Exploring Spring AI Alibaba OpenManus: Multi‑Agent Architecture and Real‑World Demo

The official release of Spring AI Alibaba OpenManus provides a complete multi‑agent task planning, reasoning, and execution flow for Java developers, enabling complex operations such as browser control, code execution, and data processing.

Demo Overview

Search for Alibaba's latest weekly stock price, plot the trend, and save the chart locally.

Generate a detailed five‑day travel itinerary from Hangzhou to Seoul for a 10,000‑yuan budget, outputting an HTML travel guide with maps, descriptions, basic Korean phrases, and tips.

Translate a set of Chinese documents located in /tmp/docs and store the translated files in /tmp/endocs.

Overall Architecture and Principles

Spring AI Alibaba OpenManus shares the design philosophy of the Python OpenManus project. Its architecture consists of a Planning Agent that decomposes user queries into sequential steps, a chain of Manus Agents that execute each step using a ReAct pattern with tool calls, and a Summary Agent that aggregates the final result.

Planning Agent : Responsible for task decomposition and planning, generating a serial Manus Agent sub‑workflow via the planning tool.

Manus Agents : Form a chain where each agent corresponds to a step from the planning phase, employing a ReAct architecture to complete sub‑tasks through multiple tool invocations.

Summary Agent : Produces the final summary of the overall task.

Current Implementation Issues

Approximately 80% of the repository code deals with workflow orchestration (linking Manus agents, message memory, tool forwarding, global state changes), which could be abstracted into a higher‑level agent framework to reduce complexity.

Tool coverage and execution effectiveness are limited, especially for browser automation and script execution.

The planning and workflow lack mechanisms for manual review, dynamic modification, or rollback.

Debugging the current OpenManus implementation is relatively difficult.

Future Plans and Solutions

Spring AI Alibaba aims to provide a Java‑centric open‑source AI application framework tightly integrated with the Spring ecosystem, allowing developers to build new AI applications or enhance existing Spring Boot projects with intelligence.

Beyond the core framework abstractions, the roadmap includes a multi‑agent framework, visual evaluation platform, and debugging Studio. An upcoming Spring AI Alibaba Graph multi‑agent framework and an enhanced OpenManus implementation are expected to reduce code size by over 70%, improve readability, and boost overall performance.

The project already supports MCP tool integration, and future work will incorporate a mature MCP server to further enhance OpenManus capabilities.