CN121455468A - Agent construction and operation system and method based on AI large model technology - Google Patents

Abstract

The invention discloses an Agent construction and operation system and method based on an AI large model technology, which relate to the technical field of AI large models, wherein the Agent construction and operation system based on the AI large model technology comprises an interactive node configuration system and a visual flow development system, wherein the interactive node configuration system outputs a standardized component unit, and the visual flow development system realizes the combination arrangement of a business flow through a component quoting mechanism; the interactive node configuration system and the visual flow development system are provided with dynamic update links, the interactive node configuration system and the visual flow development system adopt interactive closed-loop design, different delivery forms are automatically generated and adapted based on the configured node flow, the interactive node configuration system is adapted to the development requirements of the whole scene, the layered development system reduces the development cost, the complicated flow arrangement capability breaks through the traditional limitation, and the system can automatically generate the delivery forms which are adapted to the different requirements, so that the efficiency is improved.

Description

Agent construction and operation system and method based on AI large model technology

Technical Field

The invention relates to the technical field of AI large models, in particular to an Agent construction and operation system and method based on an AI large model technology.

Background

In the age of rapid development of digitalization nowadays, the application of agents in various fields is becoming wider, the agents can simulate human behaviors and autonomously complete specific tasks, and great convenience is brought to enterprises and users. However, the existing Agent construction and operation system based on the AI large model technology has a plurality of defects, and is difficult to meet the actual use requirements.

From the research and development perspective, the traditional system lacks a layered development system, so that the development process is disordered and the standardization degree is low. For example, when building agents, it is difficult for different functional modules to achieve efficient assembly arrangement, and the lack of uniform standards between individual components makes development inefficient and difficult to maintain and upgrade. Meanwhile, most systems do not provide standardized component units, and developers need to repeatedly write a large amount of codes, increasing development cost and period.

In terms of operation management, the existing system cannot realize dynamic updating of links, and when the service requirements change, the system is difficult to adjust and optimize quickly. Moreover, due to the lack of interactive closed-loop design, users cannot timely obtain feedback in the configuration and development processes, and configuration errors and development direction deviation are easy to occur. In addition, the traditional system does not have a good visual flow development function, and developers can only build business flows through complex codes.

In terms of delivery morphology, existing systems are difficult to adapt to different demands. Many systems can only generate a single delivery form, and cannot be flexibly adjusted according to different terminals and service scenes, so that the application range of the Agent is limited. For example, when interfacing with a third party, the interfacing process is cumbersome due to lack of functions such as security protocol encapsulation and message format conversion, and security problems are likely to occur.

Therefore, in view of this current situation, development of an Agent construction and operation system and method based on AI large model technology is urgently needed to meet the actual needs.

Disclosure of Invention

In view of the defects existing in the prior art, the main purpose of the invention is to provide an Agent construction and operation system and method based on an AI large model technology, wherein an interactive node configuration system is adopted to give consideration to flexibility and usability, adapt to full scene development requirements, break through the limitation of a single configuration mode of the existing low-code platform, build standardized multiplexing ecology by a layered development system, greatly reduce development cost, output standardized assembly units of a node design layer can be multiplexed across processes, a process arrangement layer forms complete business logic by dragging or code combination nodes, complex process arrangement capability breaks through traditional limitation, adapt to multi-industry business scenes, and the system can automatically generate API interfaces adapting to different requirements based on node processes of configuration completion, thereby improving front-end and back-end team cooperation efficiency.

The technical scheme is that the Agent constructing and operating system based on the AI large model technology comprises an interactive node configuration system and a visual flow development system, wherein the interactive node configuration system and the visual flow development system are in a layered development system, the interactive node configuration system outputs standardized component units, the visual flow development system realizes the combined arrangement of service flows through a component quotation mechanism, the interactive node configuration system and the visual flow development system are provided with dynamic update links, the interactive node configuration system and the visual flow development system adopt interactive closed-loop design, and delivery forms adapting to different requirements are automatically generated and released based on the configured node flows.

The method comprises the steps that a screen-division type interaction architecture is adopted in the interactive node configuration system, one side of the screen-division type interaction architecture is a parameter configuration area, the other side of the screen-division type interaction architecture is a real-time rendering area, the visual flow development system comprises a bidirectional editing core module used for supporting canvas editing and code editing of a workflow, the bidirectional editing core module comprises a canvas editing mode and a code editing mode, the bidirectional editing core module supports dragging a pre-configuration component from a node library, a service flow is constructed through canvas connection in the canvas editing mode, structured data of the flow is directly edited in the code editing mode, and node logic relations and data circulation rules are defined.

The dynamic updating link comprises a change tracing mechanism and a guarantee data consistency, wherein the change tracing mechanism is realized by adopting a synchronous algorithm, the guarantee data consistency is realized by adopting an operation atomicity, namely single modification is used as an independent transaction, a state snapshot, namely a configuration version before modification is saved, conflict processing, namely a priority principle is finally modified, components in a visual flow development system can trace to an interactive node configuration system, and after the bottom configuration is modified, the components are automatically synchronized to all flow instances referring to the components, so that centralized management and global effect of configuration are realized.

The interactive closed loop design is realized through mode switching logic and seamless connection, scene judgment is carried out through the mode switching logic, parameter batch modification, automatic screen splitting mode switching, complex logic editing, code mode switching proposal and user habit analysis through operation history are carried out, the seamless connection comprises a shared memory data pool, view state serialization storage and automatic focus element positioning, a screen splitting interactive framework in an interactive node configuration system and a bidirectional editing core module in a visual flow development system form complementation, a developer is supported to select an operation mode according to a scene, and coverage is facilitated from basic configuration to complex logic development.

The standardized component unit is characterized in that the standardized component unit comprises a node classification comprising an input node, a processing node and an output node, wherein the parameter specification comprises an input parameter and an output parameter, the input parameter comprises a filling field mark and a basic type check, and the output parameter is a structured data template.

The method comprises the following steps of generating an API interface, generating an Agent and an AI model tool (MCP), wherein the generation of the API interface comprises an interface configuration node and a terminal self-adaptive node, the terminal self-adaptive node comprises web page end adaptation, mobile end adaptation and third party docking, and the third party docking comprises security protocol encapsulation and message format conversion.

