A modern, open-source intelligent customer service system designed for teams implementing both website chat widgets and agent workbenches. AI-CS offers a hybrid AI-human approach, private deployment, and robust observability, aiming to streamline customer interactions and support operations. It is suitable for businesses seeking a configurable and self-hostable solution for real-time chat, file transfer, and knowledge base Q&A.

How It Works

The system employs a dual-architecture: a Go backend handles APIs, WebSockets, AI processing, and Retrieval-Augmented Generation (RAG), while a Next.js frontend manages the public website, visitor chat widget (embeddable via iframe), and the agent workbench. Core functionalities include AI-driven responses via OpenAI-compatible APIs, RAG for knowledge base integration (supporting PDF/DOCX imports and Milvus vector stores), real-time messaging synchronized across instances using Redis, and optional web search capabilities. This modular design allows for private deployment and deep customization.

Quick Start & Requirements

The recommended installation method is using pre-built Docker images via docker-compose.prod.yml up -d. Essential configuration involves creating a .env file from .env.example and setting database credentials, an admin password, and a 64-character hex encryption key. For local development or manual installation, prerequisites include Go 1.24+, Node.js 20.9.0+, and MySQL 8.0+. Online demos are available at demo.cscorp.top for the official website, demo.cscorp.top/chat for the visitor chat, and demo.cscorp.top/agent/login for the agent login.

Highlighted Details

• Hybrid AI/Human Support: Seamlessly integrates AI chatbots with human agents, allowing for flexible interaction models.
• Knowledge Base (RAG): Supports document import (PDF/DOCX), chunking, and vectorization with Milvus. Prioritizes FAQ matches before vector retrieval.
• LLM Agnostic: Integrates with any service offering an OpenAI-compatible Chat Completions API.
• Web Search Integration: Optional support for web search via Serper or function calling, enabling AI to access real-time information.
• Real-time Communication: Utilizes WebSockets for instant messaging and includes features like shared draft input and visitor location tracking.
• Observability: Provides structured logging and data reports on visitor activity, AI performance, and knowledge base usage.

Maintenance & Community

Community interaction is primarily facilitated through GitHub Issues for bug reports and feature suggestions. A QQ group (1106804464) is also available for discussion. Contributions via Pull Requests are welcomed, with guidelines provided for reporting issues, including necessary logs and sanitized configuration details.

Licensing & Compatibility

The project is released under the MIT license, which permits broad use, modification, and distribution, including for commercial purposes and integration into closed-source applications.

Limitations & Caveats

Production deployments require an external Nginx or Caddy server to proxy /api requests to the backend and forward specific custom headers. Pre-built Docker images may have historical port binding dependencies, making local builds preferable for full port customization. Admin password management is primarily handled during initial setup; subsequent changes necessitate direct database manipulation. Enabling RAG without disabling Milvus (MILVUS_REQUIRED=true) will cause startup failure if the vector store is unavailable. Browser audio playback requires explicit user interaction.