Intelligent agent collaborative execution system based on task dialogue link, electronic device

Abstract

This application relates to the field of artificial intelligence and discloses an intelligent agent collaborative execution system and electronic device based on a task dialogue link. The system includes: an intelligent agent workbench; an intelligent agent management console; a task scheduling and execution module, used to generate corresponding dialogue nodes each time a task scheduling is triggered during the execution of a periodic task by the intelligent agent, with multiple dialogue nodes corresponding to the periodic task forming a dialogue link in chronological order; an intelligent agent access and collaboration interface module, used to realize the interaction between the intelligent agent collaborative execution system and the intelligent agent; and an intelligent agent workbench including a task management and execution viewing module, used to respond to periodic task execution viewing requests and display the execution process of the periodic task in the form of a dialogue link. The dialogue link, formed by the orderly connection of dialogue nodes, provides a unified and presentable data organization format for the execution process of the periodic task, which is beneficial for the backtracking, auditing, and reuse of the execution process.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an intelligent agent collaborative execution system and electronic device based on a task dialogue link. Background Technology

[0002] With the development of artificial intelligence technology, enterprises have begun to apply intelligent agents to business scenarios such as email processing, inspection, report generation, and work order workflow. In email processing scenarios, it is typically necessary to configure intelligent agents to perform tasks periodically, such as processing new emails in the inbox every other day. However, traditional intelligent agent scheduling systems suffer from difficulties in backtracking, auditing, and reusing each round of execution when scheduling agents to perform periodic tasks. Summary of the Invention

[0003] The purpose of this application is to provide at least one intelligent agent collaborative execution system and electronic device based on task dialogue links, in order to solve one of the aforementioned technical problems.

[0004] To address the aforementioned technical problems, at least one embodiment of this application provides an intelligent agent collaborative execution system based on a task dialogue link, comprising: an intelligent agent workbench for providing a unified human-computer interaction entry point to users; an intelligent agent management platform for unified management of intelligent agents; a task scheduling and execution module for receiving registration and deregistration of periodic tasks, scheduling intelligent agents to execute registered periodic tasks, and generating corresponding dialogue nodes each time task scheduling is triggered during the execution of periodic tasks by the intelligent agent, wherein multiple dialogue nodes corresponding to periodic tasks constitute a dialogue link in chronological order; and an intelligent agent access and collaboration interface module for realizing the interaction between the intelligent agent collaborative execution system and the intelligent agents; the intelligent agent workbench includes a task management and execution viewing module and a human-computer interaction and notification module, wherein the task management and execution viewing module is used to create periodic tasks in response to periodic task creation requests, register periodic tasks to the task scheduling and execution module, and receive and deregister periodic tasks to the task scheduling and execution module based on periodic task termination instructions; The system also responds to requests to view the execution of periodic tasks by displaying the execution process of the periodic tasks in the form of a dialogue link. The human-computer interaction and notification module responds to requests for human intervention during the execution of periodic tasks by the agent and issues a human intervention notification based on a preset method. The agent collaborative execution system also employs a human-computer collaboration and scheduling linkage mechanism. This mechanism works as follows: during the execution of any periodic task, if the agent determines that the execution input of the current round of tasks does not meet the automatic processing conditions of the agent corresponding to that periodic task, the agent sends a request for human intervention to the human-computer interaction and notification module and sends a request to the task scheduling and execution module to suspend the scheduling of that periodic task. Upon completion of human intervention, the agent sends a request to the task scheduling and execution module to resume the scheduling of that periodic task. The request for human intervention corresponding to the current round of tasks includes the identifier of the dialogue node corresponding to the current round of tasks, and the human intervention notification corresponding to the current round of tasks includes the identifier.

[0005] By generating corresponding dialogue nodes each time a task is scheduled, and by using multiple dialogue nodes arranged chronologically to form a dialogue chain, the discrete task execution events traditionally recorded in log form are transformed into a dialogue chain formed by the orderly connection of structured dialogue nodes. This gives the execution process of periodic tasks a unified and presentable data organization format. By responding to requests to view the execution of periodic tasks and displaying the execution process of periodic tasks in the form of a dialogue chain, users can be directly given the ability to view the entire execution history of tasks through a unified entry point. This allows users to intuitively understand the time, order, and relationships of each execution without having to search through scattered log files or status tables, which is beneficial for backtracking, auditing, and reusing the execution process.

[0006] Furthermore, by generating a corresponding dialog node each time a task is scheduled, each independent scheduling event can be bound to a dedicated dialog node. This binding relationship is established when the scheduling is triggered, so that subsequent execution inputs, outputs, and state changes can all be recorded in this node. This avoids information gaps caused by the separation of scheduling events and execution records, and further facilitates the backtracking, auditing, and reuse of the execution process.

[0007] Furthermore, the system limits the human-computer interaction and notification module to responding to requests for human intervention, issuing notifications based on a preset method. When the agent requires human judgment during the execution of a cyclical task, the system can proactively and specifically push the intervention request to relevant personnel, rather than relying on the user to actively poll the task status. This mechanism shortens the delay time for human response and improves the timeliness of human-computer collaboration.

[0008] Furthermore, through a three-stage linkage mechanism of pausing scheduling, manual processing, and resuming scheduling, manual intervention is incorporated into the task scheduling control loop, rather than being independent of the scheduling system. This prevents periodic tasks from blindly continuing to execute new cycles after triggering manual intervention, avoiding state confusion or processing conflicts caused by the influx of new task rounds during manual processing. Additionally, the system automatically resumes cycle execution after manual intervention, eliminating the need for users to manually restart tasks, thus forming a seamless closed loop. Both the invitation to manual intervention and the notification for manual intervention include the identifier of the current round of dialogue nodes, ensuring that the pause / resumption actions of the task scheduling and execution modules precisely correspond to specific task execution rounds. This allows manual intervention to quickly locate the task that needs to be intervened based on this identifier, eliminating the need to drill down through the task list to find the execution record to be processed. This shortens the time consumed between receiving the notification and entering the processing interface, reducing context switching costs.

