This project implements a sophisticated multi-agent customer support system leveraging Retrieval-Augmented Generation (RAG) and LangGraph. It addresses complex travel-related queries and integrates with WooCommerce for e-commerce functions, offering a robust, modular, and secure conversational AI solution for businesses. The system targets developers and organizations seeking to enhance their customer support capabilities with advanced AI.

How It Works

The system employs a multi-agent architecture orchestrated by LangGraph as a stateful graph. A primary assistant routes user queries to specialized agents (e.g., flights, hotels, WooCommerce). Each agent can access tools, with sensitive operations requiring user confirmation. Novel security guardrails—including jailbreak and relevance checks—are applied to user input before processing. Sensitive actions are further secured via a GoHumanLoop-integrated human review workflow, ensuring enhanced safety and control.

Quick Start & Requirements

• Primary Install/Run:
  • Install dependencies: poetry install
  • Start Qdrant (optional): docker compose up qdrant -d
  • Generate embeddings: poetry run python vectorizer/app/main.py
  • Run CLI chat: poetry run python ./customer_support_chat/app/main.py
  • Run Web UI: poetry run uvicorn web_app.app.main:app --reload --host 0.0.0.0 --port 8000
• Prerequisites: Python 3.12+, Poetry, Docker & Docker Compose, OpenAI API Key, LangSmith API Key (optional), GoHumanLoop API Key (optional), Qdrant URL (optional), WooCommerce API credentials.
• Links: Original project: https://github.com/ro-anderson/multi-agent-rag-customer-support

Highlighted Details

• Multi-Agent Design: Specialized assistants handle distinct domains (travel, e-commerce) for modularity and expertise.
• RAG Integration: Utilizes Qdrant vector database for retrieving relevant information to ground AI responses.
• Advanced Security: Implements pre-processing "security guardrails" for jailbreak and relevance checks.
• Human-in-the-Loop: Integrates GoHumanLoop for administrator approval of sensitive operations, adding a critical safety layer.
• E-commerce Functionality: Direct integration with WooCommerce for product queries, order status, and form submissions.
• Observability: Leverages LangSmith for detailed tracing of agent interactions, tool usage, and system performance.

Maintenance & Community

Information regarding active maintenance, notable contributors, or community channels (e.g., Discord, Slack) is not explicitly detailed in the provided documentation.

Licensing & Compatibility

The specific open-source license for this project is not stated in the provided README. This absence requires further investigation for commercial use or integration into closed-source applications.

Limitations & Caveats

The project is presented as an evolving system with several areas identified for future enhancement, including adaptive RAG for tool retrieval, corrective RAG for web searches, self-RAG for answer validation, and improved memory integration using a dedicated cache database. Currently, user state and conversation history are managed via separate JSON files, indicating a less robust memory solution.