Task processing methods, cloud platforms, smart terminal devices, and task processing systems

Abstract

This application relates to the field of data processing technology, and discloses a task processing method, a cloud platform, a smart terminal device, and a task processing system. The method includes: responding to a task processing request sent by the smart terminal device; generating a task execution plan for the task to be processed in the task processing request through a classifier agent; determining all tool services required to execute the task to be processed based on the task execution plan; and executing the task to be processed by calling each tool service through a scheduler agent, generating a task execution result. Compared to existing smart terminal devices, which suffer from poor performance in processing complex tasks due to their own resource limitations, this application improves the performance of smart terminal devices in processing complex tasks by utilizing agents integrated on a cloud platform to call various tool services.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to task processing methods, cloud platforms, smart terminal devices, and task processing systems. Background Technology

[0002] With the widespread use of smart terminal devices (such as smart glasses, smartwatches, and smart toys), these devices are playing an increasingly important role in people's daily lives. However, due to the limited computing resources, storage capacity, and battery life that smart terminal devices typically have, these limitations make it difficult for them to directly run complex AI algorithms, thus restricting their ability to handle complex tasks. Summary of the Invention

[0003] The main purpose of this application is to provide a task processing method, cloud platform, smart terminal device and task processing system, which aims to solve the technical problem that the performance of existing smart terminal devices in handling complex tasks is poor due to resource limitations.

[0004] To achieve the above objectives, this application proposes a task processing method, which is applied to a cloud platform in a task processing system. The task processing system further includes a smart terminal device, and the cloud platform integrates a classifier agent and a scheduler agent. The method includes: In response to a task processing request sent by the smart terminal device, a task execution plan for the task to be processed in the task processing request is generated by the classifier agent. Based on the task execution plan, determine all the tools and services required to execute the task to be processed; The scheduling executor agent invokes each of the tool services to execute the tasks to be processed and generates task execution results.

[0005] In one embodiment, the step of generating a task execution plan for the task to be processed in the task processing request through the classifier agent includes: The classifier agent evaluates the complexity of the tasks to be processed in the task processing request to determine the task complexity of the tasks to be processed. Determine the task reasoning strategy for the task to be processed based on the task complexity; Based on the task reasoning strategy, the task to be processed is reasoned to generate a task execution plan for the task to be processed.

[0006] In one embodiment, the step of evaluating the complexity of the tasks to be processed in the task processing request through the classifier agent to determine the task complexity of the tasks to be processed includes: The classifier agent performs semantic analysis on the task text of the task to be processed in the task processing request to determine the key logical nodes and branch nodes of the key logical nodes in the task to be processed. Construct a directed graph of task execution logic corresponding to the task to be processed based on the key logic nodes and the branch nodes; The number of logical branches and the depth of dependencies of the task to be processed are determined based on the directed graph of the task execution logic. The complexity of the task to be processed is determined by evaluating the number of logical branches and the depth of dependencies.

[0007] In one embodiment, the step of reasoning about the task to be processed based on the task reasoning strategy to generate a task execution plan for the task to be processed includes: Based on the task reasoning strategy, reason about the task to be processed and generate the task execution thought chain of the task to be processed. The task to be processed is broken down according to the task execution thought chain to obtain several sub-tasks to be processed. Determine the task execution requirements for each subtask to be processed and the task dependencies between the subtasks to be processed; A task execution plan for the task to be processed is generated based on the task execution requirements and the task dependencies.

[0008] In one embodiment, the step of determining all tool services required to execute the task to be processed based on the task execution plan includes: Based on the task execution plan, determine the target tool services required to execute each of the sub-tasks in the task to be processed; The step of invoking each of the tool services through the scheduling executor agent to execute the task to be processed and generating the task execution result includes: The scheduling executor agent invokes the target tool service to execute the sub-task to be processed, and generates the sub-task execution result; The results of the subtasks are integrated to generate the final task execution result.

[0009] In one embodiment, the cloud platform further integrates a post-processor intelligent agent; after the step of integrating the sub-task execution results to generate the final task execution result, the platform further includes: The task execution result is logically verified through the task execution thought chain. If the logical verification fails, the task execution result will be corrected based on the logical verification result.

[0010] In addition, to achieve the above objectives, this application also proposes a task processing method, which is applied to an intelligent terminal device in a task processing system. The task processing system further includes a cloud platform, in which a classifier agent and a scheduler agent are integrated. The method includes: A task processing request is generated based on the task processing instructions input by the user, and the task processing request is sent to the cloud platform; The cloud platform is configured to respond to the task processing request by generating a task execution plan for the pending task in the task processing request through the classifier agent; determining all tool services required to execute the pending task based on the task execution plan; and calling each of the tool services through the scheduler agent to execute the pending task and generate a task execution result.

[0011] In addition, to achieve the above objectives, this application also proposes a cloud platform, which includes: a first memory, a first processor, and a task processing program stored on the first memory and executable on the first processor. When the task processing program is executed by the first processor, it implements the steps of the task processing method described above.

[0012] In addition, to achieve the above objectives, this application also proposes an intelligent terminal device, which includes: a second memory, a second processor, and a task processing program stored in the second memory and executable on the second processor. When the task processing program is executed by the second processor, it implements the steps of the task processing method described above.

[0013] In addition, to achieve the above objectives, this application also proposes a task processing system, which includes a cloud platform as described above and a smart terminal device as described above.