[0009] In some optional embodiments, the agent workbench further includes: an agent portal and details display module, used to display agent information of agents that have been listed, the agent information including agent name, capability description, supported task types and running status; and an instant dialogue module, used to support users to interact with agents in real time.

[0010] The intelligent agent portal and details display module centrally present the names, capability descriptions, task types, and operational status of all intelligent agents listed on the platform. Before creating recurring tasks, users can fully understand the applicable scenarios and current availability of each intelligent agent in this module, eliminating the need for external documentation or individual trials, thus lowering the barrier to entry for intelligent agents. A separate real-time dialogue module supports non-recurring, non-task-based real-time interaction between users and intelligent agents, clearly distinguishing it from the task management and execution viewing modules. This avoids mixing recurring task execution dialogues with temporary auxiliary dialogues, providing users with a clear and focused interactive interface regardless of their intended use.

[0011] In some optional embodiments, the intelligent agent management platform includes: an intelligent agent access and listing management module, used to access intelligent agents built on external platforms and to manage the listing, delisting, and versioning of intelligent agents; a permission and usage scope management module, used to configure the permissions for different objects to use intelligent agents; a platform configuration and operation management module, used to configure the system parameters of the intelligent agent collaborative execution system; and an operation record and operation analysis module, used to record the usage information, task execution frequency, and human intervention information of intelligent agents.

[0012] The intelligent agent access and deployment management module enables intelligent agents built with different external frameworks and technology stacks to be connected to this platform in a standardized manner, and performs governance operations such as deployment, decommissioning, and version management. The permission and usage scope management module configures the permissions for different objects to use intelligent agents, allowing the system to control the visibility and availability of specific intelligent agents based on dimensions such as user, department, and business scope, preventing unauthorized access or cross-business calls, and meeting enterprise-level security and compliance requirements.

[0013] In some optional embodiments, the dialogue node includes: execution input, execution result, and execution processing procedure.

[0014] The dialogue node is defined to include three types of information: execution input, execution result, and execution processing procedure. Unlike traditional task logs that only record the success / failure status, this approach binds and stores the task's triggering conditions (input), the agent's processing behavior (processing procedure), and the final output (result) within the same node. This allows a user or system to obtain a complete picture of the execution at once when accessing any node, without the need for cross-table joins or secondary inferences. By clearly defining the information elements that a dialogue node should contain, each dialogue node in the dialogue chain becomes an independently auditable complete unit. In compliance reviews or troubleshooting scenarios, reviewers do not need to reconstruct the execution process; they can directly read the "execution processing procedure" within the dialogue node to know the agent's specific operation path in that round, significantly improving traceability efficiency.

[0015] In some optional embodiments, the periodic task creation request includes agent-automatic processing conditions; the agent-cooperative execution system also applies an active reasoning and collaborative recommendation mechanism based on the periodic task dialogue link; the active reasoning and collaborative recommendation mechanism based on the task dialogue link is as follows: during the execution of any periodic task, if it is determined that the execution input of the current round of the task does not meet the agent-automatic processing conditions corresponding to the periodic task, the agent generates processing suggestion information for the current round of the task based on the dialogue link corresponding to the current round of the task; the execution processing of the dialogue node corresponding to the current round of the task includes the processing suggestion information; wherein, the processing suggestion information includes at least one candidate solution information, and the candidate solution information includes a candidate processing solution and a corresponding processing description and a summary of the expected execution result.

[0016] When an agent encounters a scenario that cannot be handled automatically, it proactively generates processing suggestions based on the historical dialogue chain of the task. This enables the agent to invoke past execution experience, transforming historical experience into reusable candidate solutions, thus improving processing efficiency after human intervention. The processing suggestions explicitly include candidate processing solutions, processing instructions, and a summary of expected execution results. This allows human operators to directly access multiple already reasoned alternative paths with expected consequences when entering a task dialogue node, eliminating the need to manually search historical records or analyze the problem's causes from scratch, effectively shortening the processing time for each human intervention. The generated processing suggestions are written into the execution process of the current dialogue node, ensuring that the agent's reasoning output and candidate solutions are no longer temporary messages independent of the task execution chain, but rather structured node content deeply bound to the current execution. Subsequent agents can further reason based on this node content, forming a continuous transfer of collaborative experience.

[0017] In some optional embodiments, the agent collaborative execution system also applies an adaptive collaboration strategy mechanism based on human behavior; the adaptive collaboration strategy mechanism based on human behavior is as follows: updating the agent automatic processing conditions corresponding to any cycle task based on the user operation behavior, so as to adjust whether subsequent round tasks meet the agent automatic processing conditions; and adjusting the sorting of candidate scheme information when the other round tasks do not meet the adjusted agent automatic processing conditions, so that when the other round tasks and the current round tasks are matched based on similar features, the other round tasks are automatically processed by the agent.

[0018] In some optional embodiments, the intelligent agent collaborative execution system also applies a dialogue node update mechanism based on human behavior; the dialogue node update mechanism based on human behavior is as follows: record the human operation behavior of inviting human participation request corresponding to the current round task, and update the operation behavior in the dialogue node corresponding to the current round task.

[0019] By recording the specific actions of each round of human intervention, the system can collect high-value feedback data, such as the specific handling solutions adopted by humans in specific scenarios, without intruding on the internal logic of the intelligent agent. The human actions are updated in the current round of dialogue nodes, ensuring that human choices, modifications, and confirmations are no longer confined to external logs or temporary session records, but become structured data strongly associated with the execution nodes in the task's dialogue chain. Information such as the handling solutions and decision preferences of each round of human intervention is precisely attached to the dialogue node that triggered the intervention, forming a complete chain of evidence. Furthermore, by using dialogue nodes to record human actions, subsequent scenarios requiring human intervention can be addressed by providing more suitable candidate handling solutions based on the records in the dialogue nodes.

[0020] At least one embodiment of this application also provides an electronic device configured with the intelligent agent cooperative execution system based on task dialogue link as described in any of the above embodiments.

