JC0v0/Customer-Agent

What it solves

Customer-Agent is a desktop application designed for e-commerce merchants to automate customer service. It replaces manual responses with an AI-driven agent that can handle product inquiries, policy questions, and sales recommendations across multiple platforms (currently supporting Pinduoduo).

How it works

The application uses a custom-built Agent framework that integrates with various Large Language Models (LLMs) via LiteLLM. It manages conversations through a multi-turn tool-calling system and session context management. To provide accurate answers, it employs a dual-knowledge base system: one for specific product details and another for general customer service policies (shipping, returns, etc.). It can automatically sync product data from platform APIs and use multimodal LLMs to extract product knowledge for the database.

Who it’s for

E-commerce sellers and customer service representatives who want to automate their store's responses while maintaining the ability to transfer complex issues to human agents.

Highlights

• Multi-channel Integration: Supports real-time message reception via WebSocket (currently Pinduoduo).
• AI-Driven Sales: The agent can proactively recommend products and send product cards to users.
• Dual Knowledge Base: Separate retrieval systems for product-specific info and general store policies.
• Automated Knowledge Sync: Automatically pulls product lists from APIs and extracts knowledge using multimodal LLMs.
• ** keluarga Toolset**: Includes tools for fetching product lists, sending links, querying product knowledge, and transferring to human agents.
• Robust Infrastructure: Features an asynchronous message queue, handler chains, and automatic WebSocket reconnection mechanisms.