The method comprises a configuration stage, a generation stage and a debugging stage, wherein the configuration stage comprises a step of dragging an interface configuration node by a developer and setting basic parameters, if a link is changed, the whole flow can be changed only by replacing one or more nodes without reconfiguration, the generation stage comprises a step of automatically connecting a system with a terminal adaptation node to generate three sets of interface schemes in parallel, and the debugging stage comprises a step of independently testing nodes by each terminal, and automatically positioning problem feedback to the corresponding nodes.

The interactive node configuration system generates a standardized component unit comprising a condition judgment node and a circulation control node; the visual flow development system realizes multi-level nested flow arrangement by referring to the standardized component unit and is used for complex flow arrangement.

The method is characterized in that the condition judgment node generated by the interactive node configuration system comprises a plurality of output ports for binding independent rule expressions, and the visual flow development system realizes multi-level nested flow arrangement by dragging or code referencing the node.

The method for constructing and operating the system by the Agent based on the AI large model technology comprises the following steps of firstly, receiving node parameter definition and outputting a standardized component unit through an interactive node configuration system;

Secondly, through a visual flow development system, a component referencing mechanism is adopted to call the generated standardized component unit, so that the combination arrangement of the business flow is realized;

thirdly, the components of the interactive node configuration system are modified, all flow instances referring to the components are automatically traced, and modified contents are synchronized to the associated flow, so that dynamic update synchronization is realized;

The fourth, interactive node configuration system adopts interactive closed loop design in the node configuration stage to realize real-time linkage of parameter configuration and component preview;

fifth, the API interface adapting to the different terminal requirements is automatically generated and issued.

Compared with the prior art, the invention has obvious advantages and beneficial effects, and in particular, the technical scheme can be as follows:

The first and the second node configuration systems give consideration to flexibility and usability, adapt to the development requirements of the whole scene and break through the limitation of the single configuration mode of the existing low-code platform;

The second and layered development systems construct standardized multiplexing ecology, so that development cost is greatly reduced, an interactive node configuration system and a visual flow development system are adopted to respectively correspond to two-layer architecture of node design and flow arrangement, a node design layer outputs standardized assembly units and can be used for multiplexing the flows, a flow arrangement layer forms complete business logic through dragging or code combination nodes, the bottom layer node configuration can be automatically synchronized to all reference flows after modification, the problem of poor code logic multiplexing in traditional development is thoroughly solved, compared with the prior art, the component multiplexing rate is remarkably improved, the system maintenance cost is greatly reduced, and the method is particularly suitable for large-scale iteration of enterprise-level application.

Thirdly, the complex flow arranging capability breaks through the traditional limitation, adapts to multi-industry service scenes, is different from the limitation that the traditional platform only supports linear flows, and adopts a flow development mode that the multistage nested logic can be quickly constructed through visual dragging;

Fourth, the data standardization mechanism guarantees the system stability, avoids the logic fault, namely, the input and output data model is uniformly defined through the standardization mechanism in the node configuration stage, and automatically completes the data mapping and format conversion in the process development, so that the problem of non-uniform data model among components in the prior art is solved, the logic error caused by manually processing the data conversion is avoided, the whole process standardization management and control is formed from field naming standards to data type verification, and the stability and compatibility of the system data flow are greatly improved.

Fifth, the multi-terminal adaptation capability realizes 'one-time configuration and multi-scene application', and improves the cooperation efficiency:

Based on the configured node flow, the system can automatically generate API interfaces adapting to different requirements, a developer can flexibly adjust elements such as interface paths, request modes, return formats and the like through parameter configuration, and multi-port calling requirements (such as front-end pages, mobile-end applications and third-party systems) can be met without repeated development;

compared with the traditional mode that the multi-port interface needs to be independently developed, the cross-platform adaptation workload is obviously reduced, and the front-end team and back-end team cooperation efficiency is improved.

In order to more clearly illustrate the structural features and efficacy of the present invention, a detailed description thereof will be given below with reference to the accompanying drawings and examples.

Drawings

FIG. 1 is a schematic diagram of data normalization and API generation logic of the present invention;

FIG. 2 is a block diagram of an interactive node configuration system and visualization process development system of the present invention;

FIG. 3 is a block diagram of a visualization process development system of the present invention;

FIG. 4 is a flow chart of a method of the present invention for an Agent build and run system based on AI large model techniques;

FIG. 5 is a schematic diagram of a linkage mechanism architecture of the interactive node configuration system and the visualization process development system of the present invention;

FIG. 6 is a flow chart of data normalization and API auto-generation according to the present invention.

Detailed Description

The invention discloses an Agent constructing and operating system based on an AI large model technology, which is shown in fig. 1 to 6, and comprises an interactive node configuration system and a visual flow development system, wherein the interactive node configuration system and the visual flow development system are in a layered development system, the interactive node configuration system outputs standardized component units, and the visual flow development system realizes the combined arrangement of service flows through a component quoting mechanism;

The interactive node configuration system and the visual flow development system adopt interactive closed-loop design, and automatically generate and release delivery forms adapting to different requirements based on the configured node flow.

The interactive node configuration system adopts a split-screen type interaction architecture, wherein the split-screen type interaction architecture is a left split-screen type interaction architecture and a right split-screen type interaction architecture, one side of the left split-screen type interaction architecture and the right split-screen type interaction architecture is a parameter configuration area, the other side of the left split-screen type interaction architecture and the right split-screen type interaction architecture is a real-time rendering area, the visual flow development system comprises a bidirectional editing core module for supporting canvas editing and code editing of a workflow, the bidirectional editing core module comprises a canvas editing mode and a code editing mode, the bidirectional editing core module supports dragging a pre-configuration component from a node library, canvas connection is carried out through the canvas editing mode to construct a service flow, and the code editing mode directly edits structured data of the flow to define node logic relations and data flow rules.

The dynamic update link comprises a change tracing mechanism and a guarantee data consistency, wherein the change tracing mechanism is realized by adopting a synchronous algorithm, and the guarantee data consistency is specifically realized by operating atomicity, single modification as an independent transaction, state snapshot, configuration version before modification, conflict treatment and final modification priority principle;

components in the visual flow development system can trace to the interactive node configuration system, and after the bottom configuration is modified, the components are automatically synchronized to all flow instances referring to the components, so that the centralized management and global effect of the configuration are realized.