[0021] At least one embodiment of this application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the intelligent agent cooperative execution system based on task dialogue links as described in any of the above embodiments.

[0022] At least one embodiment of this application also provides a computer program product, including a computer program that, when executed by a processor, implements the intelligent agent cooperative execution system based on task dialogue links as described in any of the above embodiments. Attached Figure Description

[0023] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.

[0024] Figure 1 This is a schematic diagram of an intelligent agent cooperative execution system based on a task dialogue link provided in one embodiment of this application;

[0025] Figure 2 This is a schematic diagram of an intelligent agent workbench provided in one embodiment of this application;

[0026] Figure 3This is a schematic diagram of an intelligent agent management console provided in one embodiment of this application;

[0027] Figure 4 This is a flowchart illustrating the usage process of an intelligent agent collaborative execution system based on a task dialogue link, as provided in one embodiment of this application. Detailed Implementation

[0028] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0029] Furthermore, the accompanying drawings are merely illustrative of this application and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0030] To facilitate understanding of the embodiments of this application, we will first introduce the relevant content of the intelligent agent cooperative execution system based on task dialogue link.

[0031] With the development of large language models and intelligent agent technology, enterprises have begun to apply intelligent agents to business scenarios such as email processing, inspection, report generation, and work order workflow. However, in practical applications, existing technical solutions still struggle to support the long-term, stable operation of intelligent agents as "digital employees," mainly in the following aspects.

[0032] First, intelligent agents are scattered in origin and diverse in form, lacking a unified access point and governance system. Existing systems often can only manage intelligent agents within a single platform or framework, making it difficult to achieve unified deployment, unified access control, and unified usage management, thus hindering the large-scale reuse of intelligent agent capabilities across enterprises.

[0033] Secondly, existing intelligent agents mostly remain at the level of instant dialogue or single execution, lacking the ability to continuously execute tasks for periodic periods. Although traditional task scheduling systems can trigger execution periodically, they usually only record simple execution status or logs, failing to express the contextual changes of the intelligent agent in multiple rounds of execution, making it difficult to trace, audit, and reuse the task execution process.

[0034] Secondly, there are inevitably scenarios during task execution that require human judgment. In existing technologies, once human intervention is introduced, the automated process is often forced to stop, or there is a lack of effective connection between manual processing and automatic execution, making it difficult to safely and naturally resume automatic execution after manual processing is completed.

[0035] Furthermore, existing systems typically only provide current execution information when human intervention occurs, lacking a mechanism for providing processing suggestions based on historical execution processes. Humans must then determine the appropriate course of action from scratch. Simultaneously, the results and choices made by humans are difficult for the system to effectively utilize, failing to optimize subsequent task execution or reduce the frequency of human intervention in similar scenarios. This results in human-machine collaboration efficiency remaining at a consistently low level.

[0036] Furthermore, existing intelligent agents and platforms mostly use unstructured natural language interaction, lacking standardized human-machine collaboration semantics. This makes it difficult for the platform to reliably judge task execution results, human intervention status, and scheduling and control requirements, thus affecting the system's controllability and manageability.

[0037] In summary, existing technologies have significant shortcomings in areas such as continuous execution of intelligent agents, task process representation, human-machine collaborative recommendation and learning, and platform-level governance, making it difficult to meet the needs of enterprise-level scenarios for long-term operation, secure collaboration, and continuous optimization of digital employees.

[0038] To address the aforementioned technical problems, this application proposes an intelligent agent collaborative execution system based on task dialogue links. The implementation details of the intelligent agent collaborative execution system based on task dialogue links in this embodiment are described below. The following implementation details are provided for ease of understanding and are not essential for implementing this solution.

[0039] Example 1:

[0040] The intelligent agent collaborative execution system based on task dialogue links in this embodiment can, as follows: Figure 1 As shown, it includes: an intelligent agent workbench 11, an intelligent agent management console 12, a task scheduling and execution module 13, and an intelligent agent access and collaboration interface module 14.

[0041] Among them, the intelligent agent workbench 11 is used to provide users with a unified human-computer interaction interface.

[0042] The intelligent agent management console 12 is used for unified management of intelligent agents.

[0043] The task scheduling and execution module 13 is used to receive the registration and deregistration of periodic tasks, and to schedule the agent to execute the registered periodic tasks. During the execution of the periodic tasks by the agent, a corresponding dialogue node is generated each time the task scheduling is triggered. Multiple dialogue nodes corresponding to the periodic tasks form a dialogue link in chronological order.

[0044] Task scheduling triggering refers to the system automatically starting the execution action when a periodic task reaches a preset execution time.

[0045] Each dialogue node has a corresponding identifier, and different dialogue nodes have different identifiers. In some examples, for dialogue nodes corresponding to tasks in different rounds under the same periodic task, the identifiers can have the same identifier root and different identifier numbers. For example, the identifiers of dialogue nodes corresponding to two rounds of tasks in the same periodic task are pt1_2 and pt1_3, respectively, where pt1 is the identifier root and 2 and 3 are different identifier numbers.

[0046] In some examples, the identifiers of the dialogue nodes corresponding to different rounds of the same periodic task are sequentially incremented. For example, a periodic task may have 5 dialogue nodes, identified as pt1_1, pt1_2, pt1_3, pt1_4, and pt1_5.

[0047] The agent access and collaboration interface module 14 is used to realize the interaction between the agent collaborative execution system and the agent.

[0048] The intelligent agent workbench 11 includes a task management and execution viewing module 111 and a human-computer interaction and notification module 114. The task management and execution viewing module 111 is used to create periodic tasks in response to periodic task creation requests and register them with the task scheduling and execution module 13; to receive and deregister periodic tasks with the task scheduling and execution module 13 based on periodic task termination instructions; and to display the execution process of periodic tasks in the form of a dialogue link in response to periodic task execution viewing requests. The human-computer interaction and notification module 114 is used to issue human intervention notifications based on a preset method in response to human intervention requests received during the execution of periodic tasks by the intelligent agent.

