ZhuLing
A zero-code AI agent development scaffold for Java built on Spring Boot and Spring AI, letting a developer launch a working agent from a single YAML configuration file with MCP tool calling and built-in observability.
ZhuLing lets a Java developer configure and run an AI agent by writing one YAML file rather than code, getting multiple independently registered agents, a unified interface for choosing between an agent or a workflow as the conversation target, synchronous and SSE streaming chat endpoints, multi-turn session memory persisted to a database, and MCP tool calling over local, SSE or stdio transports. Skill script packages can be loaded and executed, and lifecycle interceptors provide extension points for observation, error normalization and security policy.
Observability includes token statistics, tool-call telemetry, trace IDs and a measure of context-window usage, exposed through a versioned envelope of turn, message, reasoning and tool lifecycle events, alongside a built-in observable conversation workbench UI. A single-agent ReAct execution core supports multi-step thought, act and observe cycles with process-level observation and a real-time step trace.
It is built on JDK 21, Spring Boot 4.1, Spring AI 2.0 and MyBatis-Plus with a MySQL or PostgreSQL backend, and is released under the Apache-2.0 license.