The interactive closed loop design is realized through mode switching logic and seamless connection, scene judgment is carried out by adopting the mode switching logic, parameters are modified in batches, a split screen mode is automatically switched, a complex logic editing mode is suggested, a user habit is analyzed through operation history, the seamless connection comprises a shared memory data pool, view state serialization storage and automatic positioning of focus elements, a left split screen interactive architecture and a right split screen interactive architecture in an interactive node configuration system are complementary with a bidirectional editing core module in a visual flow development system, a developer is supported to select an operation mode according to the scene, and coverage from basic configuration to complex logic development is facilitated.

The standardized component unit has the specification that node classification comprises input nodes, processing nodes and output nodes, parameter specification comprises input parameters and output parameters, the input parameters comprise necessary field marks and basic type verification, and the output parameters are structured data templates.

The delivery form comprises an API interface, an Agent and an AI model tool (MCP), wherein the generation of the API interface comprises an interface configuration node and a terminal self-adaption node, the terminal self-adaption node comprises a webpage end adaption, a mobile end adaption and a third party docking, and the third party docking comprises a security protocol encapsulation and message format conversion.

The generating workflow of the API comprises a configuration stage, a generating stage and a debugging stage, wherein the configuration stage comprises a step of dragging an interface configuration node by a developer and setting basic parameters, if a link is changed, the whole flow can be changed only by replacing one or more nodes without reconfiguration, the generating stage comprises a step of automatically connecting a system with a terminal adaptation node to generate three sets of interface schemes in parallel, and the debugging stage comprises a step of independently testing nodes by each terminal, and automatically positioning problem feedback to the corresponding nodes.

The visual flow development system realizes multistage nested flow arrangement by referring to the standardized component unit and is used for complex flow arrangement.

The condition judgment node generated by the interactive node configuration system comprises a plurality of output ports for binding independent rule expressions, and the visual flow development system realizes multi-level nested flow arrangement by dragging or code referencing the node.

The interactive node configuration system is structurally designed, wherein a left-right split screen type interactive architecture is adopted, the left side is a parameter configuration area, a developer can input and define attributes (such as node types and input and output parameters) of functional nodes through JSON, the right side is a real-time rendering area, the form of graphical components of the nodes is dynamically displayed, the graphical component comprises node appearance, interface identification and parameter layout, and real-time synchronization of configuration data and visual effects is achieved.

The core mechanism is as follows:

the dual-mode configuration engine is used for automatically generating a standardized configuration data structure through visualization operation, enabling code editing to support depth parameter customization, and realizing bidirectional conversion through a unified data model;

And the real-time synchronous link is used for pushing configuration data to a rendering engine through a data binding mechanism, triggering component state update and ensuring millisecond-level consistency of contents at two sides.

The model adopts a three-level structure (1) a basic attribute layer, which comprises a node ID (character string type, length limit 32 characters), a node type (predefined enumeration value) and a display name (character string type, necessary filling items);

(2) An interface definition layer for recording input/output port configuration (array structure, each element contains port ID, direction enumeration, data type);

(3) And the parameter configuration layer is used for storing component-level parameters (nested structures, supporting basic types such as character strings, numerical values, boolean and the like).

Bidirectional mapping rules

The visual form- & gt data model (1) form control and model field establish strict type binding (drop-down box- & gt enumeration type);

(2) The dynamic form is automatically converted into a standardized parameter array;

(3) Value range verification (e.g., numerical range, regular expression matching) is performed in real-time.

The visual flow development system comprises a framework design, a two-way editing core module integrating canvas editing and code editing, and a graphical arrangement and structural definition of a workflow. Under the visualization mode, a developer can drag in a pre-configuration component from a node library and construct a business flow through canvas connection, and under the code mode, structured data of the flow can be directly edited, a node logic relationship and a data flow rule are defined, and seamless switching is supported by the two modes.

1. Data binding mechanism

A Proxy-based responsive system is adopted (1) a depth agent of a configuration object is established;

(2) The attribute modification triggers an accurate update notification;

(3) The change set minimizes the transfer (only the difference portion is sent).

2. The synchronization flow is as follows:

configuration change, generation of difference packets, receiving by a rendering engine, and updating of local DOM (document object model) in millisecond-level synchronous guarantee

The pure front end implementation scheme is characterized in that (1) no network transmission link (elimination of network delay) exists;

(2) Memory level data sharing (communication within the same process);

(3) Micro-task queue scheduling (ensuring UI thread priority) is employed.

Performance index:

(1) Simple attribute update, <5ms;

(2) Structure change processing is <20ms;

Exception handling design

Error isolation mechanism:

The method comprises the steps of (1) returning to a last valid state (2) when rendering fails, displaying a yellow frame warning (3) by an error component, outputting a detailed error stack (3) by a control console, and guaranteeing data consistency, wherein (1) an operation snapshot (2) is generated before each change, an atomic update strategy (3) is adopted, key operation support is adopted, JSON code is withdrawn/redone, and a data model is adopted, wherein (1) a fault-tolerant analysis strategy (non-key field is allowed to be deleted) is adopted;

(2) Type forced conversion (string "true" →boolean value true);

(3) Auto-complement default values (e.g., use type default icons when no icon field is provided).

The two-way editing synchronization is that the canvas connection line is modified to automatically update the bottom data structure, and the adjustment code logic triggers the canvas to re-render, so that the consistency of the flow logic and the visual display is ensured;

And the dynamic verification mechanism is used for automatically verifying the compatibility of the node interfaces in the configuration process, prompting logic errors in real time and guaranteeing the executable performance of the flow.

1. Bidirectional editing synchronization implementation mechanism

1.1 The conversion flow system from canvas operation to data update establishes the accurate mapping relation between canvas operation and bottom data, and the connection operation conversion includes converting graphic connection into standard connection description object including source node mark, destination node mark, data transmission type and other core attributes

Node movement processing, namely capturing node coordinate change in real time, updating corresponding layout data and maintaining a spatial index structure

Attribute modification synchronization by mapping visual edited attribute adjustments directly to corresponding fields of configuration data

1.2 The synchronization process from code editing to canvas rendering adopts a difference-driven intelligent rendering mechanism, namely, change detection, namely, recognizing data structure change through a depth comparison algorithm, and constructing a change influence range model

Differential analysis determining redrawing strategy according to change type (node topology/style attribute/connection relation)

Directional updating, namely selecting a rendering mode of full-scale layout or local redrawing based on change influence analysis result

2. Dynamic verification mechanism implementation details

2.1 The interface compatibility check rule system implements a multi-dimensional check policy that type system checks that a strict data type matching check is performed, including basic type compatibility and composite type structure consistency

Protocol compliance checking, namely verifying whether a node interface accords with a predefined communication protocol specification data flow direction check, namely ensuring that a connection direction is consistent with an input-output direction defined by the interface

2.2 The real-time error feedback system provides a multi-level error prompt scheme, wherein the visual marking system adopts a hierarchical warning mark (serious error/warning prompt) to mark an abnormal element intelligent prompt mechanism, a detailed error description and repair suggestion positioning navigation function is displayed through a suspension panel, a bidirectional mapping engine is supported to rapidly position a corresponding code position from a visual element, lossless conversion of graphic operation and structured data is realized, intelligent difference analysis is realized, the minimum set of change influence ranges is accurately identified, a hybrid verification module is combined with static type inspection and dynamic protocol verification, millisecond synchronization of editing operation and rendering update is realized, invalid redrawing operation is reduced, error detection accuracy is improved, and programming efficiency of a complex flow is improved.

