Intelligent agent management method and device based on artificial intelligence technology and related equipment

Abstract

The invention provides an intelligent agent management method and device based on an artificial intelligence technology and related equipment. The intelligent agent management method based on the artificial intelligence technology can be applied to a management platform for managing a plurality of intelligent agents. Specifically, after a target task is obtained, a first agent and a second agent can be determined according to the target task. The first agent and the second agent belong to agents managed by the management platform and can process at least one type of task based on an artificial intelligence technology. After the first intelligent agent and the second intelligent agent are determined, a dialogue chain can be generated, so that the first intelligent agent and the second intelligent agent are configured, and the first intelligent agent and the second intelligent agent process the target task according to the dialogue chain. Thus, even if the task to be completed has high complexity, the task can be completed through mutual cooperation of the multiple agents, and the requirement for completing the task with the high complexity through the agents is met.

Description

[0001] This application claims priority to Chinese patent application No. 202311253370.1 filed on September 26, 2023, and entitled “A data processing system, method, computing cluster and related equipment”, the entire contents of which are incorporated by reference into this application. Technical Field

[0002] The present application relates to the field of computer technology, and in particular to an intelligent agent management method, device and related equipment based on artificial intelligence technology. Background Art

[0003] An agent refers to an entity that can observe and make decisions in a given environment. For example, it can be implemented based on an artificial intelligence (AI) model. The agent has the ability to process data. In some scenarios, agents can be used to complete tasks instead of natural persons. For example, an agent integrated with a natural language processing (NLP) model can be used to replace a natural person to complete the task of writing an article. For another example, an agent integrated with an image recognition model can be used to replace a natural person to complete the task of finding key information in a picture. For another example, an agent integrated with a decision-making model can be used to replace a natural person to complete the task of making a decision.

[0004] However, the intelligent agent can only complete some relatively simple tasks and cannot complete tasks of higher complexity. Summary of the invention

[0005] In view of this, the embodiment of the present application provides an agent management method based on artificial intelligence technology, which is used to use models to complete tasks with high complexity. The present application also provides corresponding devices, computing device clusters, computer-readable storage media, and computer program products.

[0006] In the first aspect, the present application provides an agent management method based on artificial intelligence technology, which can be applied to a management platform that manages multiple agents so as to process more complex target tasks. Moreover, each agent managed by the management platform can process at least one type of task based on artificial intelligence technology. Specifically, after obtaining the target task, a first agent and a second agent can be determined according to the target task. The first agent and the second agent belong to the agents managed by the management platform, and both can process at least one type of task based on artificial intelligence technology. After determining the first agent and the second agent, a dialogue chain can be generated to obtain a workflow for the first agent and the second agent to process the target task, thereby configuring the first agent and the second agent so that the first agent and the second agent process the target task according to the dialogue chain. In other words, for more complex target tasks, multiple agents can be determined and a dialogue chain can be generated according to task requirements. In this way, each agent only needs to process a part of the target task, and does not need to process the completed target task, and the agent does not need to have the ability to process complex tasks. In this way, multiple agents can be used to cooperate with each other, thereby making full use of the ability of the agent to process tasks and processing more complex tasks. In this way, even if the task to be completed is of high complexity, it can be completed through the cooperation between multiple intelligent agents, meeting the demand for completing tasks of high complexity through intelligent agents.

[0007] In some possible implementations, the first agent and the second agent may be determined by parsing the target task. Specifically, the target task may be parsed first to determine the subtask information of the target task. The subtask information of the target task includes the information of each subtask in the multiple subtasks required to complete the target task. That is, by parsing the target task, the target task with higher complexity may be split into multiple subtasks with lower complexity. The subtask may be, for example, a type of task that the agent can process based on artificial intelligence technology. In this way, the agent used to process the subtask may be determined based on the information of each subtask. That is, if the multiple subtasks include the first subtask and the second subtask, the first subtask is a first type of task, and the second subtask is a second type of task. Then, according to the information of the first subtask in the subtask information, the first agent that can be used to process the first type of task can be used as the agent for processing the first subtask from among the multiple agents managed by the management platform. Similarly, according to the information of the second subtask in the subtask information, the second agent that can be used to process the second type of task can be used as the agent for processing the second subtask from among the multiple agents managed by the management platform. In this way, the complex target task is divided into type tasks that can be handled by a single agent, which provides a basis for multiple agents to cooperate with each other to handle the target task. In addition, according to the ability of the agent to handle the type task, the appropriate agent is selected for each subtask, which can also ensure that the selected agent has the ability to handle the target task, thus ensuring the processing quality of the target task.

[0008] In some possible implementations, the subtask information includes not only the information of the subtasks, but also the information between the subtasks, and the information between the subtasks may include the association relationship between the subtasks. Accordingly, the management platform can generate a dialogue chain based on the association relationship between the above subtasks, and specifically can generate a workflow for processing subtasks between agents in the dialogue chain. Specifically, after determining the first agent and the second agent, the workflow for the first agent and the second agent to process the target task can be determined based on the association relationship between the subtasks processed by the first agent (i.e., the first subtask) and the subtasks processed by the second agent (i.e., the second subtask). In this way, the workflow for the agent to process the target task is determined based on the association relationship between the subtasks processed by the agent, so that the process of the agent processing the target task can be matched with the process of the subtask, thereby ensuring the processing quality of the target task.

[0009] In some possible implementations, the dialogue chain may also include a workflow for a single agent to process a subtask. The workflow for a single agent to process a subtask indicates the number of iterations in the process of the agent processing the corresponding subtask. When an agent processes the workflow of a subtask, the number of iterations in the process of the agent processing the subtask can be determined based on the agent's ability information for processing type tasks and the information of the subtasks that need to be processed. For example, for a first agent, the number of iterations in the process of the first agent processing the first subtask can be determined based on the first agent's ability information for processing the first type of task and the information of the first subtask. In this way, iterative processing based on the number of iterations can ensure that the agent's processing of the subtask meets the requirements of the target task and ensures the processing quality of the target task.

[0010] In some possible implementations, the workflow for a single agent to process a subtask may also be user-configured. Specifically, the management platform may also obtain workflow configuration information. The workflow configuration information may indicate the target number of iterations for a subtask. Based on the workflow configuration information, the number of iterations of the agent corresponding to the subtask may be determined as the target number of iterations to determine the workflow for the agent to process the subtask. For example, if the workflow configuration information is used to indicate the target number of iterations for the second subtask, then the workflow for the second agent to process the second subtask may be determined based on the target number of iterations. The workflow for the second agent to process the second subtask indicates that the number of iterations during the process of the second agent processing the second subtask is equal to the target number of iterations. In this way, the user can flexibly set the number of iterations of one or more agents in the dialogue chain so that the processing of the target task meets their own needs.

[0011] In the implementation described above, the agent can be determined based on the subtask information. In other possible implementations, the agent can also be determined based on the target task. Specifically, multiple agent combinations can be preset, each agent combination includes at least one agent, and the agents in the agent combination can be used to process at least one type of task. The management platform can be used to manage the above-mentioned preset multiple agent combinations. After obtaining the target task, the target agent combination can be selected from multiple preset agent combinations according to the target task. The target agent combination may include a first agent and a second agent. In this way, it can be ensured that the selected agent combination has the ability to process the target task and the processing quality of the target task is guaranteed.

[0012] In some possible implementations, the agent can also be set by the user. Specifically, the management platform can also determine the third agent based on the target task. For example, the third agent can be determined based on the task configuration information of the target task. The third agent can be an agent set when the user creates the target task and used to process the target task. Accordingly, when planning a dialogue chain, planning can be performed based on the first agent, the second agent, and the third agent. In this way, the user can flexibly set the agent used to process the target task so that the processing of the target task meets their own needs.

[0013] In some possible implementations, in order to ensure the processing quality of the target task, the user can also perform verification during the processing of the target task. Specifically, after the management platform configures the first agent and the second agent, it can provide intermediate processing information. The intermediate processing information is the intermediate information obtained in the process of processing the target task. For example, the management platform can send the intermediate processing information to the client so that the client can display the intermediate processing information to the user. If the user confirms that the intermediate processing information is correct, the user can trigger a confirmation operation on the intermediate processing information displayed by the client. The management platform can continue to process the target task based on the confirmation operation on the intermediate processing information. In this way, the user confirms the intermediate processing information, which can ensure that the processing of the target task is controllable and avoid situations such as "AI hallucinations".

[0014] In some possible implementations, there may be multiple agents with similar functions. Accordingly, the management platform can select a suitable agent to process the target task based on the evaluation index parameters of the agent. For example, there may be multiple agents that have the ability to process the first type of task, then the management platform can select the agent with the best evaluation index parameters (such as the first agent) from the above multiple agents to process the target task. In some possible implementations, the evaluation index parameters of the target task can be adjusted according to the completion of the target task. Specifically, in the process of processing the target task, or after processing the target task, the management platform can obtain feedback information. The feedback information can be used to indicate the completion of the target task by the first agent. Then, the evaluation index parameters of the first agent can be adjusted according to the feedback information. For example, if the first agent has a better completion of the target task, the evaluation index parameters of the first agent can be adjusted in a more optimal direction. In this way, feedback evaluation of the agent can optimize the strategy of the management platform to select the agent.

[0015] In a second aspect, the present application provides an agent management device based on artificial intelligence technology, wherein the device is applied to a management platform, the management platform is used to manage multiple agents, each of the multiple agents is used to process at least one type of task based on artificial intelligence technology, and the device includes: an acquisition module, used to acquire a target task; an agent determination module, used to determine a first agent and a second agent according to the target task, the first agent and the second agent belonging to the multiple agents; a dialogue chain generation module, used to generate a dialogue chain, the dialogue chain is used to indicate the workflow of the first agent and the second agent to process the target task; an agent configuration module, used to configure the first agent and the second agent so that the first agent and the second agent process the target task according to the dialogue chain.

[0016] In some possible implementations, the agent determination module is specifically used to parse the target task and determine subtask information of the target task, wherein the subtask information includes information of multiple subtasks required to complete the target task, and the multiple subtasks include a first subtask and a second subtask; determine the first agent based on the information of the first subtask, wherein the first subtask is a first type of task among the at least one type of task, and the first agent is used to process the first type of task; determine the second agent based on the information of the second subtask, wherein the second subtask is a second type of task among the at least one type of task, and the second agent is used to process the second type of task.

[0017] In some possible implementations, the subtask information is used to describe the association relationship between the multiple subtasks, and the dialogue chain generation module is specifically used to determine the workflow of the first agent and the second agent in processing the target task based on the association relationship between the first subtask and the second subtask.

[0018] In some possible implementations, the dialogue chain further includes a workflow of the first agent processing the first subtask, where the workflow of the first agent processing the first subtask indicates the number of iterations in the process of the first agent processing the first subtask.

[0019] The dialogue chain generation module is also used to determine the number of iterations in the process of the first agent processing the first subtask based on the ability information of the first agent processing the first type of task and the information of the first subtask.