[0049] By generating corresponding dialogue nodes each time a task is scheduled, and by using multiple dialogue nodes arranged chronologically to form a dialogue chain, the discrete task execution events traditionally recorded in log form are transformed into a dialogue chain formed by the orderly connection of structured dialogue nodes. This gives the execution process of periodic tasks a unified and presentable data organization format. By responding to requests to view the execution of periodic tasks and displaying the execution process of periodic tasks in the form of a dialogue chain, users can be directly given the ability to view the entire execution history of tasks through a unified entry point. This allows users to intuitively understand the time, order, and relationships of each execution without having to search through scattered log files or status tables, which is beneficial for backtracking, auditing, and reusing the execution process.

[0050] Furthermore, by generating a corresponding dialog node each time a task is scheduled, each independent scheduling event can be bound to a dedicated dialog node. This binding relationship is established when the scheduling is triggered, so that subsequent execution inputs, outputs, and state changes can all be recorded in this node. This avoids information gaps caused by the separation of scheduling events and execution records, and further facilitates the backtracking, auditing, and reuse of the execution process.

[0051] Furthermore, the system limits the human-computer interaction and notification module to responding to requests for human intervention, issuing notifications based on a preset method. When the agent requires human judgment during the execution of a cyclical task, the system can proactively and specifically push the intervention request to relevant personnel, rather than relying on the user to actively poll the task status. This mechanism shortens the delay time for human response and improves the timeliness of human-computer collaboration.

[0052] In some cases, intelligent agent collaborative execution systems also employ human-machine collaboration and scheduling linkage mechanisms.

[0053] The human-machine collaboration and scheduling linkage mechanism is as follows: When the intelligent agent determines that the execution input of the current round of the task does not meet the automatic processing conditions of the intelligent agent corresponding to the task of any cycle during the execution of any cycle, the intelligent agent sends a request to the human-machine interaction and notification module to invite human participation, and sends a request to the task scheduling and execution module to suspend the scheduling of any cycle task; in response to the completion of human processing, the intelligent agent sends a request to the task scheduling and execution module to resume the scheduling of any cycle task.

[0054] The request for human intervention corresponding to the current round of tasks includes the identifier of the dialogue node corresponding to the current round of tasks, and the notification for human intervention corresponding to the current round of tasks includes the identifier.

[0055] After receiving a request for human intervention from the intelligent agent, the human-computer interaction and notification module generates a notification for human intervention carrying the identifier of the dialogue node in the request and sends the notification to invite human intervention.

[0056] By employing a three-stage linkage mechanism of pausing scheduling, manual processing, and resuming scheduling, manual intervention is incorporated into the task scheduling control loop, rather than operating independently of the scheduling system. This prevents periodic tasks from blindly continuing to execute new cycles after manual intervention is triggered, avoiding state confusion or processing conflicts caused by the influx of new task rounds during manual processing. Furthermore, the system automatically resumes periodic execution after manual intervention, eliminating the need for users to manually restart tasks, thus forming a seamless closed loop. Both invitations for manual intervention and notifications include an identifier for the current round of dialogue, ensuring that the pause / resumption actions of the task scheduling and execution modules precisely correspond to specific task execution rounds. This allows manual intervention to quickly locate the task requiring intervention based on this identifier, eliminating the need to drill down through the task list to find the current pending execution record. This reduces the time spent by manual intervention from receiving the notification to entering the processing interface, minimizing context switching costs.

[0057] In some examples, dialogue nodes include: execution input, execution result, and execution processing procedure.

[0058] The dialogue node is defined to include three types of information: execution input, execution result, and execution processing procedure. Unlike traditional task logs that only record the success / failure status, this approach binds and stores the task's triggering conditions (input), the agent's processing behavior (processing procedure), and the final output (result) within the same node. This allows a user or system to obtain a complete picture of the execution at once when accessing any node, without the need for cross-table joins or secondary inferences. By clearly defining the information elements that a dialogue node should contain, each dialogue node in the dialogue chain becomes an independently auditable complete unit. In compliance reviews or troubleshooting scenarios, reviewers do not need to reconstruct the execution process; they can directly read the "execution processing procedure" within the dialogue node to know the agent's specific operation path in that round, significantly improving traceability efficiency.

[0059] In some examples, such as Figure 2 As shown, the intelligent agent workbench 11 also includes: an intelligent agent portal and details display module 112, and an instant dialogue module 113.

[0060] Among them, the intelligent agent portal and details display module 112 is used to display the intelligent agent information of the intelligent agents that have been listed. The intelligent agent information includes the intelligent agent name, capability description, supported task types and running status, so that users can discover and use intelligent agents through a unified entry point.

[0061] The real-time dialogue module 113 supports real-time interaction between users and intelligent agents, and is suitable for non-periodic tasks or auxiliary scenarios. Real-time dialogue and task execution dialogue are distinguished from each other in terms of data structure and presentation to avoid confusion.

[0062] The intelligent agent portal and details display module centrally present the names, capability descriptions, task types, and operational status of all intelligent agents listed on the platform. Before creating recurring tasks, users can fully understand the applicable scenarios and current availability of each intelligent agent in this module, eliminating the need for external documentation or individual trials, thus lowering the barrier to entry for intelligent agents. A separate real-time dialogue module supports non-recurring, non-task-based real-time interaction between users and intelligent agents, clearly distinguishing it from the task management and execution viewing modules. This avoids mixing recurring task execution dialogues with temporary auxiliary dialogues, providing users with a clear and focused interactive interface regardless of their intended use.

[0063] In some examples, such as Figure 3 As shown, the intelligent agent management platform 12 includes: an intelligent agent access and listing management module 121, a permission and usage scope management module 122, a platform configuration and operation management module 123, and an operation record and operation analysis module 124.

[0064] Among them, the intelligent agent access and listing management module 121 is used to access intelligent agents built on external platforms and to manage the listing, delisting and version management of intelligent agents.