[0014] This application provides a task processing method, a cloud platform, a smart terminal device, and a task processing system. The method is applied to a cloud platform within the task processing system, which further includes a smart terminal device. The cloud platform integrates a classifier agent and a scheduler agent. The method includes: responding to a task processing request sent by the smart terminal device; generating a task execution plan for the pending task in the task processing request through the classifier agent; determining all tool services required to execute the pending task based on the task execution plan; and executing the pending task by invoking each of the tool services through the scheduler agent, thereby generating a task execution result.

[0015] The intelligent terminal device of this application can generate a task processing request after receiving a task processing instruction input by the user, and send the task processing request to the cloud platform. After receiving the task processing request, the cloud platform can generate a task execution plan for the task to be processed through its classifier agent, determine the tool services required to execute the task based on the task execution plan, and then execute the task by calling the tool services through the scheduling executor agent. Compared with the poor performance of existing intelligent terminal devices in handling complex tasks due to their own resource limitations, the intelligent terminal device of this application can utilize the agent integrated on the cloud platform to call various tool services to execute complex tasks, thereby improving the performance of the intelligent terminal device in handling complex tasks. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the cloud platform structure of the hardware operating environment involved in the embodiments of this application; Figure 2 This is a flowchart illustrating the first embodiment of the task processing method of this application; Figure 3 This is a flowchart illustrating the second embodiment of the task processing method of this application; Figure 4 This is a complete flowchart of the task processing method in this application; Figure 5 This is a flowchart illustrating the third embodiment of the task processing method of this application; Figure 6 This is a schematic diagram of the structure of a smart terminal device in the hardware operating environment involved in the embodiments of this application.

[0019] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0020] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0021] Reference Figure 1 , Figure 1This is a schematic diagram of the cloud platform structure of the hardware operating environment involved in the embodiments of this application.

[0022] like Figure 1 As shown, the cloud platform may include: a first processor 1001, such as a central processing unit (CPU), a first communication bus 1002, a first user interface 1003, a first network interface 1004, and a first memory 1005. The first communication bus 1002 is used to enable communication between these components. The first user interface 1003 may be connected to a display screen. Optionally, the first user interface 1003 may include a standard wired interface or a wireless interface; in this application, the wired interface of the first user interface 1003 may be a USB interface. The first network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The first memory 1005 may be a high-speed random access memory (RAM) or a non-volatile memory (NVM), such as a disk storage device. Optionally, the first memory 1005 may also be a storage device independent of the aforementioned first processor 1001.

[0023] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on the cloud platform and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0024] like Figure 1 As shown, the first memory 1005, which is identified as a computer storage medium, may include a first operating system, a first network communication module, a first user interface module, and a task processing program.

[0025] exist Figure 1 In the cloud platform shown, the first network interface 1004 is mainly used to connect to the backend server and communicate with the backend server; the first user interface 1003 is mainly used to connect to the user equipment; the cloud platform calls the task processing program stored in the first memory 1005 through the first processor 1001 and executes the steps of the task processing method provided in the first to second embodiments of this application.

[0026] It should be noted that with the widespread use of smart terminal devices (such as smart glasses, smartwatches, and smart toys), these devices are playing an increasingly important role in people's daily lives. However, due to the limited computing resources, storage capacity, and battery life that smart terminal devices typically have, these limitations make it difficult for them to directly run complex AI algorithms, thus restricting their ability to handle complex tasks.

[0027] Therefore, to address the aforementioned shortcomings, this embodiment provides a task processing method. In this embodiment, the intelligent terminal device can generate a task processing request after receiving a task processing instruction input by the user and send the task processing request to a cloud platform. After receiving the task processing request, the cloud platform can generate a task execution plan for the task to be processed through its classifier agent, determine the tool services required to execute the task based on the task execution plan, and then execute the task by calling the tool services through a scheduling executor agent. Compared to the poor performance of existing intelligent terminal devices in handling complex tasks due to their own resource limitations, this embodiment improves the performance of intelligent terminal devices in handling complex tasks by utilizing the agents integrated on the cloud platform to call various tool services.

[0028] For ease of understanding, the following is combined with Figures 2 to 5 The task processing method provided in the embodiments of this application will be described in detail.

[0029] Reference Figure 2 , Figure 2 This is a flowchart illustrating the first embodiment of the task processing method of this application. The first embodiment of the task processing method of this application is presented as follows: Figure 2 As shown, in this embodiment, the method is applied to a cloud platform in a task processing system. The task processing system further includes intelligent terminal devices, and the cloud platform integrates a classifier agent and a scheduler agent. The specific method includes: Step S20: In response to the task processing request sent by the smart terminal device, the classifier agent generates a task execution plan for the task to be processed in the task processing request.

[0030] It is understood that the method of this embodiment can be applied to the aforementioned cloud platform in a task processing system, which can be a collection of computing and storage resources deployed on a server. The execution entity of this method embodiment can utilize the aforementioned cloud platform to describe this embodiment and the following embodiments.

[0031] It is also understood that in this embodiment, the cloud platform can provide API interfaces for communication with smart terminal devices. At the same time, the cloud platform can also run various AI models (such as LLM (Large Language Model) large models) and tool services (such as MCP (Model Context Protocol) servers) for task processing.

[0032] It should be understood that the aforementioned task processing requests can be initiated by smart terminal devices and are specific task instructions that require processing by the cloud platform. These instructions may include specific functions or operations that the user wishes to achieve through the smart terminal device. In practical applications, due to the resource limitations of the smart terminal device itself, it is difficult to directly run complex AI algorithms. Therefore, after receiving a task processing request, the smart terminal device can send the request to the cloud platform to leverage the powerful computing capabilities and resources of the cloud platform to complete the processing of complex tasks.

