evi-ai 0.1.0

A flexible multi-agent framework code generator supporting multiple LLM providers and frameworks

Evi - Multi-Agent Framework Generator

Evi is a flexible, modular code generator for multi-agent systems that supports multiple agent frameworks (CrewAI, LangGraph, ReAct) and LLM providers (OpenAI, Gemini).

Features

• Multiple Agent Frameworks

  • CrewAI - Create agent crews with specialized roles and tasks
  • LangGraph - Create agent workflows using directed graphs
  • ReAct - Build agents using the Reasoning+Acting pattern

• Multiple LLM Providers

  • OpenAI (GPT-3.5, GPT-4)
  • Google Gemini (Pro, Pro Vision)

• Dual Interfaces

  • Command-line interface (CLI) for scripting and automation
  • Streamlit web interface with visualization and interactive controls

• Modular, Pluggable Architecture

  • Easily extend with new providers or frameworks
  • Customizable templates using Jinja2

• Advanced Features

  • Multiple output formats (Python code or JSON configuration)
  • Agent workflow visualization
  • Configurable model parameters

Installation

Basic Installation

pip install evi

With GUI Support

pip install evi[streamlit]

With Framework-Specific Dependencies

pip install evi[crewai]  # For CrewAI support
pip install evi[langgraph]  # For LangGraph support
pip install evi[react]  # For ReAct support

Full Installation

pip install evi[all]

For Development

pip install evi[dev]

Usage

Command-Line Interface

# Generate CrewAI code using OpenAI
evi --prompt "Create a team of researchers to analyze climate data" --framework crewai --provider openai

# Generate LangGraph configuration using Gemini
evi --prompt "Create a customer service workflow" --framework langgraph --provider gemini --output json

# Save output to a file
evi --prompt "Create a team of researchers to analyze climate data" --output-file my_agents.py

Streamlit Interface

# Launch the Streamlit web interface
evi-gui

Python API

from evi.core import EviGenerator

# Create a generator instance
generator = EviGenerator(provider="openai", framework="crewai")

# Generate code from a prompt
code = generator.generate(
    "Create a research team with three agents that analyze financial data",
    output_format="code"
)

# Generate JSON configuration
config = generator.generate(
    "Create a research team with three agents that analyze financial data",
    output_format="json"
)

# Print or save the generated code
print(code)
with open("my_agents.py", "w") as f:
    f.write(code)

API Key Configuration

Evi requires API keys for accessing LLM providers:

Environment Variables

# For OpenAI
export OPENAI_API_KEY=your_api_key_here

# For Gemini
export GEMINI_API_KEY=your_api_key_here

Using .env File

Create a .env file in your project directory:

OPENAI_API_KEY=your_api_key_here
GEMINI_API_KEY=your_api_key_here

Then in your Python code:

from dotenv import load_dotenv

load_dotenv()  # This loads the .env file

Framework-Specific Examples

CrewAI Example

from evi.core import EviGenerator

prompt = """
Create a research crew with two agents:
1. A data collector that finds information online
2. An analyst that summarizes and provides insights
The crew should analyze recent developments in quantum computing.
"""

generator = EviGenerator(provider="openai", framework="crewai")
code = generator.generate(prompt, output_format="code")

# Save the generated code
with open("quantum_research_crew.py", "w") as f:
    f.write(code)

LangGraph Example

from evi.core import EviGenerator

prompt = """
Create a customer support workflow that:
1. Classifies the initial customer query
2. Routes to either technical support or billing department
3. Generates a response based on the department's knowledge base
"""

generator = EviGenerator(provider="openai", framework="langgraph")
code = generator.generate(prompt, output_format="code")

# Save the generated code
with open("customer_support_workflow.py", "w") as f:
    f.write(code)

ReAct Example

from evi.core import EviGenerator

prompt = """
Create a reasoning agent that can:
1. Answer questions about physics
2. Break down complex problems into steps
3. Generate analogies to explain difficult concepts
"""

generator = EviGenerator(provider="gemini", framework="react")
code = generator.generate(prompt, output_format="code")

# Save the generated code
with open("physics_tutor_agent.py", "w") as f:
    f.write(code)

Architecture

Evi is built with a modular, pluggable architecture:

• Core: Central controller that orchestrates the generation process
• Providers: Interfaces to LLM providers (OpenAI, Gemini)
• Frameworks: Handlers for different agent frameworks (CrewAI, LangGraph, ReAct)
• Templates: Jinja2 templates for code generation

This architecture makes it easy to extend Evi with new providers or frameworks by implementing the appropriate interfaces.

Contributing

We welcome contributions! Please see CONTRIBUTING.md for details on how to contribute to Evi.

License

This project is licensed under the MIT License - see the LICENSE file for details.