[0065] The usage scope management module 122 is used to configure the permissions for different objects to use the smart agent.

[0066] For example, the visibility and availability of intelligent agents can be configured across different users, departments, or business scopes to ensure the security and compliance of intelligent agent usage.

[0067] The platform configuration and operation management module 123 is used to configure the system parameters of the intelligent agent collaborative execution system.

[0068] In some cases, system parameters may include scheduling strategies, notification methods, and human-machine collaboration rules.

[0069] The operation record and operation analysis module 124 is used to record the usage information of the intelligent agent, the frequency of task execution, and the information of human intervention, so as to provide a data foundation for subsequent optimization.

[0070] The intelligent agent access and deployment management module enables intelligent agents built with different external frameworks and technology stacks to be connected to this platform in a standardized manner, and performs governance operations such as deployment, decommissioning, and version management. The permission and usage scope management module configures the permissions for different objects to use intelligent agents, allowing the system to control the visibility and availability of specific intelligent agents based on dimensions such as user, department, and business scope, preventing unauthorized access or cross-business calls, and meeting enterprise-level security and compliance requirements.

[0071] In some examples, the periodic task creation request includes conditions for automatic agent processing. The agent collaborative execution system also employs an active reasoning and collaborative recommendation mechanism based on the periodic task dialogue link. This mechanism works as follows: during the execution of any periodic task, if the agent determines that the execution input of the current round of the task does not meet the automatic agent processing conditions corresponding to that periodic task, the agent generates processing suggestion information for the current round of the task based on the dialogue link corresponding to that periodic task. The execution process of the dialogue node corresponding to the current round of the task includes the processing suggestion information; wherein, the processing suggestion information includes at least one candidate solution information, which includes a candidate processing solution, its corresponding processing description, and a summary of the expected execution result.

[0072] In some examples, when selecting agents and configuring them to perform periodic tasks, agent auto-processing conditions are configured to indicate under what circumstances the agent can automatically handle the task. These conditions can be included in the prompts given to the agent. When the agent is executing a periodic task and scheduling the current round of tasks, if it finds that the execution input of the current round of tasks meets the agent auto-processing conditions, the agent will automatically process the task based on the execution input of the current round of tasks. If, during the execution of any periodic task, the agent determines that the execution input of the current round of tasks does not meet the agent auto-processing conditions corresponding to any periodic task, the agent can check the dialogue link corresponding to that periodic task and generate processing suggestions for the current round of tasks based on the content of that dialogue link.

[0073] Among them, the candidate processing scheme is the processing scheme for the current round of tasks, the processing description is used to explain the candidate processing scheme, and the expected execution result summary is a summary of the execution result produced by executing the candidate processing scheme.

[0074] When an agent encounters a scenario that cannot be handled automatically, it proactively generates processing suggestions based on the historical dialogue chain of the task. This enables the agent to invoke past execution experience, transforming historical experience into reusable candidate solutions, thus improving processing efficiency after human intervention. The processing suggestions explicitly include candidate processing solutions, processing instructions, and a summary of expected execution results. This allows human operators to directly access multiple already reasoned alternative paths with expected consequences when entering a task dialogue node, eliminating the need to manually search historical records or analyze the problem's causes from scratch, effectively shortening the processing time for each human intervention. The generated processing suggestions are written into the execution process of the current dialogue node, ensuring that the agent's reasoning output and candidate solutions are no longer temporary messages independent of the task execution chain, but rather structured node content deeply bound to the current execution. Subsequent agents can further reason based on this node content, forming a continuous transfer of collaborative experience.

[0075] In some examples, candidate solution information is generated based on the execution results and / or human actions (if any) in at least one dialogue node corresponding to the periodic task. Here, the "link" in the dialogue chain refers to a decision structure, not a presentation structure.

[0076] In some cases, the intelligent agent collaborative execution system also employs a dialogue node update mechanism based on human behavior.

[0077] The dialogue node update mechanism based on human behavior is as follows: record the human's operation behavior for the invitation request corresponding to the current round of tasks, and update the operation behavior in the dialogue node corresponding to the current round of tasks.

[0078] By recording the specific actions of each round of human intervention, the system can collect high-value feedback data, such as the specific handling solutions adopted by humans in specific scenarios, without intruding on the internal logic of the intelligent agent. The human actions are updated in the current round of dialogue nodes, ensuring that human choices, modifications, and confirmations are no longer confined to external logs or temporary session records, but become structured data strongly associated with the execution nodes in the task's dialogue chain. Information such as the handling solutions and decision preferences of each round of human intervention is precisely attached to the dialogue node that triggered the intervention, forming a complete chain of evidence. Furthermore, by using dialogue nodes to record human actions, subsequent scenarios requiring human intervention can be addressed by providing more suitable candidate handling solutions based on the records in the dialogue nodes.

[0079] In some cases, agent-cooperative execution systems also employ adaptive collaboration strategies based on human behavior.

[0080] The adaptive collaboration strategy mechanism based on human behavior is as follows: the automatic processing conditions of the agent corresponding to any cycle task are updated based on the user's operation behavior to adjust whether the results of other round tasks meet the automatic processing conditions of the agent; and the ranking of candidate solution information is adjusted when other round tasks do not meet the adjusted automatic processing conditions of the agent, so that when other round tasks are matched with the current round task based on similar features, the other round tasks are automatically processed by the agent.

[0081] It is important to emphasize that the updated automatic processing conditions of the agent can directly affect the execution path determination results of subsequent rounds of tasks.

[0082] In some examples, updating the conditions for automatic processing by an agent may include modifying at least one of the following: changing the decision rule that triggers human intervention, updating the automatic processing threshold, or adjusting the decision logic of the execution path.

[0083] Among them, the logic for adjusting the execution path is used to determine whether the task is handled automatically by the intelligent agent or whether human intervention is invited.

[0084] The embodiments of this application do not limit how the automatic processing conditions of the agent corresponding to any periodic task are updated based on the user's operation behavior. For example, the agent can update the automatic processing conditions based on the user's operation.

