Tgo

Tgo is an open source AI customer service platform built around teams of cooperating agents. Instead of a single chatbot, an operator assembles several agents with different roles and lets them work together on customer inquiries, after sales support and lead generation, with a handoff to human agents when a conversation needs one. The project documentation is published in Chinese and English. The platform includes a built in retrieval augmented knowledge base that accepts PDF and Word uploads and crawled web pages, support for the Model Context Protocol so agents can call external APIs, databases and business tools, and connections to major language model providers. Customers can be reached through a website widget, an API and messaging channels including WeChat, Lark and DingTalk. A React based management console lets support staff run the service. Tgo is aimed at businesses that want to host their own customer service stack. It is deployed with Docker Compose on Linux, macOS or WSL2, and the documentation lists a minimum of 2 CPU cores and 8 GiB of memory, with 4 cores and 16 GiB recommended. The architecture is split into services for the core API, agent orchestration, retrieval, platform integrations, the web console, the visitor widget and a real time messaging layer. The source code is on GitHub. The vendor does not publish paid plan prices.