Policy-Aware RAG Multi-Agent AI System

---

🚀 Quick Start

Option 1: Docker (Recommended)

git clone https://github.com/pooyaphoenix/RA3G-Agent.git
cd RA3G-Agent
docker compose up --build

Option 2: Local Installation

git clone https://github.com/pooyaphoenix/RA3G-Agent.git
cd RA3G-Agent
python3 -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt
python ra3g.py --api-port 8010 --ui-port 8501

Access the application:

• 🌐 Web UI: http://localhost:8501
• 📡 API Docs: http://localhost:8010/docs
• 🔍 Health Check: http://localhost:8010/health

---

✨ Features

• 🔍 Local RAG System - Query your documents without external APIs
• 🛡️ Governance Agent - Automatically filters sensitive information
• 🤖 Multi-Agent Architecture - Retriever, Reasoning, and Governance agents
• 💾 Session Memory - Remembers previous queries and context
• 📊 Real-time Logs - Monitor agent activity with live log streaming
• 📄 PDF Document Upload - Automatic PDF upload and vector store building
• 🎨 Streamlit UI - Beautiful web interface for easy interaction
• 🔌 REST API - Full programmatic access via FastAPI
• ⚙️ Fully Customizable - Easy configuration via config.yml

---

🎬 Demo

---

🏗️ Architecture

Agent Flow:

1. Retriever Agent - Finds relevant passages from your corpus
2. Reasoning Agent - Generates answers using Ollama LLM
3. Governance Agent - Validates and filters responses based on policies

---

📖 Usage Examples

Web UI

1. Open http://localhost:8501
2. Navigate to the Chat tab
3. Type your question and get instant answers

API Request

curl -X POST 'http://localhost:8010/query' \
  -H 'Content-Type: application/json' \
  -H 'session-id: my-session' \
  -d '{
    "query": "What are the benefits of regular hand washing?",
    "top_k": 5
  }'

API Response

{
  "query": "What are the benefits of regular hand washing?",
  "answer": "Preventing the spread of infections is one of the simplest ways.",
  "governance": {
    "approved": true,
    "reason": "approved"
  },
  "trace": [
    {
      "index": 0,
      "note": "relevant passage about benefits of hand washing"
    }
  ],
  "retrieved": [
    {
      "id": "corpus_medical_general.txt#p0",
      "text": "...",
      "source": "corpus_medical_general.txt",
      "score": 0.51
    }
  ],
  "confidence": 0.512,
  "session_id": "my-session"
}

Python Client Example

import requests

response = requests.post(
    'http://localhost:8010/query',
    headers={'session-id': 'my-session'},
    json={'query': 'What is machine learning?', 'top_k': 5}
)
print(response.json()['answer'])

---

🔌 API Endpoints

Method	Endpoint	Description
POST	/query	Ask a question through RAG
GET	/health	Health check for all agents
GET	/health/{agent}	Health check for specific agent
GET	/trace	Get session query history
DELETE	/memory/clear	Clear session memory
GET	/docs	Interactive Swagger UI
GET	/logs/stream/{log_type}	Stream logs in real-time (SSE)

Try it live: http://localhost:8010/docs

---

🛡️ Governance Rules

The governance agent automatically blocks or approves queries based on:

❌ Blocked:

• Personal names, phone numbers, addresses
• Medical records and patient data
• Confidential corporate information
• Banned phrases (configurable in config.yml)

✅ Allowed:

• General medical, scientific questions
• Educational content
• Public information

---

📂 Adding Documents

1. Place your documents in data/corpus/ directory (.txt or .md files)
2. Automatic indexing - The system builds the FAISS index on startup
3. Manual indexing (optional):

   python indexer.py --corpus data/corpus

Configuration: Edit config.yml to customize corpus directory and indexing behavior.

---

⚙️ Configuration

All settings are in config.yml:

# Ollama Configuration
OLLAMA_URL: http://localhost:11434/api/generate
OLLAMA_MODEL: qwen2.5:7b-instruct

# Embedding Model
EMBED_MODEL: all-MiniLM-L6-v2

# Confidence Thresholds
THRESHOLDS:
  retriever: 0.72
  reasoner: 0.81

# Auto-build index on startup
AUTO_BUILD_FAISS: true
CORPUS_DIR: data/corpus

# Banned phrases for governance
BANNED_PHRASES:
  - diagnosis
  - prescription
  - classified
  - confidential

---

🧪 Testing

# Health check
curl http://localhost:8010/health

# Test query
curl -X POST http://localhost:8010/query \
  -H 'Content-Type: application/json' \
  -H 'session-id: test' \
  -d '{"query": "Hello", "top_k": 3}'

---

📝 Requirements

• Python 3.8+
• Ollama (for LLM inference)
• Docker (optional, for containerized deployment)

See requirements.txt for Python dependencies.

---

🤝 Contributing

See CONTRIBUTING.md for guidelines.

---

📧 Contact

pooyachavoshi@gmail.com

---

📄 License

See LICENSE file for details.