[0033] It should also be understood that the aforementioned tasks to be processed can be specific tasks that need to be processed by the cloud platform, that is, specific work that the user wants the smart terminal device to complete, such as data analysis, text generation, image recognition, etc. This embodiment does not limit this.

[0034] It should be noted that the aforementioned classifier agent can be a module integrated into the cloud platform for analyzing, classifying, and initially planning task processing requests sent by smart terminal devices. It can understand the intent of the task, assess the complexity of the task, and generate a corresponding task execution plan for subsequent task execution.

[0035] It should also be noted that the above task execution plan can be generated by the classifier agent to guide the cloud platform in executing the task in a detailed sequence of steps. This sequence specifies the execution order, dependencies, and required resources to ensure that the task can be executed efficiently and accurately.

[0036] In practical use, after receiving a task processing instruction from a user via voice or text, a smart terminal device can generate a corresponding task processing request and send it to a cloud platform via a predefined communication protocol (such as RESTful API, MQTT, etc.). Upon receiving the task processing request, the classifier agent on the cloud platform can parse the request, obtain a task description, and perform intent recognition and complexity assessment on the task to be processed based on the task description. Then, the classifier agent can select an appropriate inference strategy based on the task complexity assessment results to perform task inference, determine the execution information of each subtask in the task to be processed, such as requirements, execution order, and dependencies, and finally generate a task execution plan based on this information. For example, if a user inputs the task text "Summarize the main work of the past week regularly (every Friday), output the summary text in Markdown format, and send it to the supervisor via email" on a smart terminal device, the cloud platform, upon receiving a task processing request carrying this task description, can first parse the task description and identify that one of the user's intentions is "to generate this week's work report." For the task "to generate this week's work report," the classifier agent may identify that: the task is highly complex, involving multiple data sources (such as Jira, calendar, Git); and requires multiple steps such as data acquisition, data processing, and text generation. Based on this, the classifier agent can break down this complex task into multiple subtasks, for example: Subtask 1: Determine the time range for this week; Subtask 2: Obtain this week's work data from multiple data sources; Subtask 3: Summarize and refine the acquired data; Subtask 4: Convert the summary content into Markdown format; Subtask 5: Construct the email content and send it, and plan the execution order and dependencies of each subtask, ultimately generating a task execution plan.

[0037] Furthermore, in order to accurately select an appropriate inference strategy based on task complexity and ensure that the task is processed using the most suitable method during execution, thereby improving task processing efficiency and accuracy, the step of generating a task execution plan for the task to be processed in the task processing request through the classifier agent includes: Step S21: The classifier agent evaluates the complexity of the tasks to be processed in the task processing request to determine the task complexity of the tasks to be processed.

[0038] It should be noted that the aforementioned task complexity can be an indicator used to reflect the difficulty and complexity of the task's execution. In this embodiment, the cloud platform can evaluate the task's complexity by analyzing multiple dimensions of the task, including the task's semantic complexity, the number of logical branches, the depth of dependencies, the number and type of data sources, and the scope of domain knowledge, etc. This embodiment does not impose any limitations on these aspects.

[0039] It should also be noted that the semantic complexity of a task can refer to the semantic complexity of the task description, which can be determined by judging whether the task involves knowledge from multiple domains or requires understanding complex contexts. In this embodiment, the classifier agent can use a deep learning model based on the Transformer architecture to perform semantic analysis on the task input, extract the semantic features of the task, and determine the semantic complexity of the task based on the semantic features using a semantic complexity scoring model, thereby evaluating the complexity of the task. The number of logical branches can refer to the number of logical branches that the task may involve during execution, such as conditional judgments and loops; the dependency depth can be the depth of the dependency relationships between various subtasks in the task, which can be determined by detecting the number of subtasks that the task needs to execute in a specific order. In this embodiment, the cloud platform can analyze the logical structure of the task, identify key logical nodes and branch points in the task, and construct a logic tree or directed graph, thereby determining the number of logical branches and the dependency depth of the task based on the constructed logic tree or directed graph. The number and type of data sources can refer to the number and types of data sources that the task needs to access. This can be determined by detecting the external data sources (such as databases and APIs) involved in the task. In this embodiment, the classifier agent can assess the computing resources (such as CPU and memory), storage resources, and network resources required for task execution, and evaluate the complexity of the task based on its resource requirements. Complex tasks typically require more resources. The scope of domain knowledge can refer to the scope of domain knowledge involved in the task, such as whether specific domain expertise is required. In this embodiment, the classifier agent can determine whether task execution requires calling domain-specific models or rules based on the domain to which the task belongs, and evaluate the complexity of the task accordingly.

[0040] Furthermore, in order to achieve accurate complexity assessment of the task, step S21 includes: Step S211: The classifier agent performs semantic analysis on the task text of the task to be processed in the task processing request to determine the key logical nodes and branch nodes of the key logical nodes in the task to be processed.

[0041] It should be understood that the aforementioned task text can be text describing the task to be processed, such as "generate this week's work summary and send an email". In practical applications, the classifier agent can parse the received task processing request and extract the descriptive text of the task to be processed. This text can be directly entered by the user through the text box provided by the smart terminal device, or it can be obtained by the smart terminal device after recognizing the user's voice signal. This embodiment does not impose any restrictions on this.

