Quick replica of Manus in three hours, OpenManus open source project won 3,000+ stars

The emergence of Manus indirectly triggered a competition for reproduction in the open source world. In addition to the CAMEL-AI 0-day reproduction Manus universal agent OWL shared by AIbase, a project called OpenManus also realized Manus reproduction with just three hours of code .

Just last night, OpenManus, which can be used without any invitation, was born on GitHub. As of press time, it has won 3.3,000 GitHub star tags, which can be said to be "the peak of its debut". The most eye-catching thing is the concise and powerful declaration: " Just modify config.toml to play ."

The installation process of OpenManus can be described as silky. With just a few simple lines of command, you can have your own AI agent: first create a new conda environment, activate it, then clone the project, install the dependencies, and finally copy and edit the configuration file.

Intelligent collaboration, interesting with "brain"

This project is not alone, but is jointly created by the core contributors of MetaGPT in just three hours. This makes people sigh that the big guys are indeed "the time to drink a cup of coffee can change the direction of the tide."

At the heart of OpenManus is a revolutionary modular Agent system that consists of multiple highly intelligent "professional teams". You can think of it as an efficient "AI company", including a "project manager" (Manus main agent) who understands user needs, a "strategic expert" (Planning Agent) who breaks up complex tasks into clear steps, and a "technical expert" (ToolCallAgent) who is proficient in using various tools. Developers can freely combine these functional modules like building blocks according to their needs to create an exclusive AI assistant.

What is even more surprising is that OpenManus is not a "closed-door car", but seamlessly integrates multiple top-level models including Claude3.5 and Qwen VL Plus , giving full play to the advantages of each model. This open mind is worthy of praise.

Task processing ability like "Ku Ding Cuts a Ox"

When handling complex tasks, OpenManus shows the fineness of "cutting a cow". For example, when you give it a task of "in-depth research on Karpathy's website and providing SEO optimization reports", it will first think about it and break down the task into clear steps such as checking website information, analyzing SEO elements, checking technical problems, and sorting out optimization suggestions. The entire thinking process is completely visualized and every step is presented to you.

Ultimately, it generates a detailed report, even the details of Meta tag optimization, XML sitemap creation, etc. are taken into account.

Project: https://github.com/mannaandpoem/OpenManus