Recently, the 2025 Kingdee Cloud Cosmic AI Summit concluded successfully. Li Fan, Vice President of Kingdee China and General Manager of the Cosmic Platform, gave a keynote presentation titled “Kingdee Cloud\u00b7Cosmic, the Next Generation Enterprise-Level AI Platform,” where he elaborated on the evolution and technological breakthroughs of the Cosmic Platform in the AI era in an accessible manner. Below is the full text of the speech (with some omissions), shared especially with you.<\/span><\/p>\n

 <\/p>\n

I am very pleased that everyone could attend the Kingdee Cloud Cosmic AI Summit. Today, the subject I will share is “Kingdee Cloud Cosmic, the Next Generation Enterprise-Level AI Platform.” As the unified technical foundation for Kingdee’s entire product line, Cangqiong has made numerous breakthroughs in the past year\u2014we have maintained our leading position as the number one in the aPaaS domain in China and among the top ten globally for several consecutive years. What is particularly proud is that last year Kingdee became the <x1><\/span>\u552f\u4e00<\/x1> enterprise management software <x2><\/span>\u5382\u5546<\/x2> selected as one of the top ten in the Chinese market for Gartner’s generative AI models. This honor is inseparable from the collaboration with our client partners in co-creating scenarioses in enterprise management AI, and it also relies heavily on the strong support of the developer ecosystem. It is this kind of mutual effort that has achieved the continuous evolution of the Cosmic Platform.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

01<\/span><\/i><\/strong><\/span><\/p>\n

The evolution of technology drives management change.<\/span><\/strong><\/span><\/p>\n

Looking back at the development history of enterprise management software over the past 40 years from the perspective of technological evolution, the leap from MRP to enterprise management AI is essentially a revolution in management paradigms driven by technological advancements. The large-scale mainframe era of the 1960s gave rise to MRP systems, which used command-line interaction and a single-layer centralized architecture. With the advent of the local area network era, MRP II introduced a client\/server architecture, utilizing a graphical user interface for interaction. In the ERP era, the most typical technological architecture is the emergence of browsers and the web, making the mouse and keyboard the primary means of interaction. We are currently at the intersection of the mobiles cloud era and the AI era, where technologies such as touch screen interaction, natural language processing, and modal data are reconstructing the way enterprises are managed.<\/b><\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

Kingdee’s successful cloud transformation over the past ten years has made it the number one enterprise management SaaS <x1> in China, precisely by seizing the opportunity presented by technological changes such as cloud-native microservices, containerization, and DevOps. As we enter the AI era, technology-driven changes have transformed our management methods and application-driven approaches: Natural language interaction is gradually replacing traditional operation modes, and cutting-edge technologies like brain-computer interfaces are showing promise; in terms of data processing, in the past, SaaS and ERP dealt with structured data, but with the arrival of LLMs, the processing of unstructured data, the development of vector databases, and multimodal technologies have deeply released the value of enterprise data assets. The triple transformation of this interaction method, data form, and technical architecture is driving enterprise management software to complete its paradigm shift from the EBC era to the AI era.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

No technology is achieved overnight, especially AI technology. <\/span><\/b>AI technology has gone through decades, and Kingdee did not just adopt the LLM after it arrived. In fact, Kingdee has been laying out AI for over ten years, and we have gone through three stages of technological evolution: In the first phase, from 2016 to 2018, focusing on natural language processing, OCR recognition, and intelligent robots, the intelligent robot Kingdee Xiao K was published; in the second phase, continuous investment in data middle platform and intelligent middle platform, exploring low-code AI platforms, the first digital employee for enterprise management was published, exploring the implementation of AI productivity in enterprises; the current phase focuses on generative AI and LLM technology, achieving multiple technological breakthroughs between 2022 and 2025. This incremental innovation has significantly shortened the implementation cycle of AI in enterprise scenarioses. The agent applications we see today can truly be implemented in businesses, which is a concentrated presentation of technological accumulation.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

02<\/span><\/i><\/strong><\/span><\/p>\n

Cosmic Agent Platform, Kingdee Intelligent technology foundation<\/span><\/strong><\/span><\/p>\n

Kingdee Cloud Cosmic serves as the foundation for Kingdee products. In the era of intelligent agents, the Cosmic Agent platform is also the unified technical foundation for Kingdee intelligent agents. What capabilities does it possess?<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

Model Services <\/span><\/b>The Cangqiong Agent Platform is a unified model access service that supports Kingdee’s original factory, third-party, and customer self-use, as well as the invocation of Public Cloud model services, and even Private deployment.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

Trusted Service: <\/span><\/b>Through permission control, data masking, model protection, and other mechanisms, ensure that AI applications comply with enterprise-level management standards.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

Operations Service <\/span><\/b>Establish a full lifecycle management of the intelligent agent from demand collection, effect evalsuation to iterative optimization, continuously fine-tuning model accuracy through user feedback data.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