The hierarchical development system comprises a node configuration system, a visual flow development system and a control system, wherein the node configuration system outputs standardized component units, and the flow development system realizes the combination arrangement of business flows through a component reference mechanism to form a hierarchical development mode of atomic capability-composite flow;

dynamically updating links, namely enabling components in the flow to trace to a node configuration system, automatically synchronizing the modified bottom configuration to all flow instances referring to the components, and realizing centralized management and global effect of configuration;

The interactive closed loop design comprises the steps of complementary split screen interaction of node configuration and dual-mode editing of flow development, and supporting operation modes of priority or fine control of developers according to scene selection efficiency, and covering the whole scene requirement from basic configuration to complex logic development.

1. Layered development system implementation scheme

1.1 Standardized component normalizes node classification:

Input nodes, data access classes (such as API call and file uploading);

Processing nodes, logical operation classes (such as conditional branches, data conversion);

Output node-result output class (e.g., database write, message push).

Parameter specification:

Input parameters, i.e. a field mark (red x number) to be filled in, and basic type check (text/number/boolean);

output parameters structured data templates (fixed field + extension field).

1.2 Component referencing mechanism implementation:

cross-system referencing is achieved through global component IDs.

The reference relationships are stored in a central metadata repository.

The component triggers the reference check when modified.

2. Dynamic update link implementation

2.1 Change traceback mechanism synchronization algorithm:

an event driven mode (observer mode) is employed;

Modifying operation to generate event log;

the consumer (flow system) subscribes to change events.

2.2 Data consistency assurance technical measures:

operation atomicity, single modification as independent transaction;

a state snapshot, which is to store the configuration version before modification;

conflict processing, namely, finally modifying the priority principle.

3. Interactive closed loop design scheme

3.1 Mode switching logic scene judgment:

parameter batch modification, automatic switching of split screen modes;

complex logic editing-suggesting to switch code modes;

User habits are analyzed through the operation history.

3.2 Seamless joining implementation

The technical key points are as follows:

Sharing a memory data pool;

storing view states in a serialization manner;

The focus element is automatically positioned.

The left-right split screen type interaction architecture reduces development cognition cost through real-time linkage of configuration and preview, and realizes 'what you see is what you get' configuration experience;

the bidirectional editing core module integrates the advantages of visual dragging and code programming, combines development efficiency and logic flexibility, and adapts to users with different technical backgrounds;

And the hierarchical development system is used for separating a logic level of node definition and flow arrangement, improving the reusability of components and the maintainability of the flow and obviously shortening the development period of complex service scenes.

The invention adopts the left-right split-screen type interaction architecture, carries out node parameter configuration through JSON codes or visual forms on the left side, renders component forms and data interfaces on the right side in real time, realizes visual interaction of configuration and visibility, supports seamless switching of two modes of visual dragging and JSON codes in a process development link, and enables a developer to quickly build a service process through canvas connection or realize deep customization of complex logic through code editing. The combined design of split screen configuration and mode switching not only meets the fine control requirement of technicians on underlying logic, but also reduces the operation threshold of non-technicians, and realizes the dual promotion of development efficiency and logic expression capability.

The hierarchical development system builds a standardized multiplexing ecology, so that development cost is greatly reduced, namely, an interactive node configuration system and a visual flow development system are adopted to respectively correspond to a node design and a flow arrangement two-layer architecture, and a node design layer outputs standardized assembly units (such as data check nodes and model call nodes) which can be multiplexed across the flow;

After the configuration of the bottom node is modified, the configuration of the bottom node can be automatically synchronized to all reference flows, and the problem of poor code logic reusability in the traditional development is thoroughly solved. Compared with the prior art, the component multiplexing rate is obviously improved, the system maintenance cost is greatly reduced, and the method is particularly suitable for large-scale iteration of enterprise-level application.

The complex flow arranging capability breaks through the traditional limit, adapts to multi-industry business scenes, and is different from the limit that the traditional platform only supports linear flows;

Complex rule nesting and depth logic customization can be achieved through JSON codes. The complex business scene can be dealt with without depending on a custom code, the process arrangement efficiency is remarkably improved, and the problem of low development efficiency of the complex scene in the prior art is fundamentally solved.

The data standardization mechanism ensures the stability of the system, avoids logical faults, namely uniformly defines an input and output data model through the standardization mechanism in the node configuration stage, automatically completes data mapping and format conversion in process development, solves the problem of non-uniform data model among components in the prior art, avoids logical errors caused by manually processing data conversion, and forms full-process standardization management and control from field naming standards to data type verification, thereby greatly improving the stability and compatibility of system data flow.

The multi-port adaptation capability realizes one-time configuration and multi-scenario application, improves the cooperation efficiency, automatically generates an API interface adapting to different requirements based on the configured node flow, enables a developer to flexibly adjust elements such as an interface path, a request mode, a return format and the like through parameter configuration, meets multi-port calling requirements (such as a front-end page, a mobile-end application and a third-party system) without repeated development, and obviously reduces cross-platform adaptation workload and improves front-end and back-end team cooperation efficiency compared with a traditional multi-port interface mode needing independent development.

1.1 The interface configuration node inputs parameters:

basic paths (e.g., "/aip/v 1");

Protocol type (HTTP/HTTPs);

Output capability:

generating a standard interface frame;

And outputting an interface document template.

1.2 Terminal adapting node

Webpage end adaptation:

automatically adding CORS support;

Converting a field name format;

mobile end adaptation:

Filtering intelligent fields;

data compression processing;

Third party docking:

packaging a security protocol;

And (5) converting a message format.