[0020] In some possible implementations, the dialogue chain also includes a workflow for the second agent to process the second subtask, and the acquisition module is further used to obtain workflow configuration information, and the workflow configuration information is used to indicate a target number of iterations for the second subtask; the dialogue chain generation module is further used to determine the workflow for the second agent to process the second subtask based on the target number of iterations, and the workflow for the second agent to process the second subtask indicates that the number of iterations in the process of the second agent processing the second subtask is equal to the target number of iterations.

[0021] In some possible implementations, the management platform is also used to manage multiple preset agent combinations, the agents in the agent combination belong to the multiple agents, and the agent determination module is used to determine a target agent combination from the multiple preset agent combinations according to the target task, and the target agent combination includes the first agent and the second agent.

[0022] In some possible implementations, the agent determination module is also used to determine a third agent in response to an agent setting operation, and the dialogue chain is specifically used to instruct the first agent, the second agent, and the third agent to process the workflow of the target task.

[0023] In some possible implementations, the device also includes a confirmation module, which is used to provide intermediate processing information, where the intermediate processing information is intermediate information obtained during the processing of the target task; in response to a confirmation operation on the intermediate processing information, continue to process the target task.

[0024] In some possible implementations, the device also includes a feedback module, which is used to obtain feedback information, and the feedback information is used to indicate the completion status of the target task by the first agent; the evaluation index parameters of the first agent are adjusted according to the feedback information, and the evaluation index parameters of the first agent are used to select the first agent to process the task.

[0025] In a third aspect, the present application provides a computing device cluster, wherein the computing device includes at least one computing device, and the at least one computing device includes at least one processor and at least one memory; the at least one memory is used to store instructions, and the at least one processor executes the instructions stored in the at least one memory, so that the computing device cluster executes the intelligent agent management method based on artificial intelligence technology in the above-mentioned first aspect or any possible implementation of the first aspect. It should be noted that the memory can be integrated into the processor or can be independent of the processor. The at least one computing device may also include a bus. The processor is connected to the memory via a bus. The memory may include a readable memory and a random access memory.

[0026] In a fourth aspect, the present application provides a computer-readable storage medium, wherein instructions are stored in the computer-readable storage medium, and when the computer-readable storage medium is executed on at least one computing device, the at least one computing device executes the method described in the first aspect or any one of the implementations of the first aspect.

[0027] In a fifth aspect, the present application provides a computer program product comprising instructions, which, when executed on at least one computing device, enables the at least one computing device to execute the method described in the first aspect or any one of the implementations of the first aspect.

[0028] Based on the implementations provided in the above aspects, this application can also be further combined to provide more implementations. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0030] Figure 1a A schematic diagram of an application scenario provided in an embodiment of the present application;

[0031] Figure 1b Another schematic diagram of an application scenario provided in an embodiment of the present application;

[0032] Figure 2a A schematic diagram of a flow chart of an intelligent agent management method based on artificial intelligence technology provided in an embodiment of the present application;

[0033] Figure 2b Another flowchart of an intelligent agent management method based on artificial intelligence technology provided in an embodiment of the present application;

[0034] Figure 3 A schematic diagram of a process for processing a target task provided in an embodiment of the present application;

[0035] Figure 4 A schematic diagram of the structure of an intelligent agent management device based on artificial intelligence technology provided in an embodiment of the present application;

[0036] Figure 5 A schematic diagram of a structure of a computing device provided in an embodiment of the present application;

[0037] Figure 6 A schematic diagram of a structure of a computing device cluster provided in an embodiment of the present application;

[0038] Figure 7 A schematic diagram of an implementation method of a computing device cluster provided in an embodiment of the present application. DETAILED DESCRIPTION

[0039] The scheme in the embodiments provided in this application will be described below in conjunction with the drawings in this application.

[0040] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances. This is just a way of distinguishing objects with the same attributes when describing the embodiments of this application.

[0041] First, the terms involved in the embodiments of the present application are introduced.

[0042] Cloud service system: The cloud service system includes a client and a server. Users can publish tasks through the client. Tasks published by the client can be processed by the server.

[0043] Management platform: integrated into the server side of the cloud service system, with the ability to manage multiple agents. After receiving the tasks issued by the client, the management platform can call one or more agents to process the tasks.

[0044] Task: A task to be processed that is published by a user through a client (or other means). Optionally, a user may publish a task in the form of relevant information about the task. For example, a user may enter relevant information about the task on a client. The server may receive the relevant information about the task and process the task. In an embodiment of the present application, the analysis and / or processing of the task may include analyzing and / or processing the relevant information about the task.

[0045] AI agent: refers to an entity that can observe and make decisions in a given environment. For example, it can be implemented based on an artificial intelligence (AI) model. The agent has the ability to process data. That is, the agent can output a specific type of data based on a certain type of input data. In an embodiment of the present application, the agent can process at least one type of task based on artificial intelligence technology.

[0046] Dialogue chain: refers to the rules for agents to exchange information when multiple agents are called to process tasks. Dialogue chains indicate the workflow of agents when processing tasks. Dialogue chains can indicate workflows between multiple agents or between single agents.

[0047] Workflow: refers to the abstract summary of the business rules between the processing flow and its various operation steps. In the embodiment of the present application, the workflow indicated by the dialogue chain is used to describe the rules for information exchange between one or more agents when processing a task. The workflow between multiple agents refers to the connection relationship between multiple agents. For two interconnected workflows, the output data of one agent can be used as the input data of another agent. The dialogue chain between a single agent can also be called the iteration number of the agent, which indicates the number of times a single agent performs iterative processing. Iterative processing refers to re-inputting the output data of an agent as its own input data into the agent. Through iterative processing, the accuracy of data processing can be improved.

[0048] With the development of computer technology, intelligent agents including AI models have been widely used in many fields. Due to the integration of AI models, intelligent agents have the characteristics of high efficiency and fast processing speed, and can replace natural persons to complete corresponding tasks in some application scenarios. For example, in the fields of article writing, code development and image processing, it has great advantages to replace natural persons with intelligent agents.

[0049] In particular, in order to meet the needs of some application scenarios, the capabilities of the agent can be enhanced. The enhanced agent may have stronger processing capabilities in some aspects and can better handle certain types or certain types of tasks. For example, for natural language processing tasks, the NLP capabilities of the agent can be enhanced, and the enhanced agent can better process natural language. For another example, for image processing tasks, the image processing capabilities of the agent can be enhanced, and the enhanced agent can better process images. For another example, for decision-making tasks, the decision-making capabilities of the agent can be enhanced, and the enhanced agent can better make decisions. The enhanced model can be implemented, for example, by optimizing the model structure and adjusting the training data set, which will not be repeated here.

[0050] However, due to the limitations of technological development, the processing capabilities of intelligent agents can only be strengthened in one or more directions, and it is impossible to strengthen the processing capabilities of intelligent agents in all directions. This results in the capabilities of intelligent agents being relatively single and only suitable for completing certain types of tasks. For example, an intelligent agent with NLP capabilities can complete tasks related to natural language processing, but it is difficult to complete tasks related to image processing; an intelligent agent with image processing capabilities can complete tasks related to image processing, but it is difficult to complete tasks related to natural language processing; a model with decision-making capabilities can make decisions, but it is difficult to complete tasks related to image processing or natural language processing.

[0051] In actual scenarios, there may be some tasks with a high degree of complexity. Such tasks may include multiple subtasks. Different subtasks may require different processing capabilities. Therefore, such tasks with a high degree of complexity may not be completed by a single model.

[0052] For example, the task of "developing code according to requirements" can be divided into multiple subtasks such as "understanding code development requirements", "developing code", "testing developed code", "adjusting code according to test results" and "generating instruction documents based on developed code". Different subtasks require different processing capabilities and cannot be implemented by a unified agent. For example, the subtask "understanding code development requirements" may require the agent to have natural language processing capabilities. The subtask "developing code" may require the agent to have the ability to automatically generate code, and the subtask "testing developed code" may require the agent to have decision-making capabilities to select appropriate test cases. Due to the limitations of technological development, the above functions are difficult to integrate into one agent.

[0053] It can be seen that the current intelligent agents are only suitable for completing tasks with lower complexity and cannot complete tasks with higher complexity.

[0054] Based on this, the embodiment of the present application provides an agent management method based on artificial intelligence technology, which can select multiple agents according to the target task to be processed so as to process more complex target tasks. Specifically, for a management platform that manages multiple agents, after obtaining the target task, the first agent and the second agent can be determined according to the target task. The first agent and the second agent belong to the agents managed by the management platform, and both can process at least one type of task based on artificial intelligence technology. After determining the first agent and the second agent, a dialogue chain can be generated to obtain the workflow of the first agent and the second agent processing the target task, thereby configuring the first agent and the second agent so that the first agent and the second agent process the target task according to the dialogue chain. In other words, for more complex target tasks, multiple agents can be determined according to task requirements and a dialogue chain can be generated. In this way, each agent only needs to process a part of the target task, and does not need to process the completed target task, and does not need the agent to have the ability to process complex tasks. In this way, multiple agents can be used to cooperate with each other, so as to make full use of the ability of the agent to process tasks and process more complex tasks. In this way, even if the task to be completed is of high complexity, it can be completed through the cooperation between multiple intelligent agents, meeting the demand for completing tasks of high complexity through intelligent agents.

[0055] Next, various non-limiting specific implementations of the process of processing tasks are described in detail.

[0056] First, taking the cloud service scenario as an example, the application scenario of the intelligent agent management method based on artificial intelligence technology provided by the embodiment of the present application is introduced.

[0057] See also Figure 1a , Figure 1a A schematic diagram of an application scenario of an intelligent agent management method based on artificial intelligence technology provided in an embodiment of the present application. Figure 1a The application scenario shown includes a user 10, a terminal device 20, a cloud service system 30, an agent 41, an agent 42, and an agent 43. The cloud service system 30 includes a client 31 and a server 32. The client 31 may be software running on the terminal device 20. The server 32 includes a management platform 321. The management platform 321 is used to manage the agents 41, 42, and 43.

[0058] The user 10 can call the task processing service provided by the cloud service system 30 through the client 31. Specifically, the user 10 can input the relevant information of the target task on the client 31. The client 31 can send the relevant information of the target task to the server 32. The management platform 321 in the server 32 can execute the intelligent agent management method based on artificial intelligence technology provided in the embodiment of the present application to call one or more intelligent agents to process the target task proposed by the user 10. Among them, the one or more intelligent agents can include any one or more of the intelligent agents 41, 42 and 43, and can also include other intelligent agents.

[0059] Optionally, the agents managed by the management platform 321 (such as the above-mentioned agents 41, 42, and 43) can be integrated into the cloud service system 30, or can be independent of the cloud service system 30. For example, the agent 41 can run on the server 32 of the cloud service system 30. The agent 42 can be independent of the cloud service system 30. When the management platform 321 manages the agent 42, or calls the agent 42 to process a task, the management platform 321 can exchange information with the agent 42 through the network.

