Domestically-made intelligent agent MasterAgent: Create a dedicated AI team in one sentence for professional collaborative delivery.
MasterAgent adheres to the independent research and development of core technologies to ensure the safety and reliability from the underlying architecture to the upper - layer applications.
If large language models have shown the world the potential of AI, then AI Agents are the core tools in this productivity revolution.
As a proxy program driven by large language models that can autonomously call tools to help humans complete complex tasks, the emergence of AI Agents has extended the capabilities of AI from single human - machine dialogue to facing the real world and performing complex tasks. Whether it's academic research, investment and financial management, information collection, or ticket booking and grocery shopping, it can handle them all. All humans need to do is give a single instruction in natural language.
The year 2025 is generally regarded as the inaugural year of Agents. From Manus at the beginning of the year to ChatGPT Agent recently launched by OpenAI, the number of AI Agent products has exploded. All the leading companies in the entire AI field are deploying AI Agents. According to IBM statistics, 99% of AI developers are shifting their focus to the exploration or development of Agents. The research institution Gartner estimates that by 2028, 15% of daily work will be autonomously completed by Agents, indicating extremely broad market prospects.
However, at present, most general - purpose AI Agent products targeting the consumer market still have many problems. For example, the process of building an Agent is relatively complex, requiring a certain foundation in coding, which sets a high threshold. Moreover, the functions of a single Agent are relatively limited, lacking universality and unable to complete complex tasks. Tuning is also a major challenge. Once an Agent is deployed, it is very difficult to modify its settings. These pain points have always hindered the integration of AI Agents into production and daily life scenarios.
Now, a local team from the University of Science and Technology of China, Shenyuan Artificial Intelligence, has independently developed the world's first L4 - level Agent mother system, MasterAgent. With its three characteristics of "rapid multi - Agent generation, professional collaboration, and customized tuning", it enables Agents to truly enter users' lives.
The World's First L4 - level Multi - Agent Generation and Collaboration Platform
The core technology team of Shenyuan Artificial Intelligence is from the University of Science and Technology of China. Since 2018, they have been focusing on the research of Agent autonomous decision - making. It is a home - grown domestic AI technology company.
On May 18, 2025, Shenyuan Artificial Intelligence officially released a fully self - developed Agent product, the MasterAgent Agent mother system. It is the world's first L4 - level multi - Agent generation and collaboration platform, specifically designed to solve complex problems. Here, the L4 level corresponds to the concept in autonomous driving technology, indicating that MasterAgent can complete tasks autonomously throughout the process without human intervention. And "multi - Agent generation and collaboration" is one of the core technologies that enables MasterAgent to autonomously complete complex tasks.
Thanks to the dual - engine architecture of Master Builder and Agent Group independently developed by Shenyuan Artificial Intelligence, MasterAgent can automatically generate professional Agents with complex behavior strategies, long - term memory, professional knowledge systems, and API operation capabilities within a few minutes after receiving a natural - language command. Subsequently, MasterAgent can organize multiple professional Agents into a fully digital team with clear division of labor, and conduct autonomous division of labor and collaborative decision - making according to the task objective, ultimately delivering a complete solution to the user. This is equivalent to MasterAgent being able to hire a group of AI experts instantly according to the user's instruction and organize them into an AI company dedicated to serving a single person.
From understanding the requirements to delivering the results, the entire process is autonomously completed by MasterAgent, with the whole process being AI - driven, fast, and efficient. Currently, MasterAgent has been implemented in specific work scenarios in multiple industries such as law, finance, and healthcare, helping enterprises restructure their work processes and improve production efficiency.
One - click Generation, Professional Collaboration, and Intelligent Tuning
Let's take a look at how it works.
Open the official website of MasterAgent. Different from the homepages of other Agent platforms, which are full of various instructions and buttons, its interface is as simple as that of early large language models. The difference is that you can input more complex tasks in natural language in the dialogue box, such as "Create an Agent that can summarize the content of a paper, evaluate its influence, industry prospects, and commercial value, and recommend some other relevant papers." After a few seconds, MasterAgent will ask some more detailed questions to further clarify the requirements, which saves the user the time of figuring out the Agent settings themselves. If you don't want to supplement the requirements yourself, you can choose "Let AI supplement".