2. Actual workflow

Configuration phase:

the developer drags the interface configuration node;

Setting basic parameters (path/protocol/format);

if the link is changed, only one or more nodes are replaced, and the whole flow does not need to be reconfigured;

the generation stage:

The system automatically connects with a terminal adapting node;

generating three sets of interface schemes in parallel;

Debugging:

Each terminal independently tests nodes;

the problem feedback is automatically located to the corresponding node.

The multiplexing rate of the interface development nodes is improved, and the adaptation and adjustment time is shortened.

In the node design level, a left-right split screen type interaction architecture is adopted, a left side area allows a developer to finely configure node parameters through two modes of JSON codes or visual forms, and a right side area renders component forms of corresponding nodes in real time, including node appearance, data input and output interfaces and the like, so that the developer can intuitively see configuration effects, and convenience and accuracy of configuration are greatly improved. In the process development link, seamless switching of two modes of visual dragging and JSON code editing is supported in a breakthrough manner. The developer can quickly build a basic framework of the business process by simply dragging nodes on the canvas and connecting the nodes, and can switch to a code editing mode when needed to carry out depth customization on the bottom logic of the process. The layered interactive architecture design remarkably improves the development efficiency of personnel with different technical levels.

The left and right split screens of node development are popup windows, 50% of the left and right split screens can be full screen, and the initial width is 80% of the screen. The process development JSON and view mode can select one at the same time, and the view only converts the JSON into the real component or JSON.

The standardized component unit multiplexing and dynamic updating mechanism is that the output of the node design layer is a standardized component unit, and the components have high universality and reusability and can be repeatedly used in different flows.

Such as select nodes, model nodes, etc. More importantly, when the configuration of the underlying node is modified, the system can automatically synchronize the modifications to all the flow instances referencing the node, thereby realizing centralized management and global validation of the configuration. The mechanism effectively avoids resource waste caused by repeated development of the same functional module in the traditional development mode, and simultaneously ensures the consistency and stability of the system.

The analysis and execution logic of canvas data includes front end transmitting canvas data comprising two core parts of node set and connection relation, and back end executing engine to realize automatic execution of canvas data structure

1. Node set, each node contains unique ID, type identification (such as "data check" and "model call"), input/output port definition (describing port ID, data type and connection target), and configuration parameters (such as check rule and model address).

2. The connection relation is that the data flow direction between nodes is recorded, and the data transmission path (such as 'output port 1 of node A → input port 2 of node B') is defined through the mapping of 'source node ID+output port' and 'target node ID+input port'.

Execution flow

1. The engine automatically analyzes node dependence according to the connection relation to generate a loop-free execution sequence (such as executing the starting node without dependence and then executing the downstream node), so as to avoid deadlock caused by cyclic dependence.

2. Input parameter assembly, namely, for each node to be executed, the engine automatically collects the results of all output ports of the upstream nodes, and maps the results to the input parameters of the current node according to port definition (such as the input parameters of node B = the results of output port 1 of node A).

3. And (3) node chained execution, namely sequentially calling the execution logic of the nodes according to the ordering result, and storing the output parameters of the current node into a global data pool for the downstream node to take until all the nodes execute.

(II) unified specification of BaseNode base class all nodes inherit from BaseNode base class, and unified attribute and interface are defined by base class so as to ensure that execution engine can be universally scheduled by base class core attribute

1. The node IDs are in one-to-one correspondence with the node IDs in the canvas and are used for positioning the nodes.

2. And (3) inputting a parameter set, namely storing input data transmitted by an upstream node, wherein a key is an input port ID, and a value is corresponding data.

3. And outputting a parameter set, namely storing a result after the node is executed, wherein a key is an output port ID, and a value is processed data.

4. The execution state is marked as the stage (non-execution/success/failure) of the node for the engine to monitor the flow.

Base class core interface

1. And the parameter checking interface is used for automatically checking whether the input parameters accord with port definitions (such as data types and necessary filling items), and terminating execution and marking a failure state when the checking fails.

2. And the execution interface is used for receiving the input parameter set and returning the output parameter set by core logic (such as data verification and model call) which is needed to be realized by the subclass.

3. And the result acquisition interface is used for reading the node output parameters by the engine and transmitting the node output parameters to the downstream node.

After the individuation of the child nodes realizes the child node inheritance base class, only the service logic of the child node needs to be expanded, and the universal function, namely the data check node, does not need to be repeatedly developed

1. The specific attribute is a check rule list (such as 'field A needs to match the mobile phone number format', 'field B needs to be more than 100').

2. And executing logic, namely after the base class parameter checking interface is called to pass, checking the input parameters according to a rule list, and storing the checking result (whether the base class parameter checking interface passes or not and the error reason) into an output parameter set.

Model call node

1. The specific attributes are model address, timeout time, input-output mapping rule (e.g. "map field X of input port 1 to model parameter Y").

2. And executing logic, namely converting the input parameters according to the mapping rule, calling an external model interface and converting the returned result into an output parameter set.

The method has the advantages of universality, unified specification of base class, capability of enabling an engine to schedule any type of node, no need of modifying engine logic for newly added nodes, high efficiency, capability of ensuring that nodes execute according to a dependency sequence, avoiding invalid waiting, automatic input and output transmission, reducing manual intervention, and reliability, namely, capability of ensuring that the nodes execute only when input is legal and accurately locate problem nodes when failure occurs.

The invention breaks through the limitation that the traditional platform only supports linear flow, and a developer can easily construct multi-stage nested logic in a visual drag mode. Taking the financial industry as an example, in the loan approval process, a plurality of links such as credit rating, loan amount calculation, risk assessment and the like are involved, and each link possibly comprises different conditional branches and sub-processes. Under the JSON code editing mode, a developer can realize complex rule nesting and deep logic customization, high-end requirements such as intelligent contract execution, complex business algorithm realization and the like are met, and the aim of dealing with various complex business scenes without depending on a large amount of custom codes is really achieved. The core logic of the complex flow arrangement is realized by (a) multistage conditional branch logic branch trigger logic:

1. The process is provided with a condition judgment node which presets a plurality of output ports (for example, 3 ports correspond to 3 client grades).

2. Each port binds an independent judgment rule (such as 'credit score not less than 800', '600 not less than credit score < 800', 'credit score < 600').

