Zhejiang Province Issues the "Guidelines for the Construction of Industrial Agent Platforms (Trial)": Comprising Four Major Modules—Infrastructure, Core Engine, Agent Construction, and Application Interaction

To implement the decisions and arrangements of the Party Central Committee and the State Council, and to fulfill the relevant requirements of the "Opinions on Deeply Implementing the 'AI+' Initiative," the "Implementation Opinions on the 'AI + Manufacturing' Special Initiative," and the "Zhejiang Province Manufacturing Digital Transformation Action Plan," this Guide is hereby formulated to accelerate the construction of industrial agent platforms in our province and promote the deep integration and application of artificial intelligence technology in the manufacturing sector.

I. General Requirements

With the deep integration of informatization and industrialization as the main line, data elements as the core driver, artificial intelligence technology as the key support, and scenario-driven approaches as an important lever, we will adhere to the principles of "technology empowerment, production efficiency improvement, scenario-driven approaches, and industrial upgrading," advance the construction of industrial agent platforms that are technologically advanced, autonomously collaborative, secure and reliable, open and shared, and jointly built by the ecosystem, provide one-stop AI-enabled services, lower the threshold for small and medium-sized enterprises to apply agents, develop, deploy, and operate industrial agents with autonomous closed-loop capabilities in perception, memory, decision-making, interaction, and execution, and support the leap from automation to intelligent production.

II. Platform Definition

III. Construction Content

(I) Overall Architecture

The industrial agent platform mainly comprises four core modules: infrastructure, core engine, agent construction, and application interaction.

The platform provides foundational support capabilities such as computing power, models, and storage; with industrial large models as the foundation, it integrates industrial mechanisms and domain knowledge and possesses core capabilities such as data governance, ontology construction, knowledge learning, model training, and model interaction; it supports users in customized development and management of agents; it supports the construction of applications for full business scenarios; and it integrates security assurance, standards and specifications, and talent cultivation systems to achieve a full-process closed loop of perception, memory, decision-making, interaction, and execution in industrial scenarios.

The platform should build a highly reliable, elastically scalable foundational support environment to provide basic guarantees for the platform's computing power supply, data governance, IoT access, and model operation, and to meet the high-concurrency, low-latency business requirements of industrial scenarios. It should build a computing power resource pool, construct heterogeneous computing power clusters, integrate diverse computing power resources, and establish an elastic computing power scheduling system to support large model training and inference and ensure on-demand allocation and dynamic scaling of computing power. It should support distributed data storage and build a distributed storage system covering time-series databases, graph databases, object storage, relational databases, and others, to achieve persistent storage, efficient read/write, and disaster recovery backup of massive industrial data. It should possess IoT access capabilities, deploy industrial IoT gateways and protocol conversion modules, support mainstream industrial protocols, and achieve ubiquitous connectivity and data collection for production equipment, sensors, and various industrial control systems. It should support edge computing, enabling data preprocessing, real-time analysis, and instruction issuance in close proximity at production sites, reducing cloud transmission pressure and improving response speed and reliability in industrial scenarios. It should build a trusted computing environment, achieve trusted startup and dynamic measurement of infrastructure, and provide a fully controllable trusted operating environment for model services. It should build a foundational large model base, access or deploy foundational large models adapted to industrial scenarios, form the platform's intelligent capability base, and provide model support for the development and application of industrial agents.

(III) Core Engine

The platform should establish core capability engines driven by data governance, data integration, and ontology, be equipped with high-efficiency model capabilities deeply adapted to the industrial domain, and support the closed-loop iterative resonance of data, models, and applications driven by data, empowered by models, and guided by applications.

1. Data Processing

It should possess data collection and cleaning capabilities, be able to collect and aggregate multi-source data, build automated cleaning processes, support deduplication, completion, noise reduction, and format standardization of industrial data, and continuously improve data quality. It should implement data annotation, provide visual annotation tools for development, support manual and semi-automatic annotation of industrial data, and provide high-quality datasets for model training. It should support data synthesis, be able to use generative technology to synthesize scarce data in industrial scenarios, and expand the scale of datasets.

2. Data Integration

It should possess metadata scanning capabilities, support automatic collection of metadata information from various business systems and equipment, build data catalogs and lineage relationships, and achieve visual management of data assets. It should implement multimodal fusion, be able to integrate structured data such as business reports, semi-structured data such as equipment logs, and unstructured data such as images and audio, and support the formation of a unified data view. It should support virtual data access, be able to use data virtualization technology to achieve federated queries across systems and devices, and enable data interconnection without physical migration.