[0085] In some cases, the matching between other rounds of tasks and the current round of tasks is based on similarity features. This can be achieved by calculating text similarity, such as by calculating the cosine similarity of the respective execution inputs. When the similarity is greater than a similarity threshold, it is considered that the matching between other rounds of tasks and the current round of tasks is based on similarity features.

[0086] The embodiments of this application do not limit the specific value of the similarity threshold.

[0087] In other examples, the specific implementation of similarity feature matching between other rounds of tasks and the current round of tasks can also be any other way of representing that other rounds of tasks and the current round of tasks are the same or similar tasks.

[0088] By limiting the automatic processing conditions of the agent to be updated based on user actions for any given periodic task, the system proactively uses the results of human intervention to modify the rules (agent automatic processing conditions) by which the agent determines whether a task can be automatically processed in the future. When a human processes an input that did not originally meet the automatic processing conditions in a certain round, the system can update the automatic processing conditions to include such inputs within the scope of future automatic processing. The automation capability of the agent is no longer fixed to the initial configuration, but continues to expand with the accumulation of human collaboration, gradually covering more scenarios that originally required human intervention.

[0089] By adjusting the sorting of candidate solutions when other rounds of tasks do not meet the adjusted conditions for automatic processing by the agent, the system indicates that when subsequent rounds still require human intervention (i.e., still do not meet the updated automatic processing conditions), the system will adjust the display order of each candidate solution based on historical human operation behavior. Solutions that are frequently selected and modified less frequently in historical rounds will receive higher priority in subsequent recommendations. This means that the first information encountered by human intervention is no longer a fixed default option, but a personalized recommendation list that is closer to the user's / the task's historical preferences, thereby shortening the time for human browsing, comparison, and selection.

[0090] The ultimate goal of updating the agent's automatic processing conditions is to facilitate automatic processing in subsequent rounds when matching similar features with the current round. This allows the system to transform the correct processing method demonstrated by the human in the current round into rules that the machine can directly execute in similar future scenarios. Thus, each human intervention not only solves the current problem but also serves as a demonstration and teaching exercise. After learning from this demonstration, the system can automatically reproduce the human processing path when encountering similar inputs in the future, without needing to request human intervention again. This approach allows the system to operate without relying on external retraining or downtime upgrades. During continuous operation, the frequency of human intervention in the same / similar scenarios gradually decreases through natural human-machine collaboration. Furthermore, because the updates are based on actual human operations rather than preset rules, the expansion of the automation boundary is always subject to human verification, avoiding the risk of errors introduced by over-automation.

[0091] To facilitate understanding of the periodic task scheduling system of the above embodiment, in some embodiments, such as Figure 4 As shown, the process of using the periodic task scheduling system can be carried out through the following steps S401-S410.

[0092] S401, Agent Access and Task Configuration.

[0093] The periodic task scheduling system first completes the access and governance configuration of intelligent agents through the intelligent agent management console, including the management of intelligent agents' listing, permission configuration, and the setting of indicators for whether task scheduling is supported.

[0094] Users can select an already listed agent in the agent workbench and configure it as a periodic task, setting task prompts, execution cycle, and start and end times, thereby giving the agent the attributes of a digital employee.

[0095] S402, Task initialization and scheduling start.

[0096] After receiving a task creation request, the periodic task scheduling system generates a corresponding task instance and registers the task with the task scheduling and execution module. The task scheduling and execution module enters the pending execution state according to the task configuration and triggers the agent to execute when the scheduling conditions are met.

[0097] S403, Task conversational execution and conversation node generation.

[0098] Whenever the task management and execution viewing module triggers an execution, the periodic task scheduling system creates a new dialog node for that execution and assigns a unique identifier.

[0099] This dialogue node is used to carry the execution input (input information), execution process, and execution result (output result) for this round of execution. Multiple dialogue nodes form a dialogue link for this cycle task according to the execution order (time order), and the dialogue link can describe the continuous work process of the digital employee.

[0100] S404, Automatic Execution and Execution Result Callback.

[0101] The agent completes the task execution in the context of the dialogue node and sends the execution result callback to the periodic task scheduling system according to the execution status.

[0102] The execution results are presented as a brief text description of the current execution process. These results can be filtered and viewed within the dialogue nodes of the dialogue chain. The platform associates and stores the execution results with their corresponding dialogue nodes.

[0103] If no human intervention is required for this round of execution, the task scheduling and execution module will remain in normal scheduling status and proceed to the next cycle.

[0104] S405, proactive reasoning and human collaborative recommendation based on dialogue links.

[0105] When the agent determines during execution that there are situations that cannot be fully handled automatically in the current round, the agent analyzes the historical execution information based on the dialogue link formed in this cycle task and generates processing suggestion information related to the current round task execution stage. This processing suggestion information can be used as part of the execution process in the dialogue node.

[0106] The processing suggestion information includes at least one candidate solution information. The candidate solution information includes the candidate processing solution, the corresponding processing description, and a summary of the expected execution results. The processing suggestion information is used to provide decision-making reference for human participation.

[0107] S406, Manual intervention request and scheduling paused.

[0108] The agent sends an invitation to human participation to the agent collaborative execution system and notifies the task scheduling and execution module to suspend the subsequent scheduling of tasks in this cycle.

[0109] The intelligent agent collaborative execution system sends a notification to the user to allow manual intervention, guiding the user to the corresponding task dialogue node for processing. During the scheduling pause, the periodic task will not be automatically executed again.

[0110] S407, Human intervention processing and conversational recording.

[0111] Users enter the current dialogue node in the agent's workbench, view the processing suggestions and historical execution information provided by the agent, and select, modify, or confirm candidate solutions. The user's actions are recorded as part of that dialogue node and stored in association with the task dialogue chain.

[0112] S408, Scheduling recovery and task continuation.

[0113] After manual processing is completed and confirmed, the agent sends a scheduling recovery request to the agent collaborative execution system.