3. When the node executes, according to the matching rule of the input data (such as credit score), only the output ports meeting the conditions are activated, and the data is transferred to the corresponding downstream nodes (such as high-quality customer branches and common customer branches).

Nested branch logic

1. The first branch (such as a common customer branch) can be inserted with a condition judgment node again (such as judging whether the approval limit is more than or equal to 10 ten thousand) to form a second branch.

2. The triggering rule of the child branch can depend on the output data of the parent branch (such as 'approval limit' generated by 'limit computing node' of the parent branch), so as to realize hierarchical transfer and logical nesting of the data.

(Two) circular structure logic

The process is provided with a circulation control node, and two core conditions are defined, namely 1. When a circulation continuation condition (such as 'material supplement times < 3') is met, the nodes in the circulation body (such as 'material supplement and review') are repeatedly executed.

2. When the loop exit condition (such as 'supplementing material times more than or equal to 3') is met, the loop is terminated, and the data is transferred to an exit node (such as 'refusal notice').

3. When the node in the circulation body executes successfully (such as 'review pass'), the rest circulation can be skipped directly, and the data is transmitted to the flow end point (such as 'pay').

Loop count logic

1. The circulating node automatically maintains a circulating times counter of +1 after the node in the circulating body is executed each time.

2. The counter data is used as an input parameter to participate in condition judgment, so that the controllable number of circulation times (such as not more than 3 times) is ensured.

Collaborative logic visualization drag mode logic for two modes (III)

1. Dragging the condition judging node and the circulating control node to canvas, and defining the upstream and downstream relation between the nodes through connecting lines.

2. The system automatically converts the rules into executable logic (without manual coding) by double clicking on the node configuration rules (e.g., entering a "credit score ≡800").

3. The system verifies the link rationality in real time (e.g., avoiding loop dependencies) to ensure that the flow logic is executable.

JSON code mode logic 1 to structure JSON description flow, defining node attribute (including judgment rule and circulation condition) with 'nodes' array, and defining data flow direction between nodes with 'flows' array.

2. Supporting nested expressions (e.g., "if (score > 0.8) then pass") may directly reference the output data of other nodes (e.g., "approval limit").

3. When the system analyzes the JSON, the association relation between the rule grammar and the node is automatically checked, and if the rule grammar and the node association relation are wrong, a specific position is prompted (if 'node ID does not exist').

(IV) integral execution logic

1. After the flow is started, the method is sequentially executed according to the node connection sequence, and the initial node (such as 'receiving application') is executed first, and then the output data is transmitted to the next node (such as 'credit inquiry').

2. When encountering the condition judgment node, only branches conforming to the rule are triggered, and other branches are not executed temporarily.

3. And repeatedly executing the nodes in the circulation body when encountering the circulation control node until the exit condition is met or the nodes in the circulation body are successfully executed.

4. After the end nodes of all branches are executed, the process ends (such as "paying money" or "refusing notice").

Data standardization and compatibility guarantee, namely starting from the node configuration stage, the invention strictly and uniformly defines the input and output data model through a standardization mechanism. When data transmission and interaction are carried out between nodes, the system can automatically complete data mapping and format conversion, and the problem of inconsistent data models between components in the prior art is effectively solved. For example, in a business process including data collection, cleaning and analysis, the requirements of different nodes on the data format may be different, but by the data standardization mechanism of the present invention, smooth data circulation between the nodes can be ensured, and logic errors caused by manually processing data conversion can be avoided. From field naming standards to data type verification, the whole flow realizes comprehensive standardized management and control, and the stability and compatibility of system data flow are greatly improved.

Standardized mechanism of node configuration phase:

1. The definition method system of the input and output data model supports the input and output data model of the custom node through the visual interface. The user can flexibly configure the metadata such as parameter names, data types, nested structures, default values, imperfection and the like of each node.

And all node parameters adopt uniform structural description, support multi-layer nesting and type labeling, and facilitate multiplexing and maintenance of a data model. The custom data model specification provides a basic guarantee for subsequent data flow and integration between nodes.

2. The data mapping and format conversion algorithm integrates a flexible data mapping and format conversion mechanism for realizing seamless flow and format compatibility of data among nodes. The data mapping relation among the nodes can be defined by an expression, a mapping table or a visual configuration mode, and static mapping and dynamic mapping are supported. The system can automatically complete the conversion of data types (such as character strings, integers, boolean, objects, arrays and the like) according to the definition of the node parameters, automatically analyze parameter references in the data circulation process, and realize the dynamic acquisition and transmission of data. For complex data structures, the system supports recursive parsing and nesting processes, ensuring consistency and correctness of data formats.

3. Field naming convention and data type checking mechanism

In order to ensure the standardability and maintainability of the data model, the system introduces a data type automatic conversion and compatible processing mechanism in the node configuration stage. The system can automatically convert the input data into a target type according to the parameter definition, and perform fault-tolerant processing or throw exception when the types are incompatible. In terms of field naming, the system encourages users to follow a uniform naming convention (e.g., small humps, underlines, etc.), and naming convention prompts can be made during the parameter configuration and preservation phases. Through the measures, the system effectively improves the normalization and consistency of the data model, and reduces the integration risk caused by inconsistent naming or types.

4. The standardized mechanism of the technical integrity and credibility guarantee system covers the whole processes of data model definition, data mapping, format conversion, naming standards, type verification and the like. Through unified data structure description, flexible data mapping mechanism and automatic data type processing, the system can effectively improve data consistency and verifiability in a node configuration stage, and provides a solid technology base multi-port adaptive API automatic generation technology for flow automation and inter-node integration. The developer can flexibly adjust key elements such as interface paths, request modes (such as GET, POST, PUT and the like), return formats (such as JSON, XML and the like) and the like by simple parameter configuration. The technology enables the same set of node processes to be easily adapted to various front-end application scenarios, including but not limited to web page end, mobile end application and third party system docking. Compared with the traditional mode that multiple ports are required to be independently developed, the workload of cross-platform adaptation is obviously reduced, the collaboration efficiency of front and back end teams is greatly improved, and the efficient development targets of one-time configuration and multiple-port calling are truly realized.