3. Ontology Engine

It should support ontology construction, be able to build domain ontology models based on industrial domain knowledge, and define entities, attributes, and relationships. It should maintain hierarchical, associative, and constraint rules among ontologies, support knowledge reasoning and graph retrieval, and implement relationship management. It should support skill encapsulation and reuse, be able to encapsulate industrial mechanism models and AI models as intelligent skills, form standardized and reusable intelligent skill components, and provide reliable service invocation capabilities for upper-layer agent construction and multi-agent collaboration.

4. Knowledge Learning

It should accumulate professional knowledge such as industrial physical mechanisms, production processes, and equipment principles, and form a reusable and tunable industrial mechanism model library. It should support the construction of industrial knowledge graphs, cover industrial mechanism models and production operation data, associate core elements such as equipment, processes, and materials, and implement semantic reasoning and knowledge retrieval. It should support multimodal industrial knowledge queries, be able to integrate semantic retrieval, vector retrieval, and graph retrieval technologies, and achieve precise queries of cross-modal data.

5. Training and Tuning

It should have model fine-tuning capabilities and provide low-code fine-tuning tools, support domain adaptation of foundation models based on industrial data and industrial mechanisms, and effectively improve the application performance of models in specific scenarios. It should support reinforcement learning training, be able to build reinforcement learning environments for decision-making scenarios such as production scheduling and resource optimization, and enable autonomous iterative optimization of industrial agents. It should support retrieval augmentation, be able to combine knowledge retrieval with large model generation capabilities, and improve the accuracy and professionalism of generated content by retrieving external knowledge bases.

6. Model Interaction

It should have model management capabilities, support model version control, lifecycle management, and performance monitoring, and support visual management of the entire model process. It should support dynamic adjustment of model operating parameters and dynamic allocation of computing resources to ensure stable model operation. It should support model collaboration, enable task scheduling, information fusion, and unified collaboration among multiple models, and support multi-model collaboration in complex business scenarios.

(IV) Agent Construction

The platform should provide modular and extensible agent development and collaboration capabilities, support the rapid construction of dedicated agents for different industrial scenarios, realize full lifecycle management of industrial agents, and empower their efficient collaboration.

1. Modular Development and Management

It should build a workflow engine, provide visual drag-and-drop design tools, support hybrid orchestration of business processes and large model inference chains, and realize an automated closed loop from data processing to decision execution. It should support component management, encapsulate reusable industrial intelligent components such as fault prediction and quality inspection, and form a component library. It should provide plugin tools and support the access and integration of third-party tools, algorithms, and custom models. It should support testing and verification, provide scenario-based test environments and automated testing tools, and ensure high availability and security of agents before launch. It should support configuration management and realize unified configuration and distribution of agent parameters, permissions, and resources. It should carry out regular iterative updates, establish an agent version iteration mechanism, and support feature updates, performance optimization, and problem fixing.

2. Customized Applications

It should have customized agent adaptation capabilities that meet different business needs. Relying on the platform's multi-dimensional foundational capabilities such as model training, inference deployment, and iterative optimization, it should support enterprises in flexibly building customized agent application scenarios according to their own business needs and industry characteristics. It should support industrial protocol parsing, API integration, and MCP access, and rely on standardized protocols to achieve unified connection between industrial agents and business systems and field devices, reducing the threshold for system integration. It should support agents in calling tools on demand and achieve efficient connection with production equipment, industrial software, and industrial robots. It should have flexible deployment capabilities and simultaneously support multiple deployment methods such as public cloud, private cloud, hybrid cloud, and edge, meeting the differentiated needs of different enterprises for data security and resource scheduling.

3. Multi-Agent Collaboration

It should support personalized creation of multi-agent collaboration modes and interaction mechanisms and standardize collaboration processes. It should support task decomposition, have task planning and orchestration capabilities, and support the transformation of complex business objectives into subtask flows assigned to different agents for execution. It should have operation management capabilities, be able to monitor the operating status of multiple agents, and realize task load balancing, exception fault tolerance, and conflict resolution. It should support memory sharing, be able to build a shared memory library for multiple agents, and realize the exchange of knowledge, experience, and task status across agents.