Capacity kit: <\/span><\/b>Provides a complete suite for the development and operation of agents, offering not only general prompts, task orchestration, knowledge bases, and tool libraries, but also a rich library of templates that are deeply integrated with business scenarioses, significantly lowering the development threshold.<\/span><\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

The diverse presentation forms of the Agent developed on the Cangqiong platform can be on the APP side, in H5, embedded, or even without an interface, provided to the SaaS layer through APIs, creating various application scenarioses for Kingdee’s intelligent agents.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

This flexibility has been fully validated in customer co-creation cases\u2014such as the HR self-service <Agent> developed in collaboration with Tongwei Group, which enhances employee service efficiency through natural language interaction; in a large equipment manufacturing enterprise, we built a knowledge <Agent> platform that integrates over 4,000 policies, reports, and evalsuation documents, achieving quick retrievals of quality issues and operation manuals, as well as recommendations for corrective actions, fundamentally changing the traditional “master-apprentice” model of knowledge transfer.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

03<\/span><\/i><\/strong><\/span><\/p>\n

An intelligent platform that understands enterprise management better.<\/span><\/strong><\/span><\/p>\n

Last year, we published version 1.0 of the Sky Agent platform, and today we officially launch version 2.0, which is identified as “an agent platform that better understands enterprise management,” accelerating the inclusive development of enterprise agents. This upgrade focuses on four major dimensions to accelerate the widespread development of intelligent agents for enterprises:<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

Richer templates and tools: <\/span><\/b>Built-in rich task flow templates, prompt templates, MCP tools, as well as APIs and business operation tools, quickly connect enterprise data with business process flows, achieving agile implementation of AI applications.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

Deeper SaaS <Connect>: <\/span><\/b>Based on the unified metadata architecture and relying on the MCP protocol, achieve connections for basic capacity, tool capacity, and process capacity.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

Safer enterprise-level protection: <\/span><\/b>The Kingdee Cangqiong large model is the first enterprise management large model filed with the National Internet Information Office, and the platform has established a complete protection system covering data permissions, content review, and operation auditing.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

More open technical standards: <\/span><\/b>Based on the MCP and A2A (Agent2Agent) open protocols, create a “USB-C” dock for the multi-agent era, achieving seamless collaboration across platforms and systems.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

Taking the order processing scenario of small and micro enterprises as an example, through the Cangqiong Agent platform 2.0 + MCP, users only need to input “Process today’s order emails and arrange for shipment,” and the agent can automatically call the email query tool to extract order information, recognize the content of attachments through OCR, and connect to the SaaS system to complete customer credit verification, inventory checks, order generation, and other 7 steps, ultimately triggering the shipping process and generating a notification email. The entire process requires no manual intervention, and behind it is the orchestration of the Agent + external calls of the MCP service, as well as the perfect collaboration of internal SaaS calls.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

In the analysis scenario, we made key breakthroughs in the technology challenges of ChatBI. Through extensive enterprise practice, we have summarized four major technical challenges: The SQL language of relational databases has a natural gap with natural language, and it is necessary to address the precision of semantic conversion; the semantic matching of massive data indicators can easily exceed the conversation window of large models, and it must rely on enhanced retrievals from vector databases; the specialized terminology and business context unique to enterprises require the construction of private knowledge graphs; and management decisions not only require data queries but also need composite capabilities such as attribution analysis and predictive recommendations. Kingdee ChatBI can convert users’ semantics into a data query language that can be understood, allowing large language models and fine-tuned models to more accurately comprehend and recognize internal jargon and technical terms of enterprises such as “Cloud Hub Appliances BU,” enabling AI to truly “understand” the language of the business.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

04<\/span><\/i><\/strong><\/span><\/p>\n

Closing Remarks<\/span><\/strong><\/span><\/p>\n

Standing at the forefront of the AI technology revolution, Kingdee will continue to promote the development of the Agent ecosystem. “Cannot conquer the heart, how can one conquer the cosmic?” This is the source of the cosmic. <\/b>Kingdee has continuously accumulated and innovated in the field of enterprise management for over 30 years, which is our greatest gene. We firmly believe that: Only by deeply understanding the real needs of customers in the context of enterprise management can we create true business value in the AI era.<\/span><\/p>\n

<\/p>\n

<\/span><\/p>\n

Thank you, everyone.<\/span><\/p>\n

\u00a0<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

Recently, the 2025 Kingdee Cloud Cosmic AI Summit concluded successfully. Li Fan, Vice President of Kingdee China and General Manager of the Cosmic Platform, gave a keynote presentation titled “Kingdee Cloud\u00b7Cosmic, the Next Generation Enterprise-Level AI Platform,” where he elaborated on the evolution and technological breakthroughs of the Cosmic Platform in the AI era in\u2026<\/p>\n","protected":false},"author":30,"featured_media":3093,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_links_to":"","_links_to_target":""},"categories":[171],"tags":[],"by_industry":[],"by_product":[],"by_topic":[28,48],"class_list":["post-2327","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog","by_topic-ai-artificial-intelligent","by_topic-cloud-erp-applications"],"yoast_head":"\n