[0042] Understandably, the aforementioned key logical nodes can be core steps or decision points that must be completed during the execution of the task to be processed. For example, for the task of "generating this week's work summary and sending an email", the key logical nodes may include: obtaining this week's work data, summarizing and refining key work, generating a summary text in Markdown format, and constructing and sending the email content.

[0043] It is also understood that the aforementioned branch nodes can be multiple selection paths or conditional branches that may appear during the execution of the task to be processed. These nodes usually correspond to conditional judgments or multiple possible execution paths in the task to be processed. For example, for the task of "generating this week's work summary and sending an email", the branch nodes may include: checking whether the data source is available (if not, selecting an alternative data source or reporting an error); checking whether there is enough data to summarize (if not, generating a short "no data" report).

[0044] Step S212: Construct a directed graph of task execution logic corresponding to the task to be processed based on the key logic node and the branch node.

[0045] It should be noted that the above directed graph of task execution logic can be a directed graph used to describe the logical structure and execution order of a task. The nodes can be the key logical nodes and branch nodes of the task, and the directed edges can represent the execution order and dependencies between nodes.

[0046] Step S213: Determine the number of logical branches and the depth of dependencies of the task to be processed based on the directed graph of the task execution logic.

[0047] It should be noted that the number of logical branches mentioned above can be the number of branch nodes in the directed graph of task execution logic, which can reflect the number of possible execution paths in the task.

[0048] It should also be noted that the aforementioned dependency depth can be the length of the longest path from the starting node to the farthest node in the directed graph of the task execution logic. It can reflect the hierarchical depth of the task logic structure, that is, the most complex dependency chain in the task execution process.

[0049] In this embodiment, the classifier agent can determine the number of logical branches and the depth of dependencies in the task to be processed by traversing the directed graph of task execution logic, counting the number of branch nodes, and finding the longest path length from the start node to the end node. For example, for the task of "generating this week's work report", the number of logical branches may be 2, including: whether the data source is available; whether there is enough data to summarize; the depth of dependencies may be 4, including: obtaining this week's work data → summarizing and refining key work → generating a summary text in Markdown format → constructing email content and sending it.

[0050] Step S214: Evaluate the complexity of the task to be processed based on the number of logical branches and the depth of the dependency relationship, and determine the task complexity of the task to be processed.

[0051] In practical applications, after receiving a task processing request, the classifier agent parses the request, extracts the task text, and uses a deep learning model based on a Transformer architecture to perform semantic analysis on the task text, extracting key information such as task objectives, operation steps, and conditional judgments. Based on this key information, it identifies key logical nodes and branch nodes within the task. Then, the classifier agent uses these key logical nodes and branch nodes as nodes in a graph, adding directed edges according to the task's execution order and dependencies to construct a directed graph of the task's execution logic. Subsequently, the classifier agent uses graph theory methods to calculate the number of logical branches in the directed graph and uses depth-first search (DFS) or topological sorting to calculate the dependency depth. Finally, it combines the number of logical branches and the dependency depth to evaluate the complexity of the task and determine its overall complexity.

[0052] Step S22: Determine the task reasoning strategy for the task to be processed based on the task complexity.

[0053] It should be noted that the aforementioned task reasoning strategy can be a specific method and logical path for processing and executing tasks. It determines how tasks are broken down, reasoned, and executed to ensure efficient and accurate task completion. The task reasoning strategy in this embodiment can be determined based on the task type to be processed, and may include, but is not limited to, single-layer thought chain reasoning strategies, multi-level recursive thought chain reasoning strategies, and knowledge-enhanced thought chain reasoning strategies. A single-layer thought chain reasoning strategy refers to a reasoning strategy that directly generates output from task input through a single logical reasoning chain, without recursive calls or hierarchical divisions. It is suitable for tasks with relatively simple logical structures that do not require in-depth recursive decomposition, such as simple data queries and basic text analysis. A multi-level recursive thought chain reasoning strategy refers to a reasoning strategy that solves problems step-by-step through multiple logical reasoning chains. This strategy may require recursive calls to the reasoning chain and is suitable for complex tasks, such as multi-step decision-making, complex logical reasoning, and multi-level data association analysis. A knowledge-enhanced thought chain reasoning strategy refers to a strategy that requires combining external knowledge bases or domain knowledge for reasoning. It is suitable for tasks requiring professional knowledge support, such as medical diagnosis, legal analysis, and scientific research.

[0054] In practical applications, after determining the task complexity of the task to be processed, the classifier agent can classify the task according to its complexity and match the corresponding task inference strategy according to the task category. In this embodiment, the task categories can include: low-complexity tasks, medium-complexity tasks, high-complexity tasks, and domain-specific tasks. Low-complexity tasks are those with simple logic and low resource requirements. These tasks can usually be responded to directly or completed through simple logical processing without complex reasoning, such as querying a simple factual question: "What is today's date?". Medium-complexity tasks are those with relatively complex logical structures (potentially involving multiple steps) but do not require deep recursion or complex models. These tasks can use a single-layer thought chain reasoning strategy based on simple logic, such as generating a simple report based on user input. High-complexity tasks are those with complex logic, potentially involving multi-layer recursion, complex model reasoning, or large-scale data processing. These tasks can use a multi-level recursive thought chain reasoning strategy based on hierarchical logic, such as generating a comprehensive report containing information from multiple data sources. Professional domain tasks can involve knowledge in a specific domain and require the support of domain expert models or knowledge graphs. Such tasks can employ knowledge-enhanced thinking chain reasoning strategies to combine domain knowledge graphs and expert models for reasoning, such as generating a medical diagnostic report.