[0114] The task scheduling and execution module resumes the execution of the cycle based on the scheduling recovery request, enabling the digital employee to continue running automatically after human intervention and enter the next round of task scheduling.

[0115] S409, Human behavior data collection and collaboration strategy optimization.

[0116] The intelligent agent collaborative execution system summarizes the operational behaviors during human intervention, including the frequency of human intervention, processing time, and solution selection results. Based on these operational behaviors, the intelligent agent collaborative execution system can adjust subsequent task execution strategies, such as optimizing the triggering conditions for human intervention or the priority of processing suggestions, thereby gradually reducing the number of human interventions while ensuring task safety.

[0117] S410, Summary of Execution Results and View of Running Status.

[0118] Users can view the execution status of periodic tasks through the task management and execution viewing module, including the dialogue chain, summaries of each round of execution results, and records of human intervention. This unified display of the task execution process enables visualization and management of the digital employee's operational status.

[0119] The following example of intelligent email management illustrates this point. The application scenario of intelligent email agents will be used to specifically explain the intelligent agent collaborative execution system in this application.

[0120] The intelligent email agent periodically checks user mailboxes and processes emails automatically or in collaboration with humans. As an agent instance within an intelligent agent collaborative execution system, the intelligent email agent is configured to perform tasks periodically, thus possessing the attributes of a digital employee.

[0121] Step 1: Access and governance configuration of the intelligent email agent.

[0122] First, connect the Smart Email agent to the platform through the Smart Agent Management Console and complete the following governance configurations:

[0123] 1. Synchronously obtain the name, version, and capability description information of the intelligent email agent;

[0124] 2. Set up the business categories and usage instructions for the intelligent agent;

[0125] 3. Mark the agent as supporting periodic task execution;

[0126] 4. Configure the user scope and permissions that can use this intelligent agent.

[0127] After completing the above configuration, the Smart Email Agent is uploaded to the Agent Portal and Details Display module for users to access.

[0128] Step 2: Creation and initialization of digital employee tasks.

[0129] Users can select the available Smart Email agent in the Smart Agent Workbench and configure it as a digital employee task, specifically including:

[0130] 1. Set task prompts for email processing to describe the types of emails that can be processed automatically and the conditions that require human intervention;

[0131] 2. Set the task execution cycle and start and end times;

[0132] 3. Submit the task configuration.

[0133] The platform generates corresponding periodic tasks and registers them with the task scheduling and execution module, at which point the periodic task enters the pending execution state.

[0134] Step 3: The cyclical execution process of task dialogue.

[0135] 1. Dialogue node generation.

[0136] The task scheduling and execution module triggers the execution of the periodic task when the scheduling conditions are met.

[0137] The intelligent agent collaborative execution system creates an independent dialogue node for each task execution and assigns a unique identifier to the dialogue node to carry the context information of this round of execution.

[0138] Multiple dialogue nodes, arranged in execution order, form a task dialogue link for the task, which can describe the continuous execution process of the smart mailbox digital employee.

[0139] 2. Automatic execution in scenarios without emails.

[0140] When the intelligent mailbox agent detects in a certain execution round that there are no pending emails in the mailbox:

[0141] a. Summary of the agent's execution results for this round;

[0142] b. Send an execution result callback to the intelligent agent collaborative execution system, indicating that the execution status of this round is "no email";

[0143] c. The intelligent agent collaborative execution system associates and stores the execution results with the corresponding dialogue nodes.

[0144] The task scheduling and execution module remains in normal condition and will proceed to the next cycle.

[0145] 3. Execution scenarios that can automatically process emails.

[0146] When an email matching the automatic processing rules is detected:

[0147] a. The intelligent email agent automatically processes emails within the context of the current conversation node;

[0148] b. Generate a summary of the execution results;

[0149] c. Send an execution result callback to the intelligent agent collaborative execution system, indicating the completed processing content.

[0150] The intelligent agent collaborative execution system records the processing results to the task dialogue link, and the task enters the next cycle.

[0151] Step 4: Trigger the task execution process with human intervention.

[0152] 1. Active reasoning and recommendation based on task dialogue links

[0153] When the intelligent email agent determines that a certain email cannot be processed automatically:

[0154] a. The intelligent agent analyzes the historical execution status based on the task dialogue links already formed for this cycle task;

[0155] b. Generate multiple candidate processing schemes with different processing strategies;

[0156] c. Generate corresponding processing instructions and expected results for each candidate solution.

[0157] The candidate solution is associated with the current dialogue node and used for subsequent manual collaboration.

[0158] 2. Manual intervention requests and scheduling are suspended.

[0159] The intelligent email agent sends a request to the task scheduling and management system to invite human intervention and notifies the task scheduling and execution module to suspend the subsequent scheduling of tasks in this cycle.

[0160] The task scheduling and management system sends a notification to the user to allow manual intervention, guiding the user to the current dialogue node for processing.

[0161] 3. Users can access the corresponding task dialogue node through the intelligent agent workbench to view:

[0162] Current email content;

[0163] Candidate processing schemes for agent generation;

[0164] Historical execution records of the current round of tasks.

[0165] Users can select one of the candidate solutions, modify the solution, or input their own processing information. The results of manual operations are recorded as part of the current dialogue node and associated with the task dialogue chain.

[0166] Step 5: Schedule recovery and task continuation.

[0167] Once manual processing is completed and the results are confirmed:

[0168] 1. The intelligent mailbox agent sends a scheduling recovery request to the task scheduling management system;

[0169] 2. The task scheduling and execution module resumes the periodic execution of the task;

[0170] 3. The intelligent mailbox agent continues to execute automatically in subsequent scheduling cycles.

[0171] The above method enables a smooth switch between automatic execution and manual processing.

[0172] Step Six: Accumulation of Human Behavior and Optimization of Collaboration Strategies.

[0173] The task scheduling and management system records the operational behavior during the manual intervention process, including: the round in which manual intervention occurred, the processing plan selected or modified by the human, and the duration of manual processing.