[0060] In an embodiment of the present application, the cloud service system 30 can provide multiple types of cloud services to the user 10. Moreover, the multiple types of cloud services include task processing type cloud services. Optionally, the cloud service system 30 can also be a cloud computing system, and accordingly, the cloud services provided to the user 10 by the cloud service system 30 can include cloud computing type services. Optionally, the management platform 321 can be implemented in software. For example, it can be implemented in software running on one or more computing devices. If the management platform 321 runs on multiple computing devices, the multiple computing devices can belong to the same area or different areas.

[0061] The following is a detailed introduction to the specific implementation methods of the intelligent agent management process based on artificial intelligence technology.

[0062] See also Figure 2a , Figure 2a A flow chart of an intelligent agent management method based on artificial intelligence technology provided in an embodiment of the present application. The method can be specifically executed by a management platform. The management platform can be run on the server side of the cloud service system to call the intelligent agent to process the task. The management platform can also be called an intelligent agent management device based on artificial intelligence technology. Optionally, the method can be applied to the above Figure 1a The method may be used in the application scenario shown in FIG. 1 , or may be used in other applicable application scenarios. Figure 1a In the application scenario shown, the method can be implemented by the management platform 321. Figure 2aThe agent management method based on artificial intelligence technology is implemented by the management platform 321. The management platform can be as follows: Figure 1b The following is applied to Figure 1a The application scenario shown is used as an example for explanation.

[0063] Specifically, Figure 2a The agent management method based on artificial intelligence technology shown may specifically include:

[0064] S201: Obtain target tasks.

[0065] If the user wants to call the agent to process the target task, the user can send the target task to the management platform through the client. The management platform can obtain the target task, generate a dialogue chain based on the target task, and call the agent to process the target task.

[0066] In an embodiment of the present application, the management platform obtains the target task, which may include the management platform obtaining relevant information of the target task. The relevant information of the target task may be relevant information input by the user through the client to describe the target task. Optionally, the relevant information of the target task may also be referred to as task configuration information of the target task.

[0067] The task configuration information of the target task is used to describe the relevant configuration of the target task. Based on the task configuration information, the management platform can determine what kind of task the target task to be processed is and what kind of capabilities are required to process the target task. For example, the task configuration information of the target task can be used to describe the input of the target task and the output of the target task, and can also be used to describe the requirements in the process of processing the target task. Optionally, if the target task includes multiple links, the task configuration information of the target task can also include the configuration information corresponding to each link. The configuration information corresponding to the link can be used to describe the input of this link, the output of this link, and the requirements in the process of processing the task of this link.

[0068] In an embodiment of the present application, the task configuration information of the target task may include the to-be-processed data corresponding to the target task so that the server can process the target task. In addition, the task configuration information of the target task may also include multiple configuration information of the target task, for example, any one or more of the workflow configuration information of the target task, the role configuration information of the target task, and the agent configuration information of the target task. The following will be introduced respectively.

[0069] First, the workflow configuration information is introduced.

[0070] The workflow configuration information of the target task is used to describe the workflow of the target task. The workflow of the target task refers to the subtasks that need to be completed by multiple participants in order to process the target task, as well as the rules for information interaction between participants. Optionally, if the target task is originally completed by multiple natural persons, then the above-mentioned "participants" can be natural persons, and the workflow information of the target task can be the subtasks that multiple natural persons need to complete respectively when multiple natural persons process the target task, as well as the rules for interaction between natural persons. In this way, the workflow configuration information in the process of natural persons processing the target task is used as the task configuration information of the target task, so that the intelligent agent determined in the subsequent steps can simulate the natural person to process the target task.

[0071] For example. Assuming that the target task is a task of the code development type, the task configuration information of the target task may include the workflow configuration information of the task of the code development type. Specifically, it may include the relevant information of the personnel required in the code development process, the work that each person needs to do in the code development process, and the interaction information between each person in the code development process. Among them, the personnel information required in the code development process can be used to describe the personnel in the code development process, for example, it can be used to describe personnel such as executives, product managers, architects, code writers, testers, and user interface (UI) designers. The work that each person needs to do in the code development process can be used to describe what work each of the above multiple personnel needs to complete, for example, it can be used to describe that the product manager needs to perform demand analysis, the architect needs to give the code architecture, the code writer needs to write code, and the tester needs to test the code. The interaction information of each person in the code development process can be used to describe the interaction process between the above multiple personnel, for example, it can be used to describe the interaction process of the product manager sending requirements to the code writer, the interaction process of the code writer sending the code to the tester after writing the code, and the interaction process of the tester sending the test result to the code writer after completing the test.

[0072] Optionally, the workflow configuration information may also include process attribute configuration information. The process attribute configuration information may be used to describe attributes related to the processing flow of the target task, and may represent the user's specific requirements for the processing flow of the target task.

[0073] In some possible implementations, the process attribute configuration information may include information describing specific rules for information exchange between roles or agents. That is, the user may set the information exchange rules between two agents in the process flow. Alternatively, the user may set the number of times the agents exchange information, or set the specification of prompt information used in the information exchange process between agents.

[0074] For example, in the process of code development by a team of natural persons, multiple interactions may be required between code writers and testers to identify possible problems in the code and optimize it. Accordingly, the user can instruct the agent used to complete the subtasks completed by the "code writer" in the target process through the process attribute configuration information to interact multiple times with the agent used to complete the subtasks completed by the "tester". Specifically, the process attribute configuration information may include dialogue turn information, which corresponds to the rounds of interaction between the above two agents, and may specifically include the number of dialogues between the agent used to complete the subtasks completed by the "code writer" and the agent used to complete the subtasks completed by the "tester".

[0075] In some possible implementations, the process attribute configuration information may include requirement information for describing the overall processing flow. That is, if the user has special requirements for the processing flow, the user may carry his or her requirements in the process attribute configuration information and send it to the server. The server may analyze the user's requirements, thereby determining the interaction relationship between agents according to the user's requirements and establishing the target flow.

[0076] For example, in the traditional code development process, there are many development methods such as waterfall development, extreme programming development, and agile development. The interaction methods between natural persons in different development methods may be different, and the information transmitted in the interaction may also be different. When publishing a target task, the user can select a development method according to their own needs. The process attribute configuration information may include the identifier of the development method selected by the user. When creating a processing flow, the user can deploy the interaction relationship between intelligent agents according to the identifier of the development method selected by the user, and define the information transmitted by the intelligent agents in the interaction.

[0077] The following describes the role configuration information.

[0078] In an embodiment of the present application, the determination of the agent may rely on the role. That is, the management platform may first determine the role played by the agent, and then select a suitable agent to play the corresponding role. In some possible implementations, the above-mentioned role may be obtained by parsing the workflow configuration information or other configuration information of the target task by the server. Alternatively, in some other possible implementations, the task configuration information of the target task may also include role configuration information. The role configuration information is the relevant information given by the user who publishes the target task to describe one or more specific roles. In this way, the user can freely set the role corresponding to the target task, which improves the flexibility of processing the target task and enhances the user experience.

[0079] That is, if the user wants to specify the characteristics of a certain role (or multiple roles) in the processing flow of the target task, the task configuration information published by the user may include role configuration information. Role configuration information may be used to describe the attribute information of the role, indicating the user's specific requirements for the role. Optionally, role configuration information may be used to describe the capabilities of the role, or to describe the hierarchical relationship between roles.

[0080] For example. Assuming that the target task is a code development type task, the role may include a role responsible for code development. In order to improve the efficiency of code development and the performance of the developed code, the task configuration information of the target task may include configuration information for the role responsible for code development, for example, information describing that the role "has strong code development capabilities" or information describing that the role "has rich software development experience in a certain programming language". For another example, in the above code development process, the role corresponding to the "executive officer" may be responsible for coordinating and scheduling various personnel in the code development process, then the role configuration information of the target task may include information describing that the role "has the ability to coordinate and schedule", and for another example, assuming that the target task is a task of writing legal documents, the target task may include a role corresponding to a junior lawyer, a role corresponding to an intermediate lawyer, and a role corresponding to a senior lawyer, then the role configuration information of the target task may include information describing that "the intermediate lawyer has the ability to judge the documents given by the junior lawyer", and information describing that "the senior lawyer has the ability to judge the documents given by the intermediate lawyer and the junior lawyer".

[0081] The agent configuration information is introduced below.

[0082] In an embodiment of the present application, the target task can be processed by one or more agents. In some possible implementations, the above-mentioned agent can be obtained by parsing the task configuration information of the target task by the server. Alternatively, in some other possible implementations, the user can specify the specific information of the agent. In an embodiment of the present application, the information specified by the user for describing the agent can be referred to as agent configuration information. In this way, the user can freely set the role of processing the target task, which improves the flexibility of processing the target task and enhances the user experience.

[0083] That is, if the user wants to specify a certain (or more) agent or agent characteristics in the processing flow of the target task, the task configuration information published by the user can include agent configuration information. Agent configuration information can be used to specify an agent or to describe the attribute information of an agent, indicating the user's specific requirements for the agent.

[0084] In the embodiment of the present application, the user can configure the agent configuration information by triggering the agent setting operation. That is, the user can trigger the agent setting operation on the client to set one or more specific agents in the process of processing the target task.

[0085] The above describes the task configuration information of the target task. The following describes how to obtain the task configuration information of the target task.

[0086] In the embodiment of the present application, the task configuration information of the target task can be set by the user. Specifically, if the server provides the user with the function of processing tasks in the form of cloud services, the user can set the task configuration information of the target task on the client. After the user confirms to use the cloud service for processing tasks, the client can send the task configuration information of the target task to the server. That is, Figure 1a In the implementation shown, the user 10 can input the task configuration information of the target task on the client 31. The client 31 can send the task configuration information of the target task to the server 32 through the network, so that the management platform 321 in the server 32 can obtain the target task.

[0087] In the embodiment of the present application, the user can set the task configuration information of the target task on the interface of the client.

[0088] For example, if the task configuration information of the target task includes workflow configuration information, the user can input or import the workflow configuration information on the client interface. Specifically, the user can pre-plan the workflow corresponding to the target task, and then import the workflow into the client so that the client obtains the workflow configuration information of the target task. Optionally, the user can import the workflow file into the client to obtain the workflow configuration information of the target task. Alternatively, the client can also display a workflow editing interface, and the user can mark the workflow in the workflow editing interface. Alternatively, the user can also describe the workflow corresponding to the target task in natural language or non-natural language. The information input by the user to describe the workflow corresponding to the target task can be regarded as the aforementioned workflow configuration information.

[0089] For another example, if the task configuration information of the target task includes process attribute configuration information, the user can input the process attribute configuration information on the client. Specifically, the client can display an input box to the user, and receive the process attribute configuration information input by the user through the input box. Alternatively, one or more optional process attribute configuration information can be pre-set. If the user wants to call an agent to process the target task, the client can display a menu to the user, and the menu includes the above-mentioned pre-set optional process attribute configuration information. The user can select the process attribute configuration information through the menu. For example, if the server provides software development services, the process attribute configuration information can be set in advance according to the above-mentioned multiple development methods. When the user wants to use an agent for software development, the user can select the desired development method from the pre-set development methods in the menu.