[0055] Step S23: Based on the task reasoning strategy, reason about the task to be processed and generate a task execution plan for the task to be processed.

[0056] In practical applications, after determining the task reasoning strategy for the task to be processed, the task can be decomposed according to the task reasoning strategy, breaking it down into multiple subtasks. The execution order and dependencies of each subtask are planned, and then a detailed task execution plan is generated based on the subtask descriptions, execution order, dependencies, and other information. Specifically, for a single-level thought chain reasoning strategy, the task can be decomposed into several independent subtasks, each corresponding to a simple logical operation. Simple reasoning is performed using simple logical rules or pre-set models to determine the execution order, execution logic, and other information of the subtasks, ultimately generating the corresponding task execution plan. For a multi-level recursive thought chain reasoning strategy, the task can be decomposed into multiple levels of subtasks. Each level of subtasks may be further decomposed into finer-grained subtasks, and complex reasoning is performed step-by-step through recursive calls to the reasoning chain to determine the execution order, execution logic, and other information of the subtasks, ultimately generating the corresponding task execution plan. The knowledge-enhanced thinking chain reasoning strategy can integrate the logical reasoning of a task with the knowledge in the knowledge base. It guides the selection of reasoning paths through knowledge and dynamically adjusts the reasoning strategy based on the information in the knowledge base during the reasoning process. This enhances the accuracy and depth of reasoning, determines the execution order and execution logic of the task, and ultimately generates the corresponding task execution plan.

[0057] Step S30: Determine all the tool services required to execute the task to be processed according to the task execution plan.

[0058] It should be understood that the aforementioned tool services can be external tools and resources used to complete specific functions or operations during task execution. These tool services can be services provided by the cloud platform, or third-party APIs, databases, model services, file storage services, etc. This embodiment does not impose any restrictions on them.

[0059] In this embodiment, the tool services may include, but are not limited to, data acquisition tools (such as database query tools, web crawler tools, API interfaces, etc.), data processing tools (such as data analysis tools, text processing tools, image processing tools, etc.), task execution tools (such as email sending tools, file generation tools, etc.), and enhancement tools (such as virtual machine or sandbox environments, long short-term memory enhancement tools, etc.).

[0060] In practical applications, the execution of a task may rely on the functionality of multiple tool services. Therefore, this embodiment can first determine all the tool services required to execute the task based on the task's execution plan. For example, for the task of "generating this week's work report," the tool services that may be needed include: Jira API: used to obtain data on tasks completed this week; Calendar API: used to obtain meeting minutes this week; Git API: used to obtain code commit records this week; Markdown renderer: used to convert the summary content into Markdown format; Email client: used to send the generated weekly report. After determining all the tool services required to execute the task, the scheduler agent can call the corresponding tool services to execute specific tasks during task execution.

[0061] Step S40: The scheduling executor agent calls each of the tool services to execute the task to be processed and generates the task execution result.

[0062] It should be noted that the aforementioned scheduling executor intelligent agent can be a module integrated into a cloud platform for calling tool services and actually executing tasks. It can schedule resources, execute tasks, and interact with external systems.

[0063] It should also be noted that the above task execution result can be the final output generated after the task is completed, which may include the task execution status, generated data, log information during the execution process, etc. This embodiment does not impose any restrictions on this.

[0064] In practical applications, after receiving the task execution plan from the classifier agent, the scheduler agent analyzes each subtask in the plan, determines the necessary tool services to complete them, and selects appropriate tools from the tool service library based on the type and requirements of the subtasks. After identifying all the necessary tools for executing the task, the scheduler agent sequentially calls the required tools according to the order of the subtasks in the execution plan, while handling dependencies between subtasks to ensure correct execution. Upon completion of the task, the scheduler agent integrates all outputs and status information generated during execution, ultimately generating and outputting the task execution result.

[0065] In this embodiment, the smart terminal device can generate a task processing request after receiving a task processing instruction input by the user, and send the task processing request to the cloud platform. Upon receiving the task processing request, the cloud platform can generate a task execution plan for the task through its classifier agent, determine the necessary tool services for executing the task based on the task execution plan, and then execute the task by calling the tool services through the scheduling executor agent. Compared to existing smart terminal devices whose performance in handling complex tasks is poor due to their own resource limitations, this embodiment improves the performance of the smart terminal device in handling complex tasks by utilizing the agent integrated on the cloud platform to call various tool services.

[0066] Reference Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the task processing method of this application. Based on the first embodiment described above, a second embodiment of the task processing method of this application is proposed.

[0067] Considering that different tasks have different complexities and logical structures—simple tasks may only require one operation, while complex tasks may involve multiple steps and conditional branches—this embodiment can dynamically select an appropriate task reasoning strategy to generate a task execution thought chain based on the task's complexity. This adapts to tasks of varying complexity and improves the accuracy of task execution. Figure 3 As shown, in this embodiment, step S23 includes: Step S231: Based on the task reasoning strategy, reason about the task to be processed to generate the task execution thought chain of the task to be processed.

[0068] It should be noted that the aforementioned task execution thought chain can be a logical reasoning path formed during the task reasoning process, reflecting the steps and logical sequence required to complete the task. In practical applications, the process of reasoning about a task is similar to the human thought process. The scheduling executor agent can determine the specific steps and logical sequence to complete the task through step-by-step reasoning, thereby forming the task execution thought chain for the task to be processed.

[0069] Step S232: The task to be processed is split into several sub-tasks to be processed according to the task execution thought chain.

