What is apboa-next?

Apboa Next is a Chinese language platform for building and running AI agents, meant for companies rather than individual hobby projects. It uses the ReAct approach, where an agent reasons about a task and then takes actions in a loop, and it bundles together the pieces a team would otherwise assemble separately: filtering out sensitive words, managing prompts, connecting to multiple AI model providers such as OpenAI, DashScope, Anthropic, Gemini, and Ollama, integrating external tools, building automated workflows, and connecting a knowledge base for the agent to search. The backend is split into five separate services, a console for administration, a runtime that does the actual AI reasoning, a proxy for running shell commands safely, a file service for syncing skill files, and a websocket service for real time messages. This split lets the runtime service scale out to handle more load while the others stay as single instances. The frontend adds multi tenant support, meaning multiple separate customer organizations can use the same deployment with their own permissions and data kept apart. Users can run multiple chat sessions in parallel, and if a page refreshes mid conversation, the session state and any in progress streaming response can pick back up rather than being lost. A newer document recognition feature lets the agent read uploaded documents. A visual workflow builder lets non programmers assemble business processes by dragging and connecting over 30 kinds of building blocks, covering things like conditional branches, loops, database operations, calling AI agents or other tools, and formatting data, without writing code. There is also a scheduling feature for running agents or workflows automatically on a timer, similar to a cron job, with support for manual triggers and a log of past runs. To try it, you need Docker Engine and Docker Compose installed, then clone the repository and run a startup script, which spins up all five services plus supporting databases in a few minutes with a default admin login. Setting it up for real local development instead requires Java 21, Maven, MySQL, Redis, and Node.js. The project is released under the MIT license, which allows free use, modification, and commercial use as long as the copyright notice is kept.