[0090] For another example, if the task configuration information of the target task includes role configuration information, the user can input the role configuration information on the client. Specifically, the client can also display an input box, and the user can manually input a natural language description of the role through the input box. The natural language description corresponds to the role configuration information. Alternatively, the user can also set the role configuration information based on the workflow configuration information. For example, the user can set the role configuration information in the workflow editing interface. Alternatively, one or more optional role configuration information can be pre-set. If the user wants to call the agent to process the target task, the client can display a menu to the user, and the menu includes the above-mentioned pre-set optional role configuration information. The user can select the role configuration information through the menu.

[0091] For another example, if the task configuration information of the target task includes agent configuration information, the user can input the agent configuration information on the client. Specifically, the client can display a menu to the user, and the menu includes relevant information of one or more optional agents. The user can set the agent configuration information through the menu. Alternatively, the agent configuration information can also be set in other ways. The specific setting method can be found in the introduction of the role configuration information inconsistency above, which will not be repeated here.

[0092] It should be noted that the above implementation method of obtaining the task configuration information of the target task is only an example. In actual scenarios, the task configuration information of the target task can also be obtained by other methods.

[0093] S202: Determine the first agent and the second agent according to the target task

[0094] After obtaining the target task, the management platform can determine the first agent and the second agent according to the target task, so as to use the first agent and the second agent to process the target task. The first agent and the second agent are both agents managed by the management platform, and both the first agent and the second agent can be used to process at least one type of task based on artificial intelligence technology.

[0095] Optionally, the type of tasks that the first agent can handle can be the same as or different from the type of tasks that the second agent can handle. In the embodiment of the present application, to process the target task, it is necessary to process the type of tasks that the first agent can handle and the type of tasks that the second agent can handle. Therefore, the management platform can complete the processing of the target task by configuring the first agent and the second agent to respectively process various types of tasks corresponding to each other when processing the target task.

[0096] Among them, the first agent is used to process at least one type of task based on artificial intelligence technology, which means that the first agent has the ability to process the above-mentioned "at least one type of task". In the embodiment of the present application, the first agent can be said to correspond to the above-mentioned "at least one type of task". That is, the agent has the ability to process its corresponding type of task. For example, if the first agent integrates an NLP model, then the first agent can be used to process natural language recognition type tasks or natural language generation type tasks based on artificial intelligence technology; if the first agent integrates an image recognition model, then the first agent can be used to process image recognition type tasks based on artificial intelligence technology.

[0097] Optionally, the correspondence between the agent and the type of task can be obtained when registering the agent. For an introduction to this part, please refer to the following text and will not be repeated here.

[0098] The following describes how the management platform determines the first intelligent agent and the second intelligent agent.

[0099] Implementation method (1): The management platform determines the first agent and the second agent by selecting a combination of agents.

[0100] In a first implementation, the first agent and the second agent may belong to the same agent group. The management platform may determine an agent combination including the first agent and the second agent based on the target task.

[0101] Specifically, multiple agent combinations can be preset. Each agent combination can include multiple agents. Moreover, each agent combination can be used to process a preset type of task. The preset type can be a pre-configured function of the management platform. The agent combination is an agent combination obtained by analyzing the preset type of task. The agent corresponding to the agent combination can process the preset type of task corresponding to the agent combination. Accordingly, after obtaining the target task, an agent combination matching the target task can be selected from multiple preset agent combinations according to the target task, so as to use the agent combination to process the target task. In this way, it can be ensured that the selected agent combination has the ability to process the target task and the processing quality of the target task is guaranteed.

[0102] That is to say, assuming that the cloud service system provides users with task processing functions, then before providing users with type A task processing functions, type A tasks can be analyzed in advance to determine which types of agents are needed to process type A tasks, and then agent combination a can be established based on these agents. Similarly, before providing users with type B task processing functions, type B tasks can be analyzed in advance to determine which types of agents are needed to process type B tasks, and then agent combination b can be established based on these agents. In this way, if a user needs to process type A tasks, the management platform can select agent combination a from multiple preset agent combinations (including agent combination a and agent combination b). In this way, by setting the agent combination in advance according to the requirements of the task, the agent combination selected by the management platform has the ability to process the task.

[0103] Optionally, the agent combination may include the identifier of the first agent and the identifier of the second agent. That is, when setting the agent combination, one (or more) agents in the agent combination may be preset.

[0104] Or, optionally, the agents in the agent combination can also be selected by the management platform. Accordingly, when processing the target task, the management platform can first select the agent combination template according to the target task. There may be one or more items to be configured in the agent combination template. Each item to be configured corresponds to one or more optional agents. Then, the management platform can determine the specific value corresponding to each item to be configured in the agent combination template to obtain an agent combination including a specific agent. Among them, each item to be configured in the agent combination template corresponds to one or more tasks that need to be completed by the agent. The optional agent corresponding to the item to be configured has the ability to complete the task corresponding to the item to be configured.

[0105] Let's take an example. Suppose natural language processing is required in the process of processing the target task. Then the agent combination template corresponding to the target task can include item A to be configured. Item A to be configured corresponds to the above-mentioned "natural language processing" step, and the optional agents corresponding to item A to be configured all have the ability of natural language processing. When determining the agent combination corresponding to the target task, the management platform can select an agent from multiple agents as the agent actually corresponding to item A to be configured according to the actual situation of the target task, so as to use the agent to complete the relevant part of natural language processing in the process of processing the target task.

[0106] Regarding the method of selecting an agent from multiple optional agents, please refer to the introduction of the section on selecting an agent based on subtask information below, which will not be repeated here.

[0107] Implementation method (2): The management platform determines the first agent and the second agent based on the subtask information.

[0108] In the second implementation, the management platform can parse the target task to determine the first agent and the second agent. By parsing the target task, the management platform can determine what conditions are required to complete the target task, and thus determine the agent used to process the target task based on the conditions required to complete the task. As described above, the agent managed by the management platform can process at least one type of task based on artificial intelligence technology.

[0109] Specifically, when determining the intelligent agent for processing the target task, the management platform can first parse the target task, split the target into multiple subtasks, obtain the subtask information of the target task, and then determine the intelligent agent based on the subtask information of the target task.

[0110] The subtask information of the target task indicates the subtasks required to complete the target task, that is, the subtask information may include information about each subtask. The subtask information is used to describe the subtasks that the agent needs to complete. A subtask refers to a task that the agent needs to complete. Each subtask can be completed by an agent, and an agent can complete one or more subtasks. In other words, a subtask is a task that a single agent can complete.

[0111] If the target task includes multiple subtasks, the subtask information may also include information between the subtasks. The information between the subtasks may be used to describe the association relationship between the subtasks and to generate a dialogue chain. In other words, in order to complete the target task, each subtask needs to be completed separately according to the association relationship between the subtasks. Optionally, the association relationship between the subtasks may also include the information exchange process between the subtasks.

[0112] Optionally, if the task configuration information of the target task includes workflow configuration information, then the subtask information can be determined based on the workflow configuration information. Specifically, the information between subtasks in the subtask information can be used to determine based on the workflow configuration information. According to the previous introduction, the workflow configuration information can be used to describe the subtasks that a natural person needs to complete in the process of processing the target task. When determining the subtask information, the subtasks that need to be completed in the process of a natural person processing the target task can be referred to to determine the subtasks that need to be completed in the process of an intelligent agent processing the target task. Optionally, the subtask processed by each natural person can be determined as a subtask in the subtask information, or the subtasks completed by a single natural person can be split.

[0113] In some possible implementations, the server can determine the subtask information through an agent. That is, the management platform can be integrated with an agent or have the function of calling an agent, so as to use the agent to determine the subtask information corresponding to the target task. According to the previous introduction, the task configuration information of the target task may include relevant information based on natural language description. Accordingly, when determining the subtask information of the target task, an agent with NLP capabilities can be called to process the part of the task configuration information related to the natural language, and convert the task configuration information into a structured language before subsequent processing.

[0114] Take the example of the management platform calling the agent through the network to determine the subtask information. Specifically, the management platform can generate prompt information according to the task configuration information of the target task, and then send the prompt information to the agent. The agent can parse the task configuration information in the prompt information and return the subtask information to the management platform.

[0115] After determining the subtask information, the agent corresponding to each subtask can be determined according to the subtask information. The agent corresponding to the subtask is used to complete the subtask. That is, if the multiple subtasks obtained by splitting the target task include a first subtask and a second subtask, the management platform can determine the first agent according to the information of the first subtask, and determine the second agent according to the information of the second subtask. The first agent has the ability to complete the type of task corresponding to the first subtask, and the second agent has the ability to complete the type of task corresponding to the second subtask.

[0116] That is to say, the management platform can select an agent (i.e., the first agent) from multiple agents that has the ability to complete the type of task corresponding to the first subtask based on the information of the first subtask, and can select an agent (i.e., the second agent) from multiple agents that has the ability to complete the type of task corresponding to the second subtask based on the information of the second task.

[0117] For ease of explanation, the type of task corresponding to the first subtask is referred to as the first type of task, and the type of task corresponding to the second subtask is referred to as the second type of task. That is, the first subtask is a first type of task in at least one type of task, and the first agent is used to process the first type of task. The second subtask is a second type of task in at least one type of task, and the second agent is used to process the second type of task. That is, if a plurality of subtasks include a first subtask and a second subtask, the first subtask is a first type of task, and the second subtask is a second type of task. Then, based on the information of the first subtask in the subtask information, the first agent that can be used to process the first type of task can be selected from the multiple agents managed by the management platform as the agent for processing the first subtask. Similarly, based on the information of the second subtask in the subtask information, the second agent that can be used to process the second type of task can be selected from the multiple agents managed by the management platform as the agent for processing the second subtask.

[0118] In some possible implementations, the role corresponding to the subtask can be determined based on the information of the subtask, and then the agent can be determined based on the role. This is described in detail below.

[0119] Specifically, after determining the subtask information, the management platform can parse the subtask information to determine the condition information of multiple roles. The role refers to the role that completes the subtask, and the agent is used to play the role to complete the subtask corresponding to the role. The condition information of the role indicates the conditions required to play the role, that is, the conditions required to complete the subtask corresponding to the role. Optionally, the step of determining the conditions of multiple roles according to the subtask information can be implemented by the agent.

[0120] In the scenario where a natural person processes a target task, the natural person may play one or more roles to complete the subtasks corresponding to the roles. In the embodiment of the present application, an agent may play a role to complete the subtasks corresponding to the role. In other words, the condition information of the role is used to indicate the conditions that the agent playing the role needs to have. It is understandable that "playing a role" here means playing the role of the role in the process of processing the target task.

[0121] As described above, the task description information of the target task can include role configuration information. Accordingly, when determining the condition information of the role, it can be determined in combination with the role configuration information and the subtask information. That is, on the one hand, the condition information corresponding to the role can be determined based on the subtasks completed by the role, and on the other hand, the condition information corresponding to the role can be determined based on the instructions given by the user. In this way, the determined condition information has the ability to complete the subtasks that the role needs to complete and can meet the needs of the user.

