EntraFlow: Enterprise Agent Orchestration System

EntraFlow is a powerful, production-ready AI agent orchestration system designed to simulate and automate complex enterprise workflows. It leverages modular agent architecture, intelligent DAG-based orchestration, and seamless integration with Large Language Models (LLMs) to provide deep business insights and automated decision-making.

---

🚀 Overview

EntraFlow coordinates a suite of specialized agents through a central Orchestrator. Each agent is responsible for a specific domain—from data ingestion and machine learning analysis to strategic decision-making and quality assurance.

The Agent Ecosystem

• 📡 DataFetcher: Ingests real-time data from external APIs (OpenWeather, NewsAPI). Features robust caching and mock data fallbacks for development.
• 🔍 Analyzer: Conducts multi-layer analysis using Transformer models (DistilBERT) for sentiment analysis and LangChain-powered LLMs for deep semantic insights.
• ⚖️ DecisionMaker: Applies complex business logic and priority rules. It uses LLMs to provide Strategic AI Advisory, translating data analysis into actionable corporate strategy.
• 📢 Notifier: The reporting engine. Generates stakeholder-ready reports in HTML, JSON, and Markdown/Text formats, including executive summaries and alert logs.
• 🛡️ Supervisor: The QA layer. Validates agent outputs against schemas, resolves data conflicts, and ensures final reports meet enterprise quality standards.

---

🧠 Key Features

• Intelligent Orchestration: Uses a Directed Acyclic Graph (DAG) for dependency management and topological sorting.
• LLM-Agnostic Core: Native support for OpenAI, HuggingFace, and Ollama via LangChain integration.
• Fault-Tolerant Engine: Automatic retries with exponential backoff, state persistence for workflow resumption, and graceful degradation.
• Extensible Configuration: Entire workflows, agent thresholds, and API settings are managed via a centralized config.yaml.
• Comprehensive Logging: Multi-tier logging with console colorization and persistent storage for audit trails.
• Enterprise-Grade Analysis: Combines classical NLP (keywords) with modern Deep Learning (Transformers) and GenAI (LLMs).

---

🛠️ Installation

1. Clone & Setup

git clone https://github.com/hashimkalam/EntraFlow.git
cd EntraFlow

2. Environment Configuration

Create a .env file or set environment variables:

# API Keys (Optional - fallbacks provided)
export OPENWEATHER_API_KEY="your_key"
export NEWS_API_KEY="your_key"
export OPENAI_API_KEY="your_key"  # If using OpenAI provider

3. Install Dependencies

pip install -r requirements.txt

Note: On first run, the system will automatically download the DistilBERT model (~260MB) for local analysis if not present.

---

🔑 Configuration

EntraFlow is highly configurable via config.yaml. Key sections include:

• api: Set your preferred LLM provider (huggingface, openai, ollama) and model name.
• agents: Tune confidence thresholds, priority rules, and output formats.
• workflows: Define custom sequences of agent execution.

---

💻 Usage

EntraFlow provides a powerful CLI for managing your agent workforce.

Run a Workflow

Execute the full enterprise analysis pipeline:

python main.py run --workflow enterprise_analysis

Manage Workflows

# List all defined workflows
python main.py list-workflows

# Check status of previous runs
python main.py status

Debugging & Testing

# Test a specific agent in isolation
python main.py test --agent Analyzer

---

📂 Project Structure

EntraFlow/
├── agents/             # Modular agent implementations
│   ├── base_agent.py   # Base class with core lifecycle logic
│   ├── analyzer.py     # ML & LLM analysis logic
│   └── ...             # Other specialized agents
├── orchestrator/       # The "Brain" of the system
│   ├── orchestrator.py # DAG execution and scheduling
│   └── workflow.py     # Workflow definition parsing
├── utils/              # Shared utilities
│   ├── llm.py          # LangChain & LLM provider management
│   └── logger.py       # Enterprise logging system
├── cli/                # Terminal interface logic
├── data/               # Persistent states and generated reports
└── config.yaml         # Centralized system settings

---

📝 Output & Reporting

EntraFlow doesn't just process data; it communicates results. Check data/outputs/ after a run for:

• HTML Reports: Polished, visual summaries for executives.
• JSON Data: Raw, structured output for downstream integration.
• Alert Logs: Immediate notification of critical business triggers.

---

📄 License

Distributed under the MIT License. See LICENSE for more information.