[0070] Understandably, the aforementioned subtasks to be processed can be specific and independent operational steps broken down from the task to be processed. These subtasks all have clear goals, inputs, and outputs, and may depend on the outputs of other subtasks. For example, the task of "generating this week's work report" can be broken down into multiple subtasks, including: Subtask 1: Obtain this week's work data, the input of which can be: the time range of this week, and the output can be: raw work data obtained from multiple data sources (such as Jira, calendar, Git); Subtask 2: Summarize and extract key work, the input of which can be: raw work data, and the output can be: a structured summary of key work; Subtask 3: Generate a summary text in Markdown format, the input of which can be: a structured summary of key work, and the output can be: a weekly report text in Markdown format; Subtask 4: Construct email content and send it, the input of which can be: a weekly report text in Markdown format and the recipient's email address, and the output can be: the email sending status.

[0071] Step S233: Determine the task execution requirements corresponding to each subtask to be processed and the task dependencies between the subtasks to be processed.

[0072] It should be understood that the above-mentioned task execution requirements may include the resources, tools, and services required to complete each subtask, as well as the specific requirements for these resources. These requirements may include computing resources, data sources, external tool calls, etc., and this embodiment does not impose any limitations on them. For example, the execution requirements for subtask 1 "obtaining this week's work data" may include: access to the Jira API and network connectivity; the execution requirements for subtask 2 "summarizing and extracting key work" may include: text processing models and computing resources; the execution requirements for subtask 3 "generating a summary text in Markdown format" may include: a Markdown generator and formatting tools, etc.

[0073] It should also be understood that the above task dependencies can be the order and conditions of subtask execution. In practical applications, the execution of some subtasks in a task to be processed often depends on the output of other subtasks, and these dependencies determine the order of task execution. For example, subtask 2 "summarizing and extracting key work" depends on the output of subtask 1 "obtaining this week's work data"; subtask 3 "generating a summary text in Markdown format" depends on the output of subtask 2 "summarizing and extracting key work"; and subtask 4 "building email content and sending it" depends on the output of subtask 3 "generating a summary text in Markdown format".

[0074] Step S234: Generate a task execution plan for the task to be processed based on the task execution requirements and the task dependencies.

[0075] In practical applications, the scheduling executor agent first selects an appropriate task reasoning strategy (such as single-layer thought chain reasoning, multi-level recursive thought chain reasoning, or knowledge-enhanced thought chain reasoning) based on the complexity of the task to be processed and the characteristics of the domain. It then uses the selected strategy to perform logical reasoning on the task to be processed, generating a task execution thought chain. Next, the scheduling executor agent decomposes the task to be processed into several sub-tasks based on the task execution thought chain. Each sub-task corresponds to a logical node in the thought chain. It analyzes each sub-task to determine its execution requirements, including the necessary tools, resources, and data. Simultaneously, it determines the dependencies between sub-tasks, including input-output relationships and conditional dependencies. Finally, it generates a task execution plan based on the execution requirements, dependencies, and required resources or tools of the sub-tasks.

[0076] Furthermore, to ensure the logical correctness and consistency of task execution, avoid logical errors or conflicts, and improve the reliability of task execution, the cloud platform also integrates a post-processor intelligent agent; step S231 includes: performing logical verification on the task execution result through the task execution thought chain; if the logical verification fails, the task execution result is corrected based on the logical verification result.

[0077] It should be noted that the aforementioned post-processor agent can be a model integrated into a cloud platform, used for logical verification and validation of task execution results. This embodiment uses a post-processor agent to verify and validate task execution results, ensuring the logical consistency and correctness of the results, thereby improving the accuracy of task processing. Furthermore, when the post-processor agent detects logical errors or inconsistencies in the task execution results, it can correct or re-reason the results, further ensuring the logical consistency and correctness of the execution results.

[0078] Understandably, logical verification can be a process of verifying the logical consistency and correctness of the task inference results, ensuring that the task execution results correctly reflect the task's logical structure and execution path. In this embodiment, the logical verification performed by the post-processor agent may include, but is not limited to: verifying subtask logic, dependencies, resource requirements, and task completeness. Specifically, verifying subtask logic can verify whether the logic of each subtask is reasonable and whether it can achieve its goal; verifying dependencies can check whether the dependencies between subtasks are correct and whether there are circular or missing dependencies; verifying resource requirements can verify whether the resource requirements of each subtask are clear and can be met; and verifying task completeness can verify whether the task inference results contain all necessary subtasks and steps without omissions.

[0079] In practical applications, after generating the task execution results for the tasks to be processed, the scheduling executor agent can send the results to the post-processor agent. At this point, the post-processor agent can perform logical verification on the task execution results to ensure that the results are logically correct and consistent. This includes: checking whether the execution order of subtasks is reasonable; verifying whether the dependencies between subtasks are correct; and ensuring that the resource requirements of all subtasks are clear and satisfyable. If the logical verification fails, the post-processor agent can correct the task execution results based on the verification results. Specifically, if the logical verification fails, the post-processor agent can analyze the root cause of the failure (such as tool call failure, incomplete data, or incorrect inference path) and trigger task refactoring to correct the task execution results. For example, if the failure is due to a tool call failure, a simple retry or activation of a backup tool can be performed; if the failure is due to the inability to obtain valid data, the data source needs to be re-analyzed or the strategy adjusted, and the task plan re-formulated.

[0080] Furthermore, in order to efficiently process tasks with complex structures and improve the flexibility and adaptability of task execution, step S30 includes: determining the target tool services required to execute each of the sub-tasks in the task to be processed according to the task execution plan.