After completing the supplementation, click "Generate Now". With the world's first Agent generation engine, Master Builder, independently developed by it, MasterAgent immediately starts creating a multi - Agent cluster. You can see the real - time progress on the right side. Each Agent has its own avatar, the code is automatically written and output rapidly, and finally packaged into an Agent team. In less than 1 minute, an Agent team composed of a "paper analysis Agent, an evaluation expert Agent, a recommendation system Agent, and a report generation Agent" is created. The whole process truly realizes "one - click generation of a multi - Agent cluster", eliminating all the previous complex steps of creating an Agent. Thanks to the team's pioneering Few - shot Learning technology, only 1 to 3 examples are needed to quickly form a new - domain Agent template, and it supports fine - tuning through reinforcement learning, becoming more intelligent with use.
We give it a link to a latest cutting - edge neuroscience paper, and the multi - Agent team immediately starts dividing the work to execute the task, and the collaboration among the Agents is clearly visible. Under the scheduling of the Agent Group engine, the generated multi - Agent clusters do not mechanically complete their respective parts in sequence, but conduct cross - professional and autonomous collaboration. They split the tasks according to the requirements and complete complex tasks through "collective wisdom", which also reflects the technological breakthrough of MasterAgent in multi - Agent collaboration and scheduling. It is worth mentioning that Agent Group also introduces and improves the PBFT protocol to achieve rapid consensus when there are decision - making conflicts among multiple Agents, ensuring fair and efficient resource scheduling.
A few minutes later, a paper analysis and research report is generated, including the analysis of the paper content, the scoring of the industry influence and commercial value, and the recommendation of relevant papers, which can help industry novices quickly understand professional information and formulate investment strategies.
If you think the result presentation is too rigid, MasterAgent also provides two visualization presentation options: PPT and web page. Select "Generate web page with one click", and after 2 - 3 minutes, a visualized website with high - end color matching, clear layout, and even many special effects will be generated. Throughout the process, the user only needs to give a requirement, and MasterAgent will automatically handle the rest, which shows that MasterAgent has made great efforts in terms of usability.
In terms of instant tuning, MasterAgent is also more flexible. Users can add new single Agents, or enter the settings page of each single Agent at any time to add or modify the Agent settings. They can also modify the tools that the Agent can call in the plug - in section. For example, by adding an instruction of "Generate an audio report" to the report - generation Agent and then adding the "TTS voice synthesis" plug - in, we can make MasterAgent turn the research results into a podcast, allowing us to listen to the audio report at any time.
In addition, MasterAgent also has the ability of recursive self - evolution. Through periodic incremental training, it will automatically update the knowledge base and skill model every week, enabling the generated Agent cluster to learn autonomously and maintain industry - leading capabilities.
From inputting requirements to visualizing results and flexibly adjusting the Agents, MasterAgent truly realizes full - link high - quality service that requires minimal user operation, autonomous decision - making throughout the process, completion of complex tasks within minutes, and accurate delivery of results. It makes Agents no longer an elitist tool exclusive to AI experts and programmers, but an intelligent expert assistant that ordinary users can try to create and use easily.
The Future of AI Agents
As the Nobel laureate and "Godfather of AI", Geoffrey Hinton, said at the WAIC conference: AI Agents will become increasingly intelligent and may even surpass humans. In the foreseeable future, AI Agents will be further integrated into specific application scenarios in all industries. From the automated production in industrial manufacturing to the rapid decision - making in the financial market, from the precise diagnosis and treatment in the medical industry to the data processing in daily office work, AI Agents will undertake more and more tasks, liberating human resources while improving work efficiency and opening a new era of human - machine collaboration. It is conceivable that the focus of competition in the field of artificial intelligence at home and abroad will quickly shift from the computing - power competition at the infrastructure level and the competition of large models at the algorithm core to the more practical AI Agent field.
As a home - grown, purely domestic self - developed Agent product, MasterAgent adheres to the independent research and development of core technologies to ensure the safety and reliability from the underlying architecture to the upper - layer applications. It is committed to creating an Agent solution truly belonging to and serving China, and building a user - friendly and easy - to - use multi - Agent collaborative system.
Now, MasterAgent is constantly exploring along this path. Whether it is for small start - up companies to quickly build automated processes or for large enterprises to construct complex intelligent decision - making centers, MasterAgent hopes that every team or even individual in need can have their own digital assistant. It is believed that in the near future, building, managing, and using one's own Agent cluster will become a convenient experience accessible to everyone.
This article is from the WeChat official account "36Kr Pro", and is published by 36Kr with authorization.