprehensive observability through Phoenix:

• Trace Visualization - See complete request flows
• Performance Monitoring - Track response times and errors
• Conversation Context - View full conversation history
• Framework-Specific Metrics - Understand framework internals
• Real-time Dashboards - Monitor live agent interactions

🎯 Learning Path Recommendations

1. Start with OpenAI Direct - Understand the basics without framework abstractions
2. Explore Pydantic AI - Learn type-safe agent development
3. Try Groq - Experience ultra-fast inference with open-source models
4. Try LiteLLM - Experience multi-provider flexibility and cost optimization
5. Experiment with DSPy - Try declarative prompting and optimization
6. Experience CrewAI - Build multi-agent orchestration systems
7. Master LangGraph - Create complex, stateful agent workflows

🤝 Contributing

Each branch maintains the same application interface while showcasing different framework approaches. When contributing:

• Keep the core API consistent across implementations
• Update framework-specific documentation in each branch
• Ensure observability features work across all frameworks
• Maintain the same development experience (bootstrap, run scripts, etc.)

📖 Further Reading

• Phoenix Observability: Phoenix Documentation
• OpenInference Standards: OpenInference Specification
• Framework Documentation: See individual branch READMEs for framework-specific guides

---

Choose a branch above to start exploring different approaches to building AI agents! Each implementation provides the same functionality with different architectural patterns and capabilities.