[0122] After determining the condition information of multiple roles, the agent corresponding to the role can be determined according to the agent's condition information. In the process of processing the target task, the agent corresponding to the role can play the role to complete one or more subtasks required by the role. Figure 1a In the implementation shown, the management platform 321 can select an agent from agents 41, 42, and 43 to play the role corresponding to the condition information according to the condition information of the role. Optionally, the step of determining the agent according to the condition information of the role can be implemented by an agent. Optionally, the agent used to determine the agent according to the condition information of the role and the agent used to parse the target task to determine the subtask information can be the same agent or different agents.

[0123] In the implementation described above, the agent determined according to the condition information can meet the requirements of the condition information. In other words, the condition information indicates the conditions required to complete the subtasks completed by the agent, and the agent determined according to the condition information satisfies the conditions indicated by the condition information. Therefore, the agent has the ability to complete the subtasks required by the role, and the agent can play the role in the process of processing the target task and complete the processing of one or more subtasks in the target task. Among them, the ability of the agent to complete subtasks refers to the data processing capability of the agent.

[0124] In other words, the target task can be divided into multiple subtasks, and the conditions required to complete the subtasks can be determined respectively, and then the appropriate agent can be selected to complete the subtasks according to the conditions. In this way, after determining multiple agents, the multiple subtasks that make up the target task can be completed by multiple agents respectively, thereby completing the processing of the target task. In this way, the complex target task is divided into type tasks that can be processed by a single agent, which provides a basis for processing the target task through the cooperation of multiple agents. In addition, selecting a suitable agent for each subtask based on the ability of the agent to process type tasks can also ensure that the selected agent has the ability to process the target task, thereby ensuring the processing quality of the target task.

[0125] Similar to determining the condition information of the role, if the task configuration information of the target task also includes the agent configuration information. Accordingly, when there are multiple agents, the corresponding role can be determined by combining the agent configuration information and the condition information of the role. That is to say, on the one hand, the agent that completes these subtasks can be determined based on the conditions required to complete the subtasks, and on the other hand, the condition information corresponding to the role can be determined based on the instructions given by the user. In this way, the determined condition information has the ability to complete the subtasks required by the role and can meet the needs of the user.

[0126] In some possible implementations, an agent description database may be pre-established. Thus, when determining an agent based on condition information, an agent that meets the condition information may be selected from the agent description database. Specifically, the agent description database may store agent description information of multiple agents. The agent description information of an agent may include information such as the ability information of the agent, the input and output information of the agent, and the call information of the agent.

[0127] Among them, the ability information of the agent can be used to describe the processing ability of the agent. The processing ability of the agent can include the type of data that the agent is good at processing. When selecting an agent based on the condition information, the ability information of the agent can be referred to to select an agent that has the ability to process subtasks. In other words, in order to determine the agent corresponding to role A, the agent description database can be searched according to the condition information of role A, and an agent that has the ability to process the subtasks that role A needs to complete can be selected from the agent description database.

[0128] Optionally, the capability information of the agent can also be used to describe the strength of the processing capability of the agent. Optionally, the strength of the processing capability of the agent can be reflected in the speed at which the agent processes data, the amount and / or complexity of the data that the agent can process, and the accuracy of the data processed by the agent. Accordingly, if there are multiple agents that have the ability to process the subtasks that role A needs to complete, the agent with the strongest ability to process the subtasks that role A needs to complete can be selected based on the ability information of the agent as the agent that plays role A when processing the target task.

[0129] The input and output information of the agent can be used to describe the input and output conditions of the agent. In the embodiment of the present application, the processing of the target task can be achieved through information interaction between agents. Accordingly, in order to achieve information interaction between agents, the input and output of the agent can be adaptively configured so that the message sent by the upstream agent can be processed by the downstream agent. Optionally, the input and output information of the agent can, for example, include the specification of the prompt information of the agent. For an introduction to the input and output information of the agent, please refer to the following text and will not be repeated here.

[0130] The calling information of the agent may include relevant information for calling the agent, such as the calling interface of the agent. When the agent is used to process a target task, the agent may be called through the calling interface of the agent to complete the processing of the target task. Optionally, the agent may be deployed in the cloud service system or outside the cloud service system. If the agent and the server belong to the same cloud service system, the calling information of the agent may include the internal interface of the agent. If the agent and the server belong to different systems, the calling information of the agent may include the external interface of the agent.

[0131] In addition to the above three types of information, the agent description information of the agent can also include the evaluation index parameters of the agent. The evaluation index parameters of the agent indicate the quality of the agent's task processing ability. For example, the higher the evaluation index parameters of the agent, the stronger the agent's task processing ability. Accordingly, when selecting an agent, the selection can be made in combination with the evaluation index parameters of the agent. Specifically, if there are multiple agents that meet the condition information of the same role, then the evaluation index parameters of the agent can be referred to, and the agent with the strongest ability to process subtasks can be selected as the agent playing the role.

[0132] That is to say, assuming that the above-mentioned multiple roles include role A, when determining the agent corresponding to role A, first, one or more candidate agents can be determined based on the condition information of role A. Each candidate agent can meet the requirements of the condition information of role A. Then, based on the evaluation index parameters of the candidate agents, a candidate agent can be selected from the multiple candidate agents as the agent corresponding to role A. It can be understood that if only one candidate agent is determined based on the condition information of role A, the candidate agent can be used as the agent corresponding to role A without referring to the evaluation index parameters of role A.

[0133] In some possible implementations, the evaluation index parameters of the agent can be obtained by the user who requests to call the agent. For example, after calling multiple agents to process the target task, the user who publishes the target task can score each of the multiple agents. According to the user's score for the agent, the evaluation index parameters of the agent in the agent description database can be updated. Or, in some other possible implementations, the evaluation index parameters of the agent can be obtained during the operation of the agent. For example, in the process of processing the target task by multiple agents, the processing status of each agent processing subtasks can be monitored, so that the agent can be scored according to the processing status, and the evaluation index parameters can be updated according to the score of the agent. For an introduction to this part, please refer to the following, which will not be repeated here.

[0134] It should be noted that the two implementation methods for determining the first agent and the second agent described above are only examples. In actual application scenarios, agents may also be determined based on other implementation methods.

[0135] For example, in some possible implementations, the agent can also be determined according to the user's designation. Optionally, the user can specify that the management platform adopt a specific agent to process the target task when creating the target task. Accordingly, the management platform can determine the agent according to the target task. For example, if the task configuration information of the target task includes agent configuration information, and the agent configuration information indicates that the management platform processes the target task through a third agent, then the management platform can determine the third agent based on the agent configuration information. Accordingly, the dialogue chain determined below is specifically used to indicate the workflow of the first agent, the second agent, and the third agent to process the target task. In this way, the user can flexibly set the agent used to process the target task so that the processing of the target task meets their own needs.

[0136] S203: Generate a dialogue chain.

[0137] After the first agent and the second agent are determined, a dialogue chain can be generated based on the first agent and the second agent, so as to call the first agent and the second agent to process the target task according to the dialogue chain.

[0138] The dialogue chain can be used to indicate the workflow between agents or the workflow of a single agent.

[0139] The workflow between agents is used to describe the information interaction relationship between multiple agents during task processing. Figure 2a In the corresponding implementation, the management platform processes the target task through the first agent and the second agent, so the workflow between the agents can be used to indicate the information interaction relationship between the first agent and the second agent. That is, the dialogue chain is used to indicate the common workflow of the first agent and the second agent in processing the target task. In the process of processing the target task, the first agent and the second agent can exchange information according to the requirements of the dialogue chain.

[0140] The information interaction relationship between agents includes both the upstream and downstream relationships between agents and the rules for transmitting information between upstream and downstream agents. The rules for transmitting information between upstream and downstream agents may include the format of the message sent by the upstream agent to the downstream agent, the content of the message, and the information carried by the message. Among them, the upstream agent and the downstream agent are relative concepts. The upstream agent is used to send information to the downstream agent, and the downstream agent is used to process the information sent by the upstream agent.

[0141] The workflow of a single agent is used to describe the information interaction between the agent and itself during the task processing process, for example, it can include the number of iterations of the agent. Accordingly, in the process of processing the task, the agent can perform iterative processing according to the number of iterations. Iterative processing means that after the agent outputs the processing result, the processing result is converted into input content in some way and input into the agent again, so that the processing result can be processed again by the agent. For example, in Figure 2a In a corresponding implementation, the dialogue chain may indicate the workflow of the first agent and / or the workflow of the second agent. The workflow of the first agent indicates the number of iterations in the process of the first agent processing the first subtask. The workload of the second agent indicates the number of iterations in the process of the second agent processing the second subtask.

[0142] The above introduces the dialogue chain and workflow. The following introduces the method of determining the dialogue chain. Specifically, the management platform can determine the dialogue chain based on subtask information or based on process attribute configuration information. The following introduces each of them.

[0143] In a first possible implementation, the management platform may determine the dialogue chain based on the subtask information.

[0144] Specifically, if the dialog chain is used to indicate the workflow of the first agent and the second agent in processing the target task, then the dialog chain can be obtained based on the association relationship between the subtasks.

[0145] According to the previous introduction, the first agent is used to process the first subtask, and the second agent is used to process the second subtask. In the process of processing the target task, the association relationship between the first agent and the second agent can match the association relationship between the first subtask and the second subtask. Specifically, the management platform can determine the workflow of the first agent and the second agent in processing the target task based on the association relationship between the first subtask and the second subtask described by the subtask information. For example. Assuming that the output of the first subtask can be used as the input of the second subtask, when determining the dialogue chain, the first agent can be determined as the upstream agent of the second agent based on the association relationship between the first subtask and the second subtask. In this way, the workflow of the agent processing the target task is determined based on the association relationship between the subtasks processed by the agent, so that the process of the agent processing the target task can be matched with the process of the subtask, thereby ensuring the processing quality of the target task.

[0146] Alternatively, if the dialog chain is used to indicate the workflow of the agent, then the dialog chain can be determined based on the capabilities of the agent.

[0147] In the embodiment of the present application, the subtasks of the target task can be completed by the agent, which requires that the processing capability of the agent can meet the requirements of the subtask. In actual scenarios, the capabilities of the agent may be limited, making it difficult to meet the requirements of the subtask with only one iteration. To this end, multiple iterations can be used to meet the requirements of the subtask. For example, the processing accuracy required by the first subtask may be high, but the first agent cannot achieve the accuracy required by the first subtask. In this way, the accuracy of the processing result can be improved by calling the first agent multiple times, so that the final processing result can meet the requirements of the first subtask.

[0148] Accordingly, when determining the dialogue chain, the agent's ability information for processing tasks and subtask information can be obtained, and then the agent's workflow can be determined based on the agent's ability information for processing tasks and subtask information. Among them, the agent's workflow indicates the number of iterations when the agent processes subtasks. Specifically, when determining the workflow of the first agent, the first agent's ability information for processing the first type of task and the first subtask information can be obtained, and it can be judged whether the first agent's ability to process the first type of task can meet the requirements for processing the first subtask. If the requirements for processing the first subtask can be met, it can be determined that the number of iterations in the process of the first agent processing the first subtask is 1, and the first agent does not need to perform iterative processing. If the requirements for processing the first subtask cannot be met, the number of iterations that can meet the requirements for processing the first subtask can be determined. In this way, iterative processing based on the number of iterations can ensure that the agent processes the subtasks, meets the requirements of the target task, and ensures the processing quality of the target task.