The technical implementation of automatic generation and release of APIs (based on canvas release flow) (I) automatic extraction and generation logic of API metadata is based on the node configuration and flow relation of canvas, the system automatically generates core information of APIs through a canvas analysis engine without manual coding, and the specific algorithm is as follows, basic information extraction (user configuration+automatic generation of system) 1. The API path is associated with items:

In the release phase, a user selects an affiliated item (such as a financial system and a CRM system) through a release configuration panel and fills out a path prefix (such as a/api/v 1/finish), and the system automatically splices canvas unique identifiers to generate a complete path (such as a/api/v 1/finish/"). The system will check the path uniqueness in real time, and if a conflict is found, prompt the user to adjust (e.g. "path is occupied, please modify").

2. Request mode definition:

The type of the initial node in the canvas determines the request mode, namely, if the initial node is a data submitting node, the initial POST is defaulted, if the initial node is a data inquiring node, the initial GET is defaulted, a user can manually modify the initial node (supporting PUT/DELETE) on a release panel, and the system automatically checks the combination uniqueness of the path and the request mode.

The method comprises the steps of1, extracting request parameters, namely analyzing an input model of a canvas initial node, automatically converting fields in the model into API request parameters 2, and if the initial node input model comprises identification fields such as id, code and the like, selecting the path parameters by a user on a release panel, and automatically adding placeholders in a path by a system.

3. Query parameters/requester parameters-non-path parameters automatically distinguish types-simple types (character strings/numbers) are set as query parameters by default, complex types (objects/arrays) are automatically set as requester parameters (JSON format), and input model check rules (such as necessary padding and length limitation) of the initial node are associated.

4. The response format is generated by analyzing an output model of the canvas end node and automatically generating an API response structure:

5. The success response comprises a code of 200, a data field (an output model field corresponding to the end node) and a message of success.

6. The error response comprises a code (error code, such as 400/500), a message (error description) and traceId (tracking ID, used for troubleshooting the problem), wherein the error code is associated with the exception type of the node in the canvas (such as parameter verification failure corresponds to 400 and node execution exception corresponds to 500).

Visual interface and function of API parameter configuration

The publishing stage provides an API parameter configuration workbench, supports fine adjustment of parameter attributes, adapts to different front-end scenes, and has the core functions of parameter types and constraint configuration

1. The basic type is character string (configurable regular check, such as mobile phone number format), number (configurable range, such as 0< value < 1000), boolean value, date (supporting format selection, such as yyyy-MM-dd).

Complex types-objects (nested structures, associable solid models in canvas), arrays (supporting element type restrictions, such as "string arrays" and "number arrays").

2. Parameter association rules:

Supporting configuration "conditional dependencies" (e.g., "when type=1, the detail parameter must be filled out"), the system automatically generates check logic by visualizing the form selection trigger conditions and dependent parameters.

Multi-scene adaptation configuration

1. Front end type adaptation:

the differentiating parameters can be configured for a webpage end, a mobile end and a third party system, wherein the webpage end reserves all parameters by default and supports CORS cross-domain configuration (such as allowing https:// xxx. Com domain name access).

3. Support parameter reduction (conceal unnecessary fields such as create_time), data compression (auto-enable gzip).

4. Third party systems-forced open signature verification-third party systems may be of a variety, such as other business systems-business systems within an enterprise where there may be multiple different uses, such as financial management systems, human resource management systems, etc., that may require data interaction and business process collaboration with the orchestration engine. Or a partner system, namely a system owned and operated by the partner is a third party system when the enterprise performs business cooperation with an external partner. For example, an e-commerce enterprise interfaces with a system of a logistics provider, which is a third party system. But also public service systems: some systems that provide public services, such as government systems, financial institution payment systems, etc., also belong to third party systems when the orchestration engine needs to exchange data with it or invoke its services.

5. Version control, which supports the creation of API versions (such as v1/v 2), the new version can inherit the parameter configuration of the old version and modify the parameter configuration, and the influence on the stock calling party is avoided.

And (III) after the function realization of the API management page is successfully released, the system provides a unified API management page, supports full life cycle management, and has the following core functions:

List presentation and screening

1. The fields of API path, request mode, belonged item, creation time, state (enabled/disabled), call times and average response time.

2. And the screening function is used for supporting screening according to items, path keywords, states and creation time ranges and rapidly positioning target APIs.

API details and configuration modification 1. Basic information is to view/modify API paths, request modes, belonged projects and description information.

2. And (3) parameter configuration, namely visually editing request parameters and response formats, and supporting adding, deleting and checking fields and adjusting constraint rules.

3. Advanced configuration-modifying the CORS rules, current limiting policies, timeout times, etc., without reissuing the canvas.

Lifecycle management

1. State control-support enable/disable API, call return 403Forbidden after disable.

2. Version management, namely checking historical versions, rolling back to old versions and comparing version differences.

3. And the deleting function is to confirm the association relation (if the front-end application call exists) when deleting the API, and support the maintenance of the history record.

Monitoring and operation

1. And (3) monitoring in real time, namely displaying indexes such as QPS, response time, error rate and the like of the API, and supporting chart visualization.

2. And the call log is used for inquiring a history call record of the API and comprises information such as request parameters, response results, call time, caller IP and the like.

3. And (3) alarm configuration, namely setting a threshold value for error rate and response time, and automatically triggering mail/short message alarm when the error rate and response time are abnormal.

The method for constructing and operating the system by the Agent based on the AI large model technology comprises the following steps of firstly, receiving node parameter definition and outputting a standardized component unit through an interactive node configuration system;

Secondly, through a visual flow development system, a component referencing mechanism is adopted to call the generated standardized component unit, so that the combination arrangement of the business flow is realized;

thirdly, the components of the interactive node configuration system are modified, all flow instances referring to the components are automatically traced, and modified contents are synchronized to the associated flow, so that dynamic update synchronization is realized;

The fourth, interactive node configuration system adopts interactive closed loop design in the node configuration stage to realize real-time linkage of parameter configuration and component preview;

fifth, the API interface adapting to the different terminal requirements is automatically generated and issued.

The design key point of the invention is that the first and the interactive node configuration systems give consideration to flexibility and usability, adapt to the development requirement of the whole scene and break through the limitation of the single configuration mode of the existing low-code platform;

The second and layered development systems construct standardized multiplexing ecology, so that development cost is greatly reduced, an interactive node configuration system and a visual flow development system are adopted to respectively correspond to two-layer architecture of node design and flow arrangement, a node design layer outputs standardized assembly units and can be used for multiplexing the flows, a flow arrangement layer forms complete business logic through dragging or code combination nodes, the bottom layer node configuration can be automatically synchronized to all reference flows after modification, the problem of poor code logic multiplexing in traditional development is thoroughly solved, compared with the prior art, the component multiplexing rate is remarkably improved, the system maintenance cost is greatly reduced, and the method is particularly suitable for large-scale iteration of enterprise-level application.

