AgentFlow

AgentFlow is a flexible and extensible framework for building and managing AI agents and workflows. It provides a modular architecture for creating, configuring, and orchestrating AI agents with advanced features for workflow management and testing.

Latest Version

Current version: v0.1.1

• Fixed workflow transform functions to handle step and context parameters
• Added feature engineering and outlier removal transforms
• Improved test suite and type hints
• Enhanced error handling and validation

Project Structure

agentflow/
├── agents/             # Agent implementations
│   ├── agent.py       # Base agent implementation
│   └── agent_types.py # Agent type definitions
├── core/              # Core framework components
│   ├── base_types.py  # Base type definitions
│   ├── config.py      # Configuration management
│   ├── workflow.py    # Workflow engine
│   └── workflow_executor.py # Workflow execution
├── ell2a/             # ELL2A integration
│   └── types/         # Message and content types
├── transformations/   # Data transformation tools
│   └── text.py       # Text processing utilities
└── tests/            # Comprehensive test suite
    ├── unit/         # Unit tests
    ├── core/         # Core component tests
    └── performance/  # Performance tests

Features

• Flexible Agent Architecture

  • Configurable agent types and behaviors
  • Support for research and data science agents
  • Extensible agent factory system

• Advanced Workflow Management

  • Step-based workflow execution
  • Dependency management
  • Error handling and retry policies
  • Performance monitoring

• Robust Testing Framework

  • Unit and integration tests
  • Performance testing
  • Test-driven development support

• Data Transformation Tools

  • Feature engineering
  • Outlier removal
  • Text processing utilities

• Type Safety

  • Comprehensive type hints
  • Pydantic model validation
  • Runtime type checking

Installation

pip install agentflow

Quick Start

from agentflow import Agent, AgentConfig, WorkflowConfig, WorkflowStep, WorkflowStepType

# Create workflow configuration
workflow_config = WorkflowConfig(
    id="test-workflow",
    name="test_workflow",
    steps=[
        WorkflowStep(
            id="step-1",
            name="transform_step",
            type=WorkflowStepType.TRANSFORM,
            description="Data transformation step",
            config={
                "strategy": "standard",
                "params": {
                    "method": "standard",
                    "with_mean": True,
                    "with_std": True
                }
            }
        )
    ]
)

# Create agent configuration
agent_config = AgentConfig(
    name="test_agent",
    type="data_science",
    workflow=workflow_config
)

# Create and initialize agent
agent = Agent(config=agent_config)
await agent.initialize()

# Process data
result = await agent.execute({"data": your_data})

Configuration

Agents and workflows can be configured using Python dictionaries or YAML files:

AGENT:
  name: data_science_agent
  type: data_science
  version: 1.0.0
  mode: sequential

MODEL:
  provider: openai
  name: gpt-4
  temperature: 0.7
  max_tokens: 4096

WORKFLOW:
  id: transform-workflow
  name: Data Transformation Workflow
  max_iterations: 5
  timeout: 30
  steps:
    - id: step-1
      name: feature_engineering
      type: transform
      description: Feature engineering step
      config:
        strategy: standard
        params:
          method: standard
          with_mean: true
          with_std: true

Testing

Run the test suite:

# Run all tests
pytest

# Run specific test categories
pytest tests/unit/
pytest tests/core/
pytest tests/performance/

Contributing

1. Fork the repository
2. Create a feature branch
3. Write tests for your changes
4. Implement your changes
5. Run the test suite
6. Create a Pull Request

License

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