[0149] In a second possible implementation manner, the management platform may determine the dialog chain based on the process attribute configuration information.

[0150] As described above, the workflow configuration information may also include process attribute configuration information. Accordingly, when determining the dialogue chain, it may also be determined based on the process attribute configuration information in the workflow configuration information.

[0151] Optionally, assuming that the process attribute configuration information includes dialogue turn information, the processing flow can be determined based on the dialogue turn information, agent interaction information, and multiple agents. The dialogue turns between agents in the processing flow match the dialogue turn information. For example, assuming that the dialogue turn information indicates that 10 iterations are required between the "code writer" and the "tester", in the processing flow, the agent used to complete the subtask completed by the "code writer" and the agent used to complete the subtask completed by the "tester" can have 10 rounds of dialogue.

[0152] In this way, users can flexibly set the number of iterations of one or more agents in the dialogue chain so that the processing of the target task meets their own needs.

[0153] As described above, the process attribute configuration information can also include requirement information for describing the overall processing flow. If the process attribute configuration information includes requirement information for describing the overall processing flow, the processing flow can be determined by referring to the specific rules of information interaction indicated by the process attribute configuration information.

[0154] If the process attribute configuration information includes information for describing the overall requirements of the processing flow, then the information describing the overall requirements of the processing flow can be parsed first, and then the agent interaction information can be determined. For example, assuming that the target task is a software development type task, and the process attribute configuration information indicates that the user selects the "waterfall development" method as the method to complete the target task, then since the "waterfall development" method emphasizes reducing iterations and feedback, it can be determined that the content of the interaction between agents is the content output by the agent. That is, the input of the downstream agent can be the output of the upstream agent. For another example, assuming that the target task is a software development type task, and the process attribute configuration information indicates that the user selects the "extreme programming development" method as the method to complete the target task, then since the "extreme programming development" method emphasizes feedback with the user, it can be determined that the content of the interaction between agents includes not only the content output by the agent, but also the content input by the user. Optionally, the above-mentioned user input content can include the data to be processed corresponding to the target task.

[0155] In this way, the user can set or adjust the processing flow of the target task through the process attribute configuration information, which improves the flexibility of target task processing and enhances the user experience.

[0156] S104: Configuring the first agent and the second agent so that the first agent and the second agent process the target task according to the dialogue chain.

[0157] After determining the first agent, the second agent, and the corresponding dialogue chain, multiple agents can be called according to the dialogue chain to process the target task. When calling an agent to process a subtask, the data before processing can be sent to the agent, and then the processed data returned by the agent can be received. Since the agent is determined according to the target task, the target task can be processed through the cooperation of multiple subtasks.

[0158] That is to say, multiple roles and the corresponding agents for each role can be determined according to the need to complete the target task, so that the corresponding roles can be played by the agents to complete the tasks required by the roles. After determining multiple agents, multiple agents can be called to process the target task. It can be seen that by parsing the subtask information, it can be determined which roles the agents need to play to complete the target type of tasks. Then, based on the characteristics of the role and the characteristics of the agent, an agent that can perform the role is selected to complete the subtasks required by the role, and finally multiple agents are called to complete the target task. In this way, starting from the requirements of the task, an agent that can meet the requirements of the task can be selected, so that the task can be processed by calling multiple appropriate agents. In this way, even if the task to be completed has a high complexity, it can be completed by the cooperation between multiple agents, which meets the need to complete tasks with high complexity through agents.

[0159] Specifically, after determining the dialogue chain, multiple agents can be called in sequence according to the dialogue chain to process the data to be processed of the target task. Before processing the data to be processed, initialization can be performed. Through initialization, a basic execution environment such as language, event-driven, environment understanding, workflow, and log that supports the interaction between agents can be provided, so as to call the agent to process the target task.

[0160] In some possible implementations, the above-mentioned multiple agents may include internal agents and external agents. Taking the case where the management platform runs on a cloud service system as an example, the internal agent refers to an agent running on the cloud service system, and the external agent refers to an agent provided by other systems running outside the cloud service system. Accordingly, when information is exchanged, the management platform can be used as a transfer to realize the unified call of the internal agent and the external agent.

[0161] In an embodiment of the present application, tasks can be processed by an agent integrated with an artificial intelligence model. However, due to the limitations of technological development, the output of the agent may deviate from what the user wants to see. This phenomenon can be called "AI illusion". In particular, in an embodiment of the present application, the processing of tasks is completed by multiple agents, and in the process of processing tasks, multiple agents can have multiple rounds of dialogue. The dialogue between multiple agents and the multiple rounds of dialogue between agents may further amplify the AI ​​illusion, resulting in a large deviation between the processing result of the target task and the result required by the user.

[0162] In order to avoid the above problems, in some implementations, the intermediate processing information can also be displayed to the user. The intermediate processing information is the intermediate information obtained by multiple agents in the process of processing the target task. Then, according to the operation triggered by the user, the processing of the target task can be continued or interrupted. If the operation triggered by the user is a confirmation operation for the intermediate processing information, the processing of the target task can be continued. If the operation triggered by the user is a rejection operation for the intermediate processing information, the processing of the target task can be interrupted.

[0163] For example, in Figure 1a In the implementation shown, the management platform 321 can obtain the information sent by the upstream agent to the downstream agent, and send the information as the intermediate processing information to the client 31. The client 31 can call the display module of the terminal device 20 to display the Zhang Yongjian processing information to the user 10. The client 31 can also display a confirmation control and a rejection control to the user. Among them, the confirmation control is used to trigger the confirmation operation for the intermediate processing information, and the rejection control is used to trigger the rejection operation for the intermediate processing information.

[0164] Among them, the intermediate processing information may include information sent by the agent during the interaction process, for example, it may include information sent by the upstream agent to the downstream agent. By viewing the intermediate processing information, the user can determine whether the processing of the target task meets his own needs. If the user finds that the processing of the target task does not meet his own needs, the user can process the processing of the target task. If the user finds that the processing of the target task meets his own needs, the user can trigger a confirmation operation for the intermediate processing information so that the client sends a confirmation instruction to the management puzzle. According to the confirmation instruction, the management platform can continue to process the target task. In this way, the intermediate processing information is displayed to the user, so that the user can understand the intermediate data generated in the process of processing the target task, so as to determine whether the intermediate data meets his own requirements. In this way, the user is provided with the ability to intervene in the processing of the target task, so that the processing of the target task can be supervised to avoid AI hallucinations.

[0165] Optionally, taking the example of a user publishing a target task through the client of the cloud service system and the management platform calling an intelligent agent to complete the task, the intelligent agent management method based on artificial intelligence technology provided in an embodiment of the present application is introduced.

[0166] See also Figure 2b , Figure 2b Another flow chart of an agent management method based on artificial intelligence technology provided in an embodiment of the present application. The method can be specifically executed by a management platform in a cloud service system server.

[0167] Specifically, Figure 2bThe agent management method based on artificial intelligence technology shown may specifically include:

[0168] S211: The cloud service client obtains the target task configured by the user.

[0169] exist Figure 2b In a corresponding implementation, the user may configure the target task through the cloud service client. For example, the user may select or input task configuration information of the target task on the cloud service client.

[0170] S212: The cloud service client sends the target task to the management platform in the cloud service client.

[0171] After obtaining the target task, the cloud service client can send the target task to the cloud service server through the network. Through the forwarding of one or more devices or modules in the cloud service server, the management platform can obtain the target task.

[0172] S213: The management platform determines the first agent and the second agent according to the target task and generates a dialogue chain.

[0173] After obtaining the target task, the management platform can determine the first agent and the second agent according to the target task and generate a dialogue chain.

[0174] S214: The management platform configures the first agent and the second agent so that the first agent and the second agent process the target task according to the dialogue chain.

[0175] After determining the agent and the dialog chain for processing the target task, the management platform can configure the agent so that the agent processes the target task according to the dialog chain.

[0176] For detailed introduction of S213 and S214, please refer to the above Figure 2a The introduction of the corresponding embodiments will not be repeated here.

[0177] S215: The management platform sends the intermediate processing information to the cloud service client.

[0178] In order to avoid the "AI illusion", the server can also send intermediate processing information to the client for user confirmation. Specifically, the management platform can send intermediate processing information to the cloud service client. The intermediate processing information can be, for example, the output data of the first agent, the output data of the second agent, or other intermediate data generated in the process of processing the target task.

[0179] S216: The cloud service client displays the intermediate processing information.

[0180] After obtaining the intermediate processing information, the cloud service client can display the intermediate processing information to the user so that the user can trigger a confirmation operation or a rejection control.

[0181] S217: In response to the confirmation operation triggered by the user on the intermediate processing information, the cloud service client sends a confirmation instruction to the management platform.

[0182] If the user triggers the confirmation operation, it means that the user believes that the intermediate processing information meets his or her requirements for the target task. Accordingly, according to the confirmation operation, the cloud service client can send a confirmation instruction to the management platform.

[0183] S218: The management platform continues to process the target task according to the confirmation instruction.

[0184] After obtaining the confirmation instruction, the management platform can continue to process the target task according to the confirmation instruction. And after completing the processing of the target task, the management platform can send the processing result to the scope client for the user to view.

[0185] It can be seen that by confirming the multiple agents and dialogue chains required to process the target task through the management platform, the processing of the copied target task is realized. In addition, the user confirms the intermediate processing information to avoid the "AI hallucination" and ensure that the processing of the target task is within a controllable range.

[0186] Taking the target task of "developing code according to project requirements" as an example, the following introduces the process of the intelligent agent, dialogue chain and management platform configuring the intelligent agent to handle the target task.

[0187] In the process of a natural person developing code according to project requirements, it can be divided into multiple subtasks: "understanding code development requirements", "developing code", "testing the developed code", "adjusting the code according to the test results" and "generating instruction documents based on the developed code". Different subtasks require different processing capabilities. Accordingly, by analyzing the subtask information, it can be determined that four agents (hereinafter referred to as agent A, agent B, agent C and agent D) are required to process the target task.

[0188] Among them, Agent A has the ability to process natural language, specifically it can convert unstructured natural language into structured language, and can complete the subtask "understanding code development requirements". Agent B has the ability to automatically generate code, and can complete the subtasks "developing code" and "adjusting code according to test results". Agent C has the ability to simulate the running of the code and analyze the problems of the code according to the running of the code, and can complete the subtask "testing the developed code". Agent D has the ability to understand code and process natural language, and can complete the subtask "understanding code development requirements".

[0189] After determining Agent A, Agent B, Agent C, and Agent D, a dialogue chain can be generated. Assuming that Agent A has relatively weak natural language processing capabilities, the dialogue chain can instruct Agent A to perform 5 iterations before sending the parsed content to Agent B. If the user has higher requirements for the code, the user can instruct Agent B and Agent C to perform 10 iterations in the task configuration information. In addition, in order to avoid AI hallucinations, the user can also confirm after Agent A parses.

