Spring AI Alibaba releases open source Manus implementation for Java

Written by

Silas Grey

Updated on:July-09th-2025

The Spring AI Alibaba OpenManus officially released this time includes a complete multi-agent task planning, thinking and execution process, allowing developers to experience the multi-agent effect of the Java version. It can analyze user questions, operate browsers, execute codes, etc. to complete complex tasks.

Original project source code and experience address: spring-ai-alibaba-openmanus

01 Effect display

Without further ado, let's take a look at the running results. The following is the actual usage effect of Spring AI Alibaba OpenManus demonstrated through several actual question and answer records.

1. Open Baidu browser and enter Alibaba's stock price in the most recent week in the search box. Draw a stock price trend chart for the most recent week based on the searched information and save it to a local directory.

2. I plan to travel to South Korea during the upcoming Labor Day holiday. The itinerary is from Hangzhou to Seoul, South Korea, with a total budget of 10,000 yuan. I want to experience the customs, culture, and life of ordinary people in South Korea. The total itinerary is planned to be 5 days. Please provide a detailed itinerary and make it into a simple HTML travel brochure containing maps, attraction descriptions, basic Korean phrases and travel tips for my reference throughout the trip.

3. There are some Chinese documents in the /tmp/docs directory of this machine. Please translate these documents into Chinese one by one and save them into a separate file. Store the newly generated files in the /tmp/endocs directory.

02

Overall architecture and principles

The design concept of Spring AI Alibaba Openmanus is similar to that of the Python version OpenManus. Its overall architecture is shown in the following figure.

Analyzing the architecture in the figure above, we can regard it as a multi-agent intelligent automatic collaboration implementation, in which:

Planning Agent is responsible for task decomposition and planning, breaking down user problems into several sequentially executable steps. Planning Agent calls planning tool to dynamically generate a serial Manus Agent sub-workflow.
Multiple Manus Agents form a chained, sequentially executable sub-workflow. Each agent in the sub-workflow corresponds to a step in the above plan, and each agent is a ReAct architecture design, that is, it completes a specific sub-task through multiple rounds of Tool calls.
Summary Agent is used to make the final task summary.

03

Implementation Summary and Outlook

Issues in Spring AI Alibaba OpenManus implementation

The current OpenManus implementation has the following main problems:

80% of the code in the warehouse is used to solve process orchestration problems, such as connecting manus agent sub-processes, memorizing messages, forwarding tool calls, modifying global status, etc. This part of the work can be implemented by a highly abstract agent framework to simplify development complexity.
The coverage and execution effect of the tools are general, such as browser usage, script execution tools, etc.
There is no human intervention in the planning and workflow to conduct review, dynamic modification, rollback, etc.
The effects debugging currently implemented by OpenManus is relatively difficult.

Spring AI Alibaba Future Planning and Solutions

Spring AI Alibaba is an open source AI application development framework for Java developers. It is perfectly compatible with the Spring ecosystem. You can build new AI applications based on Spring AI Alibaba, or use it to upgrade traditional Spring Boot applications to intelligent ones.

From the above figure, we can see that in addition to the framework atomic abstraction, Spring AI Alibaba focuses on planning a multi-agent framework and supporting ecosystems such as a visual evaluation platform and debugging studio.

Next, we will release the Spring AI Alibaba Graph multi-agent framework and an enhanced version of OpenManus based on Spring AI Alibaba Graph. The amount of code is expected to be reduced by more than 70% compared to the current level, and the overall readability and effect will be greatly improved, allowing developers to build intelligent applications for any scenario based on this.

Currently, Spring AI Alibaba already supports MCP tool access. In the future, we will connect a more mature MCP server implementation to OpenManus to improve overall work performance.