ThorXMarvin/customer-support-agent

AI-powered customer support agent for WhatsApp, Telegram & web. Configurable knowledge base, multi-language, escalation to humans. Built for East African businesses. # Customer Support Agent — Concept Note ## Overview An AI-powered customer support agent that plugs into WhatsApp, Telegram, or web chat to handle customer inquiries 24/7. Businesses configure it with their FAQs, products, and policies — the agent handles first-line support and escalates complex issues to humans. ## Problem East African SMEs lose customers because they can't respond fast enough on WhatsApp/Telegram. Hiring 24/7 support staff is expensive. Customers expect instant replies. ## Solution A configurable AI agent that: - Answers FAQs from a knowledge base (products, services, pricing, hours, location) - Handles order status inquiries - Collects customer information and creates support tickets - Escalates to human agents when it can't help - Supports multiple languages (English, Luganda, Swahili) - Works on WhatsApp, Telegram, and web embed ## Architecture ``` ┌─────────────────────────────────────────────┐ │ Customer Support Agent │ ├─────────────────────────────────────────────┤ │ Channels │ Core Engine │ │ ├─ WhatsApp │ ├─ LLM Provider (any) │ │ ├─ Telegram │ ├─ Knowledge Base (RAG) │ │ └─ Web Widget │ ├─ Conversation Memory │ │ │ ├─ Escalation Logic │ │ Config │ └─ Analytics/Logging │ │ ├─ config.yaml │ │ │ ├─ knowledge/ │ Integrations │ │ └─ templates/ │ ├─ Email notifications │ │ │ └─ Webhook callbacks │ └─────────────────────────────────────────────┘ ``` ## Tech Stack - **Runtime:** Node.js - **LLM:** Any provider (OpenAI, Anthropic, Google, local models via Ollama) - **Knowledge Base:** Markdown/JSON files + vector embeddings (ChromaDB or in-memory) - **Channels:** Baileys (WhatsApp), Telegraf (Telegram), Express (Web widget) - **Database:** SQLite (conversations, tickets, analytics) - **Config:** YAML-based configuration ## Configuration (Developer Flow) ```bash …