[0190] In this way, the processing flow of the target task can be obtained as follows Figure 3 As shown. Figure 3 The application scenario shown includes the following steps:

[0191] Step ①: Agent A can obtain the project requirement information set by the user through the client and parse the project requirement information through 5 iterations;

[0192] Step ②: Agent A can send the parsed project requirement information to the client, and in response to the confirmation operation, Agent A sends the parsed project requirement information to Agent B;

[0193] Step 3: Agent B generates the first version of the code according to the project requirements and sends it to Agent C;

[0194] Step ④: Agent C tests the first version of the code and feeds back the test results to Agent B;

[0195] Step ⑤: Repeat the above steps 9 times;

[0196] Step ⑥: Agent B generates the 11th version of code based on the 10th feedback result of Agent C;

[0197] Step 7: Agent B sends version 11 code to Agent D;

[0198] Step ⑧, Agent D generates a description document based on the 11th version of the code;

[0199] Step 9: Send the 11th version of the code and documentation to the client.

[0200] It should be noted that in the process of developing the above code, the agent can also call third-party tools, such as calling a search engine to search, or calling test cases in a test case library to test the code.

[0201] In the implementation described above, the first agent and the second agent can be selected according to the target task to process the target task. In some implementations, the selected agent may not be able to complete the subtask that the agent needs to complete well, for example, the agent has low processing efficiency and poor processing accuracy. In this way, some subtasks of the target task may not be completed well.

[0202] To this end, in some possible implementations, evaluation index parameters of the agent can be set, and the evaluation index parameters of the agent can be updated according to the completion of the subtask by the agent. Specifically, when the target task is processed by multiple agents, feedback information can be obtained, and then the evaluation index parameters of the agent can be adjusted according to the feedback information. For example, if the higher the evaluation index parameter, the more suitable the agent is for processing the task, then the evaluation index parameters of the agent with high efficiency and high precision can be increased, and the evaluation index parameters of the agent with low efficiency and low precision can be reduced.

[0203] For example, there may be multiple agents that have the ability to handle the first type of task, then the management platform can select the agent (such as the first agent) with the best evaluation index parameters from the above multiple agents to handle the target task. In some possible implementations, the evaluation index parameters of the target task can be adjusted according to the completion of the target task. Specifically, during the process of processing the target task, or after processing the target task, the management platform can obtain feedback information. The feedback information can be used to indicate the completion of the target task by the first agent. Then, the evaluation index parameters of the first agent can be adjusted according to the feedback information. For example, if the first agent has a better completion of the target task, the evaluation index parameters of the first agent can be adjusted in a more optimal direction. In this way, feedback evaluation of the agent can optimize the management platform's strategy for selecting agents.

[0204] Optionally, the feedback information may be user feedback, for example, the user may access a feedback page on the client and enter the feedback information in the feedback page. Alternatively, the feedback information may also be obtained by monitoring the agent processing the target task. This is described in detail below.

[0205] In some implementations, the processing of subtasks by the agent can be monitored, and then the agent can be evaluated. The feedback information can include the evaluation of the agent. Optionally, an evaluation agent can be called to evaluate the agent. Specifically, when evaluating the agent, the agent can be evaluated based on one or more dimensions. The following introduces some implementations of evaluating the agent.

[0206] In a first implementation, the evaluation of the agent may include an evaluation of the work quality of the agent. The work quality of the agent indicates the quality of the agent's processing of the subtask. Optionally, if the work quality of the agent is evaluated by an evaluation agent, an evaluation index corresponding to the work quality may be first obtained, and then the evaluation index corresponding to the work quality may be input into the evaluation agent to obtain an evaluation of the work quality of the agent.

[0207] For example, assuming that the target task is a software development task, the evaluation indicators corresponding to the above work quality may include the code standardization of the developed software, code vulnerabilities, code memory leaks, software network vulnerabilities, software functional completeness, software execution performance, software performance comparison with similar software, software resource consumption, software reliability output, software corresponding document integrity, test cases and installation deployment packages, etc. Optionally, the above multiple evaluation indicators can be determined by the monitoring process deployed in the processing flow, or can be determined based on the test results given by the intelligent agent playing the role of "tester".

[0208] In a second implementation, the evaluation of the agent may include an evaluation of the agent's work efficiency. The agent's work efficiency indicates the efficiency of the agent in processing subtasks. Optionally, if the agent's work efficiency is evaluated by an evaluation agent, an evaluation index corresponding to the work efficiency may be first obtained, and then the evaluation index corresponding to the work efficiency may be input into the evaluation agent to obtain an evaluation of the agent's work efficiency.

[0209] For example, the evaluation index of the work efficiency may include the total time to process the target task, the total number of dialogue chain iterations and the number of tokens used in the process of processing the target task. Assuming that the target task is a software development task, the evaluation index corresponding to the work efficiency may also include the number of software generation times.

[0210] In a third implementation, the evaluation of the agent may include an evaluation of the agent's consumed resources. The agent's consumed resources represent the total resources consumed by the multiple agents called during the processing of the target task. Optionally, the agent's consumed resources may include, for example, the total number of tokens used during the processing of the target task, and / or the fees converted from the used tokens.

[0211] In a fourth implementation, the evaluation of the agent may include an evaluation of the consistency of the agent. The consistency of the agent indicates the consistency of the output results when the agent processes the same data multiple times. Optionally, if the consistency of the agent is evaluated by the evaluation agent, the agent may be used to perform multiple processing to obtain multiple processing results, and then the multiple processing results may be input into the evaluation agent to obtain an evaluation of the consistency of the agent. Alternatively, the consistency of the agent may be characterized by the similarity between the multiple processing results.

[0212] It should be noted that the above four implementation methods are only examples. In actual scenarios, multiple implementation methods can be used to evaluate the agent. Optionally, if the agent is evaluated in multiple dimensions, the evaluation indicators of multiple dimensions can be uniformly input into the evaluation agent to obtain the score of the target agent, or the evaluation indicators of multiple dimensions can be input into the evaluation agent separately, and then the evaluations given by multiple evaluation agents are summarized to obtain the score of the target agent. After obtaining the score of the target agent, the evaluation indicator parameters of the target agent can be updated according to the score of the target agent.

[0213] The present application also provides an intelligent agent management device based on artificial intelligence technology, wherein the intelligent agent management device based on artificial intelligence technology can be applied to Figure 1a The management platform 321 in the implementation mode is shown to realize Figure 2a The functions of the intelligent agent management device based on artificial intelligence technology in the implementation shown in the figure. Specifically, Figure 4 As shown, the intelligent agent management device 400 based on artificial intelligence technology includes:

[0214] An acquisition module 410 is used to acquire a target task;

[0215] An agent determination module 420 is used to determine a first agent and a second agent according to the target task, wherein the first agent and the second agent belong to the plurality of agents;

[0216] A dialogue chain generation module 430, used to generate a dialogue chain, wherein the dialogue chain is used to instruct the first agent and the second agent to process the workflow of the target task;

[0217] The agent configuration module 440 is used to configure the first agent and the second agent so that the first agent and the second agent process the target task according to the dialogue chain.

[0218] Among them, the acquisition module 410, the agent determination module 420, the dialogue chain generation module 430 and the agent configuration module 440 can all be implemented by software, or can be implemented by hardware. Exemplarily, the implementation of the agent determination module 420 is introduced below by taking the agent determination module 420 as an example. Similarly, the implementation of the acquisition module 410, the dialogue chain generation module 430 and the agent configuration module 440 can be the implementation of the agent determination module 420.

[0219] As an example of a software functional unit, the agent determination module 420 may include code running on a computing instance. Among them, the computing instance may include at least one of a physical host (computing device), a virtual machine, and a container. Further, the above-mentioned computing instance may be one or more. For example, the agent determination module 420 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed in the same region (region) or in different regions. Furthermore, the multiple hosts / virtual machines / containers used to run the code may be distributed in the same availability zone (AZ) or in different AZs, each AZ including one data center or multiple data centers with close geographical locations. Among them, usually a region may include multiple AZs.

[0220] Similarly, multiple hosts / virtual machines / containers used to run the code can be distributed in the same virtual private cloud (VPC) or in multiple VPCs. Usually, a VPC is set up in a region. For cross-region communication between two VPCs in the same region and between VPCs in different regions, a communication gateway needs to be set up in each VPC to achieve interconnection between VPCs through the communication gateway.

[0221] As an example of a hardware functional unit, the agent determination module 420 may include at least one computing device, such as a server, etc. Alternatively, the agent determination module 420 may also be a device implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0222] The multiple computing devices included in the agent determination module 420 can be distributed in the same region or in different regions. The multiple computing devices included in the agent determination module 420 can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the agent determination module 420 can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0223] It should be noted that, in other embodiments, the acquisition module 410 can be used to execute any step in the intelligent agent management method based on artificial intelligence technology, the intelligent agent determination module 420 can be used to execute any step in the intelligent agent management method based on artificial intelligence technology, the dialogue chain generation module 430 can be used to execute any step in the intelligent agent management method based on artificial intelligence technology, and the intelligent agent configuration module 440 can be used to execute any step in the intelligent agent management method based on artificial intelligence technology. The steps that the acquisition module 410, the intelligent agent determination module 420, the dialogue chain generation module 430 and the intelligent agent configuration module 440 are responsible for implementing can be specified as needed, and the acquisition module 410, the intelligent agent determination module 420, the dialogue chain generation module 430 and the intelligent agent configuration module 440 respectively implement different steps in the intelligent agent management method based on artificial intelligence technology to realize all functions of the intelligent agent management device based on artificial intelligence technology.

[0224] The present application also provides a computing device 500. Figure 5As shown, the computing device 100 includes: a bus 102, a processor 104, a memory 106, and a communication interface 108. The processor 104, the memory 106, and the communication interface 108 communicate with each other through the bus 102. The computing device 400 can be a server or a terminal device. It should be understood that the present application does not limit the number of processors and memories in the computing device 100.

[0225] The bus 102 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 5 The bus 102 is represented by only one line, but it does not mean that there is only one bus or one type of bus. The bus 102 may include a path for transmitting information between various components of the computing device 100 (eg, the memory 106, the processor 104, and the communication interface 108).

[0226] The processor 104 may include any one or more of a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0227] The memory 106 may include a volatile memory, such as a random access memory (RAM). The processor 104 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0228] The memory 106 stores executable program codes, and the processor 104 executes the executable program codes to respectively implement the functions of the aforementioned acquisition module 410, agent determination module 420, dialogue chain generation module 430, and agent configuration module 440, thereby implementing an agent management method based on artificial intelligence technology. That is, the memory 106 stores instructions for executing this storage method.

[0229] Alternatively, the memory 106 stores executable codes, and the processor 104 executes the executable codes to implement the functions of the aforementioned intelligent agent management device based on artificial intelligence technology, thereby implementing the intelligent agent management method based on artificial intelligence technology. That is, the memory 106 stores instructions for executing the intelligent agent management method based on artificial intelligence technology.