Thirdly, the complex flow arranging capability breaks through the traditional limitation, adapts to multi-industry service scenes, is different from the limitation that the traditional platform only supports linear flows, and adopts a flow development mode that the multistage nested logic can be quickly constructed through visual dragging;

Fourth, the data standardization mechanism guarantees the system stability, avoids the logic fault, namely, the input and output data model is uniformly defined through the standardization mechanism in the node configuration stage, and automatically completes the data mapping and format conversion in the process development, so that the problem of non-uniform data model among components in the prior art is solved, the logic error caused by manually processing the data conversion is avoided, the whole process standardization management and control is formed from field naming standards to data type verification, and the stability and compatibility of the system data flow are greatly improved.

Fifth, the multi-port adaptation capability realizes one-time configuration and multi-scenario application, and improves the cooperation efficiency, wherein based on the configured node flow, the system can automatically generate API interfaces adapting to different requirements;

compared with the traditional mode that the multi-port interface needs to be independently developed, the cross-platform adaptation workload is obviously reduced, and the front-end team and back-end team cooperation efficiency is improved.

The foregoing description is only a preferred embodiment of the present invention, and is not intended to limit the technical scope of the present invention, so any minor modifications, equivalent changes and modifications made to the above embodiments according to the technical principles of the present invention still fall within the scope of the technical solutions of the present invention.

Claims (10)

1. The Agent constructing and operating system based on the AI large model technology is characterized by comprising an interactive node configuration system and a visual flow development system, wherein the interactive node configuration system and the visual flow development system are in a layered development system, the interactive node configuration system outputs standardized component units, the visual flow development system realizes the combined arrangement of service flows through a component quotation mechanism, the interactive node configuration system and the visual flow development system are provided with dynamic update links, the interactive node configuration system and the visual flow development system adopt interactive closed-loop design, and delivery forms adapting to different requirements are automatically generated and released based on the configured node flows.

2. The Agent construction and operation system based on the AI large model technology as set forth in claim 1, wherein the interactive node configuration system adopts a split screen type interactive architecture, one side of the split screen type interactive architecture is a parameter configuration area, the other side is a real-time rendering area, the visual flow development system comprises a bidirectional editing core module for supporting canvas editing and code editing of a workflow, the bidirectional editing core module comprises a canvas editing mode and a code editing mode, the bidirectional editing core module supports dragging into a pre-configuration component from a node library, a service flow is constructed through canvas connection by the canvas editing mode, and structured data of the flow is directly edited by the code editing mode to define node logic relations and data flow rules.

3. The Agent construction and operation system based on the AI large model technology of claim 1, wherein the dynamic update link comprises a change tracing mechanism and a guarantee data consistency, the change tracing mechanism is realized by adopting a synchronous algorithm, the guarantee data consistency is specifically that operation atomicity is single modification as an independent transaction, a state snapshot is used for storing a configuration version before modification, conflict processing is carried out, a priority principle is finally modified, components in the visualization flow development system can trace to an interactive node configuration system, and after the bottom configuration is modified, the components are automatically synchronized to all flow instances referring to the components, so that the centralized management and global effect of configuration are realized.

4. The Agent construction and operation system based on the AI large model technology of claim 2, wherein the interactive closed loop design is realized through mode switching logic and seamless connection, scene judgment is performed by adopting the mode switching logic, the parameter batch modification, automatic switching of split screen mode, complex logic editing, suggesting switching of code mode and user habit analysis through operation history are performed, the seamless connection comprises shared memory data pool, view state serialization storage and automatic positioning of focus elements, the split screen type interactive architecture in the interactive node configuration system is complementary with a bidirectional editing core module in the visual flow development system, and a developer is supported to conveniently cover the basic configuration to the complex logic development according to a scene selection operation mode.

5. The Agent construction and operation system based on the AI large model technique as set forth in claim 1, wherein the standardized component unit is normalized by node classification including input nodes, processing nodes, and output nodes, parameter specification including input parameters including a fill-in field flag and a basic type check, and output parameters being structured data templates.

6. The Agent construction and operation system based on AI large model technique as set forth in claim 1, wherein said delivery modality comprises an API interface, an Agent (Agent) and an AI model tool (MCP), wherein the generation of the API interface comprises an interface configuration node and a terminal adaptation node, wherein the terminal adaptation node comprises a web page end adaptation, a mobile end adaptation and a third party interfacing, wherein the third party interfacing comprises a security protocol encapsulation and a message format conversion.

7. The Agent construction and operation system based on the AI large model technique of claim 6, wherein the API interface generation workflow comprises a configuration stage, a generation stage and a debugging stage, wherein the configuration stage comprises a step of dragging an interface configuration node by a developer and setting basic parameters, the generation stage comprises a step of automatically connecting a system with a terminal adaptation node to generate three sets of interface schemes in parallel, and the debugging stage comprises a step of independently testing nodes of each terminal and automatically positioning problem feedback to corresponding nodes.

8. The Agent construction and operation system based on the AI large model technique of claim 2, wherein said interactive node configuration system generates a standardized component unit including a condition judgment node and a circulation control node, and said visual process development system realizes a multi-level nested process arrangement for complex process arrangement by referring to said standardized component unit.

9. The Agent construction and operation system based on the AI large model technique of claim 8, wherein the condition judgment node generated by the interactive node configuration system comprises a plurality of output ports for binding independent rule expressions, and the visual process development system realizes multi-stage nested process arrangement by dragging or code referencing the node.

10. A method for constructing and operating a system based on an Agent of AI large model technique as set forth in any one of claims 1-9, comprising the steps of first, receiving node parameter definition and outputting standardized component unit by an interactive node configuration system;

Secondly, through a visual flow development system, a component referencing mechanism is adopted to call the generated standardized component unit, so that the combination arrangement of the business flow is realized;

thirdly, the components of the interactive node configuration system are modified, all flow instances referring to the components are automatically traced, and modified contents are synchronized to the associated flow, so that dynamic update synchronization is realized;

The fourth, interactive node configuration system adopts interactive closed loop design in the node configuration stage to realize real-time linkage of parameter configuration and component preview;

Fifth, automatically generating and releasing delivery forms adapting to different terminal requirements.