[0081] It is understandable that the aforementioned target tool services may be external tools or services that need to be called to complete a specific sub-task. These tool services may be internally developed modules, third-party API interfaces, or services provided by cloud platforms.

[0082] Step S40 includes: calling the target tool service through the scheduling executor agent to execute the sub-task to be processed, generating the sub-task execution result; integrating the sub-task execution results to generate the final task execution result.

[0083] It should be understood that the final output generated after the subtask execution is completed may include the success or failure status of the subtask, the generated data, execution logs, etc. In this embodiment, the scheduling executor agent can generate the final task execution result by collecting and integrating the execution results of all subtasks in the task to be processed.

[0084] In the specific implementation, refer to Figure 4 , Figure 4 This is a complete flowchart of the task processing method in this application. Figure 4As shown, the user first inputs a task processing request into the smart terminal device. Upon receiving the request, the smart terminal device transmits it to a self-developed cloud platform via a communication module. The smart terminal device and the cloud platform communicate using a customized protocol, transmitting task descriptions, status synchronization, and result feedback. Incremental updates and priority transmission mechanisms are employed to reduce bandwidth consumption. After receiving the task processing request, the Agents classifier (the classifier agent) on the cloud platform analyzes the task to be processed, determining its complexity and domain characteristics. Based on this information, it evaluates the task's complexity. Then, the Agents classifier selects an appropriate task inference strategy based on the task's complexity to generate a thought chain corresponding to the task. Based on this thought chain, it develops a detailed execution plan and outputs the generated execution plan to the Agents scheduler (the scheduler agent). After receiving the task execution plan for a pending task, the Agents scheduler determines the execution order of each subtask within the task, as well as the computing resources and tools required for each subtask, based on the execution plan. It then schedules the required computing resources and tools to execute the subtasks according to this execution order. Furthermore, during task execution, the Agents post-processor (i.e., the post-processor agent) evaluates the subtasks, checking whether each subtask was successfully executed and whether the output meets expectations. Simultaneously, the Agents post-processor verifies the thought chain, comparing the final task execution result with the thought chain generated by the Agents scheduler to check whether the entire inference path was executed completely and correctly. If the verification passes, the Agents post-processor outputs the final task execution result to the smart terminal device. Conversely, if the verification fails, the Agents post-processor analyzes the root cause (such as tool call failure, incomplete data, or incorrect inference path) and triggers task refactoring to correct the task execution result.

[0085] It should be noted that, as Figure 4As shown, in this embodiment, when the Agents scheduler executes a task, it can call corresponding modules to complete the task execution. These modules may include, but are not limited to, search enhancement, knowledge base acquisition, code mathematical calculation, and container script execution. Search enhancement can refer to obtaining external information related to the task by calling a search engine or similar tools; knowledge base acquisition can refer to retrieving information from a predefined knowledge base to support the execution of the task; code mathematical calculation can refer to executing code snippets or performing mathematical calculations to solve programming or mathematical problems involved in the task; and container script execution can refer to running scripts or programs in an isolated container environment to ensure that the execution of the task does not affect other parts of the system.

[0086] It should also be noted that the computing resources of the cloud platform can include LLM management services (such as AI models like Qwen, DeepSeek, and GPT) and tool services. The LLM management service can interface with models from all third parties and privately deployed models, providing REST API calls, uniformly abstracting the models, and exposing a unified, encapsulated interface. Tool services can include enhancement tools (such as sandbox environments (virtual machine containers, shell environments), Long Short-Term Memory (LSTM) enhancement tools like RAG, and browsers (such as web crawler services)) and general-purpose tools (such as MCP protocol services and third-party interfaces).

[0087] Furthermore, to achieve the above objectives, this application also proposes a task processing method, referring to... Figure 5 , Figure 5 This is a flowchart illustrating the third embodiment of the task processing method of this application. Based on the above embodiments, the third embodiment of the task processing method of this application is proposed.

[0088] In this embodiment, the method is applied to an intelligent terminal device in a task processing system. The task processing system further includes a cloud platform, which integrates a classifier agent and a scheduler agent. The method includes: Step S10: Generate a task processing request based on the task processing instructions input by the user, and send the task processing request to the cloud platform; The cloud platform is configured to respond to the task processing request by generating a task execution plan for the pending task in the task processing request through the classifier agent; determining all tool services required to execute the pending task based on the task execution plan; and calling each of the tool services through the scheduler agent to execute the pending task and generate a task execution result.

[0089] It is understood that the method of this embodiment can be applied to a smart terminal device in the above-mentioned task processing system. This smart terminal device can be any device with data processing, program execution, and data acquisition functions, such as a smartwatch, smart ring, earphones, smart glasses, etc., and this embodiment does not impose any limitations on it. The executing entity of this method embodiment can be the above-mentioned smart terminal device to illustrate this embodiment and the following embodiments.

[0090] It should be emphasized that the specific implementation of the task processing method executed by the smart terminal device in this embodiment can be referenced with the specific implementation of the task processing method executed by the cloud platform, and this embodiment will not elaborate on this.

[0091] In addition, refer to Figure 6 , Figure 6 This is a schematic diagram of the structure of a smart terminal device in the hardware operating environment involved in the embodiments of this application.