(V) Application Interaction

The platform should face the entire industrial value chain, create scenario-based intelligent applications and a unified interaction entry point, and improve business efficiency and decision-making levels. It should support the construction of R&D and design agents with capabilities such as industry forecasting, product design, and process simulation, assisting R&D personnel in shortening product R&D cycles and optimizing design solutions. It should support the construction of production and manufacturing agents with capabilities such as process collaboration, task scheduling, energy consumption control, and automatic production scheduling, improving production efficiency and reducing energy consumption and costs. It should support the construction of operation and management agents with capabilities such as industry analysis, business reports, supply chain optimization, and decision support, providing data-driven intelligent suggestions for enterprise business decisions. It should support the construction of customer service agents with capabilities such as intelligent Q&A, technical support, and precision marketing, able to automatically handle service requests and improve customer satisfaction and brand stickiness. It should support the construction of industrial collaboration agents, break down information barriers upstream and downstream of the industrial chain, support cross-enterprise supply-demand matching, capacity sharing, and logistics collaboration, promote supply chain collaboration, facilitate the exchange of knowledge and technical standards within industrial clusters, and empower the cultivation of the "415X" advanced manufacturing clusters. It should provide a unified interaction interface, build a visual and lightweight interaction portal, support multi-terminal access such as PC, mobile, and industrial large screens, and realize unified management, task distribution, and result display of industrial agents. IV. Guarantee System

Build a foundational security assurance system with basic security protection capabilities such as intrusion detection, vulnerability scanning, encrypted transmission, storage desensitization, access control, code inspection, and supply chain control, so as to achieve safe and reliable operation of platform data, models, and applications and effectively safeguard platform network and data security. For industrial scenarios, improve log auditing and abnormal behavior monitoring, and establish access auditing and emergency response mechanisms to prevent business interruption and operational risks. Establish a production-grade security mechanism for agents, improve multi-level verification, instruction auditing, and abnormal circuit-breaking capabilities, strengthen risk prevention and control across the "decision-execution" loop, accommodate the platform's requirements for independent and controllable technologies, and effectively ensure the operational security of agents. Establish a standards and specifications system for platform construction and operation, formulate platform technical standards, business standards, and management standards, and effectively ensure the platform's compatibility and compliance. Establish a talent cultivation system suited to the platform's development, covering roles at different levels and in different categories, including platform construction and operations and maintenance, data processing, ontology construction, model training and tuning, agent development, and application interaction, and carry out skills training and certification. Establish a collaborative mechanism among industry, academia, research, and application, and join forces with universities, research institutions, and enterprises to cultivate interdisciplinary industrial intelligence talent.

Source: Zhejiang Economic and Information Technology Department | Industrial Digitalization Promotion Division

www.smartcity.team

Disclaimer: Please indicate the source as "Smart City Industry Analysis (www.smartcity.team)" and the relevant author when reposting. Except for original content, some information on this site comes from the Internet. If the relevant content involves infringement of rights, please contact the contact information at the bottom to delete it.

[Digital Technology Special Research] Community

In-depth peer exchange, focused on research and material accumulation in the smart city and digital technology industries.

Continuously updated with the latest industry white papers, industry research, policy developments, and implementation solutions, helping industry practitioners efficiently accumulate knowledge and share experience.

A vast amount of curated content is continuously iterated, providing frontier perspectives, connecting industry talent, carrying out case co-creation and resource exchange, and jointly building a high-quality industry exchange atmosphere.

#post-##

Recommended for you

Hefei's Several Measures to Accelerate Agent-Led High-Quality Development: 25 measures, centered on improving foundational model capabilities, tackling underlying common technologies, multi-domain applications, intelligent terminal integration, scientific research, major scientific facilities, quantum technology, the innovation and entrepreneurship ecosystem, security governance, and factor supply, to promote agent technology R&D, scenario applications, and industrial ecosystem development

The Ministry of Transport issues the Guidelines for AI Application Scenarios in the Water Transport Sector: 22 AI application scenarios in the water transport sector are proposed from three aspects: safety improvement, quality and efficiency enhancement, and green transformation

Jiangsu Province's Implementation Plans for Supporting AI Innovation and Development: the Jiangsu Province Token Voucher Implementation Plan, the Jiangsu Province Corpus Voucher Implementation Plan, and the Jiangsu Province Model Voucher Implementation Plan

Hubei Province's Three-Year Action Plan for In-Depth Implementation of "AI+" (2026-2028): implement six major "AI+" special actions, build a provincial-level AI public service platform by 2028, cultivate 100 benchmark agent applications, expand 500 high-value application scenarios, and achieve a penetration rate of new-generation intelligent terminals and agents exceeding 80%

Beijing Municipal Guidelines for AI Applications in the Education Sector (2026): a "1+5+2" architecture system, deepening seven types of application scenarios: "using intelligence to assist teaching, using intelligence to assist learning, using intelligence to assist evaluation, using intelligence to assist nurturing, using intelligence to assist research, using intelligence to assist management, and using intelligence to assist construction"