[0174] The task scheduling and management system adjusts the collaboration strategy for subsequent task execution based on operational behavior. For example, it reduces the frequency of manual intervention for similar emails or optimizes the order of candidate solution generation, thereby gradually improving the level of automation.

[0175] Step 7: View the execution results and running effect.

[0176] Users can view the operation status of the smart email digital employee through the smart agent workbench, including: task dialogue links, summaries of execution results for each round, and records of human intervention.

[0177] Example 2:

[0178] Another embodiment of this application relates to an electronic device configured with an agent cooperative execution system based on a task dialogue link as described in Embodiment 1.

[0179] Example 3:

[0180] Another embodiment of this application relates to a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the intelligent agent cooperative execution system based on a task dialogue link as described in Embodiment 1 above.

[0181] That is, those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0182] Example 4:

[0183] Another embodiment of this application relates to a computer program product, including a computer program. When executed by a processor, the computer program implements the intelligent agent cooperative execution system based on a task dialogue link as described in Embodiment 1.

[0184] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0185] Furthermore, although the steps of the method in this application are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0186] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware.

[0187] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

Claims

1. A collaborative execution system for intelligent agents based on a task dialogue link, characterized in that, include: The intelligent agent workbench is used to provide users with a unified human-computer interaction interface; The agent management console is used for unified management of agents; The task scheduling and execution module is used to receive the registration and deregistration of periodic tasks, and to schedule intelligent agents to execute registered periodic tasks. During the execution of periodic tasks by intelligent agents, a corresponding dialogue node is generated each time task scheduling is triggered. Multiple dialogue nodes corresponding to periodic tasks form a dialogue link in chronological order. The intelligent agent access and collaboration interface module is used to realize the interaction between the intelligent agent collaborative execution system and the intelligent agent; The intelligent agent workbench includes a task management and execution viewing module and a human-computer interaction and notification module. The task management and execution viewing module is used to create periodic tasks in response to periodic task creation requests and register the periodic tasks with the task scheduling and execution module; receive and deregister the periodic tasks with the task scheduling and execution module based on periodic task termination instructions; and display the execution process of the periodic tasks in the form of a dialogue link in response to periodic task execution viewing requests. The human-computer interaction and notification module is used to issue human participation notifications based on a preset method in response to human participation invitation requests fed back during the execution of periodic tasks by the intelligent agent. The intelligent agent collaborative execution system also applies a human-machine collaboration and scheduling linkage mechanism; the human-machine collaboration and scheduling linkage mechanism is as follows: during the execution of any cycle task by the intelligent agent, if it is determined that the execution input of the current round task does not meet the automatic processing conditions of the intelligent agent corresponding to the any cycle task, the intelligent agent sends a request to the human-machine interaction and notification module to invite human participation, and sends a request to the task scheduling and execution module to suspend the scheduling of the any cycle task. Upon completion of manual processing, the intelligent agent sends feedback to the task scheduling and execution module to resume scheduling for any given periodic task; Wherein, the invitation for human participation corresponding to the current round of tasks includes the identifier of the dialogue node corresponding to the current round of tasks, and the human participation notification corresponding to the current round of tasks includes the identifier; The intelligent agent collaborative execution system also applies a dialogue node update mechanism based on human behavior; the dialogue node update mechanism based on human behavior is as follows: record the operation behavior of the human in response to the invitation request for human participation corresponding to the current round of tasks, and update the operation behavior in the dialogue node corresponding to the current round of tasks; The intelligent agent collaborative execution system also applies an adaptive collaboration strategy mechanism based on human behavior; the adaptive collaboration strategy mechanism based on human behavior is as follows: based on the operation behavior, the intelligent agent automatic processing conditions corresponding to any cycle task are updated to adjust whether subsequent round tasks meet the intelligent agent automatic processing conditions; and, the ranking of candidate scheme information when the other round tasks do not meet the adjusted intelligent agent automatic processing conditions is adjusted so that when the other round tasks and the current round tasks are matched based on similar features, the other round tasks are automatically processed by the intelligent agent.

2. The system according to claim 1, characterized in that, The intelligent agent workbench also includes: The agent portal and details display module is used to display agent information of agents that have been listed. The agent information includes agent name, capability description, supported task types and running status. The real-time dialogue module is used to support real-time interaction between users and intelligent agents.

3. The system according to claim 1, characterized in that, The intelligent agent management console includes: The intelligent agent access and listing management module is used to access intelligent agents built on external platforms and to manage the listing, delisting and versioning of intelligent agents. The permissions and scope of use management module is used to configure the permissions for different objects to use the intelligent agent; The platform configuration and operation management module is used to configure the system parameters of the intelligent agent collaborative execution system; The operation log and operation analysis module is used to record the usage information of the intelligent agent, the frequency of task execution, and the information of human intervention.

4. The system according to claim 1, characterized in that, The dialogue node includes three types of information: execution input, execution result, and execution processing procedure.

5. The system according to claim 4, characterized in that, The periodic task creation request includes conditions for automatic processing by the intelligent agent; The intelligent agent collaborative execution system also applies an active reasoning and collaborative recommendation mechanism based on a periodic task dialogue link; the active reasoning and collaborative recommendation mechanism based on the task dialogue link is as follows: when an intelligent agent executes any periodic task, if it is determined that the execution input of the current round of the task does not meet the automatic processing conditions of the intelligent agent corresponding to the any periodic task, the intelligent agent generates processing suggestion information for the current round of the task based on the dialogue link corresponding to the any periodic task. The execution process of the dialogue node corresponding to the current round of tasks includes the processing suggestion information; The processing suggestion information includes at least one candidate solution information, which includes a candidate processing solution, a corresponding processing description, and a summary of the expected execution results.

6. An electronic device, characterized in that, The electronic device is equipped with an intelligent agent cooperative execution system based on a task dialogue link as described in any one of claims 1-5.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent agent cooperative execution system based on task dialogue link as described in any one of claims 1-5.

8. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the intelligent agent cooperative execution system based on task dialogue link as described in any one of claims 1-5.

Images