[0092] like Figure 6 As shown, the intelligent terminal device may include: a second processor 2001, such as a central processing unit (CPU), a second communication bus 2002, a second user interface 2003, a second network interface 2004, and a second memory 2005. The second communication bus 2002 is used to enable communication between these components. The second user interface 2003 may be connected to a display screen. Optionally, the second user interface 2003 may include a standard wired interface or a wireless interface; in this application, the wired interface of the second user interface 2003 may be a USB interface. The second network interface 2004 may optionally include a standard wired interface or a wireless interface (such as a Wireless-Fidelity (Wi-Fi) interface). The second memory 2005 may be a high-speed random access memory (RAM) or a non-volatile memory (NVM), such as a disk storage device. Optionally, the second memory 2005 may also be a storage device independent of the aforementioned second processor 2001.

[0093] Those skilled in the art will understand that Figure 6 The structure shown does not constitute a limitation on smart terminal devices and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0094] like Figure 6 As shown, the second memory 2005, which is identified as a computer storage medium, may include a second operating system, a second network communication module, a second user interface module, and a task processing program.

[0095] exist Figure 6 In the smart terminal device shown, the second network interface 2004 is mainly used to connect to the backend server and communicate data with the backend server; the second user interface 2003 is mainly used to connect to user devices (such as the smart wearable device mentioned above); the smart terminal device calls the task processing program stored in the second memory 2005 through the second processor 2001 and executes the steps of the task processing method provided in the third embodiment of this application.

[0096] Furthermore, this application also proposes a task processing system, which includes a cloud platform as described above and a smart terminal device as described above.

[0097] The specific implementation of the task processing system in this embodiment can be referred to the description of the above method embodiments, and will not be repeated in this embodiment.

[0098] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0099] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0100] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as a read-only memory image (ROM) / random access memory (RAM), magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0101] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A task processing method, characterized in that, The method is applied to a cloud platform in a task processing system, which also includes intelligent terminal devices. The cloud platform integrates a classifier agent and a scheduler agent. The method includes: In response to a task processing request sent by the smart terminal device, a task execution plan for the task to be processed in the task processing request is generated by the classifier agent. Based on the task execution plan, determine all the tools and services required to execute the task to be processed; The scheduling executor agent invokes each of the tool services to execute the tasks to be processed and generates task execution results.

2. The method as described in claim 1, characterized in that, The step of generating a task execution plan for the task to be processed in the task processing request through the classifier agent includes: The classifier agent evaluates the complexity of the tasks to be processed in the task processing request to determine the task complexity of the tasks to be processed. Determine the task reasoning strategy for the task to be processed based on the task complexity; Based on the task reasoning strategy, the task to be processed is reasoned to generate a task execution plan for the task to be processed.

3. The method as described in claim 2, characterized in that, The step of evaluating the complexity of the tasks to be processed in the task processing request through the classifier agent and determining the task complexity of the tasks to be processed includes: The classifier agent performs semantic analysis on the task text of the task to be processed in the task processing request to determine the key logical nodes and branch nodes of the key logical nodes in the task to be processed. Construct a directed graph of task execution logic corresponding to the task to be processed based on the key logic nodes and the branch nodes; The number of logical branches and the depth of dependencies of the task to be processed are determined based on the directed graph of the task execution logic. The complexity of the task to be processed is determined by evaluating the number of logical branches and the depth of the dependency relationship.

4. The method as described in claim 2, characterized in that, The step of reasoning about the task to be processed based on the task reasoning strategy and generating a task execution plan for the task to be processed includes: Based on the task reasoning strategy, reason about the task to be processed and generate the task execution thought chain of the task to be processed. The task to be processed is broken down according to the task execution thought chain to obtain several sub-tasks to be processed. Determine the task execution requirements corresponding to each subtask to be processed and the task dependencies between the subtasks to be processed; A task execution plan for the task to be processed is generated based on the task execution requirements and the task dependencies.

5. The method as described in claim 4, characterized in that, The step of determining all tool services required to execute the task to be processed according to the task execution plan includes: Based on the task execution plan, determine the target tool services required to execute each of the sub-tasks in the task to be processed; The step of invoking each of the tool services through the scheduling executor agent to execute the task to be processed and generating the task execution result includes: The scheduling executor agent invokes the target tool service to execute the sub-task to be processed, and generates the sub-task execution result; The results of the subtasks are integrated to generate the final task execution result.

6. The method as described in claim 5, characterized in that, The cloud platform also integrates a post-processor intelligent agent; after the step of integrating the sub-task execution results to generate the final task execution result, it further includes: The task execution result is logically verified through the task execution thought chain. If the logical verification fails, the task execution result will be corrected based on the logical verification result.

7. A task processing method, characterized in that, The method is applied to intelligent terminal devices in a task processing system, which also includes a cloud platform, in which a classifier agent and a scheduler agent are integrated. The method includes: A task processing request is generated based on the task processing instructions input by the user, and the task processing request is sent to the cloud platform; The cloud platform is configured to respond to the task processing request by generating a task execution plan for the pending task in the task processing request through the classifier agent; determining all tool services required to execute the pending task based on the task execution plan; and calling each of the tool services through the scheduler agent to execute the pending task and generate a task execution result.

8. A cloud platform, characterized in that, The cloud platform includes: a first memory, a first processor, and a task processing program stored on the first memory and executable on the first processor. When the task processing program is executed by the first processor, it implements the steps of the task processing method as described in any one of claims 1 to 6.

9. A smart terminal device, characterized in that, The intelligent terminal device includes: a second memory, a second processor, and a task processing program stored in the second memory and executable on the second processor. When the task processing program is executed by the second processor, it implements the steps of the task processing method as described in claim 7.

10. A task processing system, characterized in that, The system includes the cloud platform as described in claim 8 and the smart terminal device as described in claim 9.

Images