[0230] The communication interface 108 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the computing device 100 and other devices or a communication network.

[0231] The embodiment of the present application also provides a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as a central server, an edge server, or a local server in a local data center. In some embodiments, the computing device can also be a terminal device such as a desktop computer, a laptop computer, or a smart phone.

[0232] like Figure 6 As shown, the computing device cluster includes at least one computing device 100. The memory 106 in one or more computing devices 100 in the computing device cluster may store the same instructions for executing the agent management method based on artificial intelligence technology.

[0233] In some possible implementations, the memory 106 of one or more computing devices 100 in the computing device cluster may also store partial instructions for executing the agent management method based on artificial intelligence technology. In other words, the combination of one or more computing devices 100 can jointly execute instructions for executing the agent management method based on artificial intelligence technology.

[0234] It should be noted that the memory 106 in different computing devices 100 in the computing device cluster can store different instructions, which are respectively used to execute part of the functions of the intelligent agent management device based on artificial intelligence technology. That is, the instructions stored in the memory 106 in different computing devices 100 can implement the functions of one or more modules among the acquisition module 410, the intelligent agent determination module 420, the dialogue chain generation module 430 and the intelligent agent configuration module 440.

[0235] In some possible implementations, one or more computing devices in the computing device cluster may be connected via a network, which may be a wide area network or a local area network. Figure 7 A possible implementation is shown. Figure 7As shown, two computing devices 100A and 100B are connected via a network. Specifically, the network is connected via a communication interface in each computing device. In this type of possible implementation, the memory 106 in the computing device 100A stores instructions for executing the functions of the acquisition module 410, the agent determination module 420, and the dialogue chain generation module 430. At the same time, the memory 106 in the computing device 100B stores instructions for executing the functions of the agent configuration module 440.

[0236] Figure 7 The connection method between the computing device clusters shown can be based on the consideration that the intelligent agent management method based on artificial intelligence technology provided in the present application can be divided into two parts: determining the intelligent agent and the dialogue chain, and processing the target task. Therefore, it is considered that the functions of determining the intelligent agent and the dialogue chain are executed by the computing device 100A, and the function of processing the target task is executed by the computing device 100B.

[0237] It should be understood that Figure 7 The functions of the computing device 100A shown in FIG. 1 may also be completed by multiple computing devices 100. Similarly, the functions of the computing device 100B may also be completed by multiple computing devices 100.

[0238] The present application embodiment also provides another computing device cluster. The connection relationship between the computing devices in the computing device cluster can be similar to that of Figure 6 and Figure 7 The connection mode of the computing device cluster is different in that the memory 106 in one or more computing devices 100A in the computing device cluster may store the same instructions for executing the agent management method based on artificial intelligence technology.

[0239] In some possible implementations, the memory of one or more computing devices 100B in the computing device cluster may also store partial instructions for executing the agent management method based on artificial intelligence technology. In other words, the combination of one or more computing devices can jointly execute instructions for executing the agent management method based on artificial intelligence technology.

[0240] It should be noted that the memory 106 in different computing devices 100A in the computing device cluster can store different instructions for executing part of the functions of the intelligent agent management device based on artificial intelligence technology. That is, the instructions stored in the memory 106 in different computing devices 100A can implement the functions of one or more devices in the cloud service system.

[0241] The embodiment of the present application also provides a computer program product including instructions. The computer program product may be software or a program product including instructions that can be run on a computing device or stored in any available medium. When the computer program product is run on at least one computing device, the at least one computing device executes an agent management method based on artificial intelligence technology.

[0242] The embodiment of the present application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that can be stored by a computing device or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk). The computer-readable storage medium includes instructions that instruct the computing device to execute an intelligent agent management method based on artificial intelligence technology, or instruct the computing device to execute an intelligent agent management method based on artificial intelligence technology.

[0243] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. An agent management method based on artificial intelligence technology, characterized in that: The method is applied to a management platform, the management platform is used to manage multiple intelligent agents, each of the multiple intelligent agents is used to process at least one type of task based on artificial intelligence technology, and the method includes: Get the target task; According to the target task, determining a first agent and a second agent, wherein the first agent and the second agent belong to the plurality of agents; Generate a dialogue chain, wherein the dialogue chain includes a workflow of the first agent and the second agent processing the target task; The first agent and the second agent are configured so that the first agent and the second agent process the target task according to the dialogue chain.

2. The method according to claim 1, characterized in that Determining the first agent and the second agent according to the target task includes: Parsing the target task to determine subtask information of the target task, wherein the subtask information includes information of multiple subtasks required to complete the target task, and the multiple subtasks include a first subtask and a second subtask; Determine the first agent according to information of the first subtask, wherein the first subtask is a first type of task among the at least one type of tasks, and the first agent is used to process the first type of task; The second agent is determined according to information of the second subtask, wherein the second subtask is a second type of task among the at least one type of tasks, and the second agent is used to process the second type of task.

3. The method according to claim 2, characterized in that The subtask information is used to describe the association relationship between the multiple subtasks, and generating a dialogue chain includes: A dialog chain is generated according to the association relationship between the first subtask and the second subtask, wherein the dialog chain includes a workflow of the first agent and the second agent processing the target task.

4. The method according to claim 2 or 3, characterized in that: The generating of the dialogue chain comprises: The dialogue chain is generated based on the ability information of the first agent in processing the first type of task and the information of the first subtask. The dialogue chain also includes a workflow for the first agent in processing the first subtask. The workflow for the first agent in processing the first subtask indicates the number of iterations in the process of the first agent processing the first subtask.

5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Acquire workflow configuration information, where the workflow configuration information is used to indicate a target number of iterations of the second subtask.

6. The method according to claim 5, characterized in that The generating of the dialogue chain comprises: The dialogue chain is generated according to the target number of iterations, and the dialogue chain also includes a workflow for the second agent to process the second subtask, and the workflow for the second agent to process the second subtask indicates that the number of iterations in the process of the second agent processing the second subtask is equal to the target number of iterations.

7. The method according to claim 1, characterized in that The management platform is further used to manage a plurality of preset agent combinations, wherein the agents in the agent combination belong to the plurality of agents, and the determining of the first agent and the second agent according to the target task comprises: According to the target task, a target agent combination is determined from a plurality of preset agent combinations, and the target agent combination includes the first agent and the second agent.

8. The method according to any one of claims 1 to 7, characterized in that: Before generating the conversation chain, the method further includes: In response to the agent setting operation, a third agent is determined, and the dialog chain is specifically used to instruct the first agent, the second agent and the third agent to process the workflow of the target task.

9. The method according to any one of claims 1 to 8, characterized in that: After configuring the first agent and the second agent, the method further includes: Providing intermediate processing information, wherein the intermediate processing information is intermediate information obtained during the process of processing the target task; In response to a confirmation operation on the intermediate processing information, the target task continues to be processed.

10. The method according to any one of claims 1 to 9, characterized in that: During the process of processing the target task, or after processing the target task, the method further includes: Acquiring feedback information, where the feedback information is used to indicate the completion status of the target task by the first agent; The evaluation index parameters of the first agent are adjusted according to the feedback information, and the evaluation index parameters of the first agent are used to select the first agent to process the task.

11. An intelligent agent management device based on artificial intelligence technology, characterized in that: The device is applied to a management platform, the management platform is used to manage multiple agents, each of the multiple agents is used to process at least one type of task based on artificial intelligence technology, and the device includes: Acquisition module, used to acquire target tasks; An agent determination module, used to determine a first agent and a second agent according to the target task, wherein the first agent and the second agent belong to the plurality of agents; A dialog chain generation module, used to generate a dialog chain, wherein the dialog chain is used to instruct the first agent and the second agent to process the workflow of the target task; An agent configuration module is used to configure the first agent and the second agent so that the first agent and the second agent process the target task according to the dialogue chain.

12. The device according to claim 11, characterized in that The agent determination module is specifically used to parse the target task and determine the subtask information of the target task, wherein the subtask information includes information of multiple subtasks required to complete the target task, and the multiple subtasks include a first subtask and a second subtask; determine the first agent based on the information of the first subtask, wherein the first subtask is a first type of task among the at least one type of tasks, and the first agent is used to process the first type of task; determine the second agent based on the information of the second subtask, wherein the second subtask is a second type of task among the at least one type of tasks, and the second agent is used to process the second type of task.

13. The device according to claim 12, characterized in that The subtask information is used to describe the association relationship between the multiple subtasks. The dialogue chain generation module is specifically used to determine the workflow for the first agent and the second agent to process the target task according to the association relationship between the first subtask and the second subtask.

14. The device according to claim 12 or 13, characterized in that The dialog chain further includes a workflow of the first agent processing the first subtask, wherein the workflow of the first agent processing the first subtask indicates the number of iterations in the process of the first agent processing the first subtask. The dialogue chain generation module is also used to determine the number of iterations in the process of the first agent processing the first subtask based on the ability information of the first agent processing the first type of task and the information of the first subtask.

15. The device according to any one of claims 12 to 14, characterized in that The dialogue chain also includes a workflow for the second agent to process the second subtask, The acquisition module is further used to acquire workflow configuration information, where the workflow configuration information is used to indicate a target number of iterations for the second subtask.

16. The device according to claim 15, characterized in that The dialogue chain generation module is also used to determine the workflow of the second agent processing the second subtask based on the target number of iterations, and the workflow of the second agent processing the second subtask indicates that the number of iterations in the process of the second agent processing the second subtask is equal to the target number of iterations.

17. The device according to claim 11, characterized in that The management platform is also used to manage multiple preset agent combinations, where the agents in the agent combination belong to the multiple agents. The agent determination module is used to determine a target agent combination from a plurality of preset agent combinations according to the target task, and the target agent combination includes the first agent and the second agent.

18. The device according to any one of claims 11 to 17, characterized in that The agent determination module is also used to determine a third agent in response to an agent setting operation, and the dialogue chain is specifically used to instruct the first agent, the second agent and the third agent to process the workflow of the target task.

19. The device according to any one of claims 11 to 18, characterized in that The device also includes a confirmation module, The confirmation module is used to provide intermediate processing information, where the intermediate processing information is intermediate information obtained during the process of processing the target task; in response to a confirmation operation on the intermediate processing information, continue to process the target task.

20. The device according to any one of claims 11 to 19, characterized in that The device also includes a feedback module, The feedback module is used to obtain feedback information, and the feedback information is used to indicate the completion status of the target task by the first agent; the evaluation index parameters of the first agent are adjusted according to the feedback information, and the evaluation index parameters of the first agent are used to select the first agent to process the task.

21. A computing device cluster, characterized in that: The computing device cluster includes at least one computing device, each computing device including a processor and a memory: The memory is used to store instructions; The processor is configured to cause the computing device cluster to execute the method according to any one of claims 1 to 10 according to the instructions.

22. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, which, when executed on a computing device, enable the computing device to perform the method according to any one of claims 1 to 10.

23. A computer program product comprising instructions which, when executed on a computing device, cause the computing device to perform the method according to any one of claims 1 to 10.

Images