Task execution method and device for multi-agent cluster, equipment and medium

Abstract

The invention discloses a task execution method and device for a multi-agent cluster, equipment and a medium, and the method comprises the steps: receiving a task request triggered by an application system, converting the task request into task description data, and speculating an agent capability type associated with the task description data; selecting a target agent from the multi-agent cluster according to the agent portrait library and the agent capability type; and constructing a task model corresponding to the task request, and executing each task node in the task model according to the target agent. According to the technical scheme provided by the embodiment of the invention, unified control and efficient management of task scheduling and execution processes in the multi-agent cluster can be realized, and the controllability, stability and engineering feasibility of the multi-agent cluster master control system are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence system control and computer data processing technology, and in particular to a task execution method, apparatus, device and medium for a multi-agent cluster. Background Technology

[0002] With the continuous development of large language models and intelligent agent technology, more and more application systems are beginning to adopt multi-agent collaborative methods to complete complex tasks, such as text understanding, information extraction, knowledge question answering, multi-turn reasoning, and automated execution.

[0003] In existing technologies, multi-agent clusters are typically triggered directly by upper-layer application logic, with a specific agent explicitly designated to handle the corresponding task. In this type of system architecture, the task scheduling logic is tightly coupled with the implementation of the specific agent. When the agent's capabilities change or its operating state becomes abnormal, adjustments to the upper-layer application logic are often required, resulting in poor system flexibility and maintainability.

[0004] Furthermore, existing multi-agent clusters typically form a sequential execution chain by calling each other, where the preceding agent triggers the following agent to continue execution after completing its processing, thus forming a fixed calling process. If any agent malfunctions, the entire calling chain can be interrupted, making it difficult to guarantee the overall stability of the system.

[0005] In addition, some multi-agent clusters rely on manually configured rule templates or static mapping relationships to complete task distribution. Because this method lacks unified management and evaluation of the agent's running status, historical execution effect and load, it cannot dynamically select and optimize agents based on system running status information. As a result, in high-concurrency or complex task scenarios, the system's execution efficiency and reliability are difficult to meet actual production needs.

[0006] In summary, existing multi-agent clusters generally suffer from problems such as high coupling between scheduling and execution logic, lack of a unified modeling method for task execution processes, unquantifiable agent capabilities and operational states, and insufficient system scalability and stability when executing tasks. As the number of agents and task complexity continue to increase, these problems will be further amplified, severely restricting the large-scale application of multi-agent clusters in real-world production environments. Summary of the Invention

[0007] This invention provides a task execution method, apparatus, device, and medium for a multi-agent cluster, which can realize unified control and efficient management of task scheduling and execution processes in a multi-agent cluster, and improve the controllability, stability, and engineering feasibility of the multi-agent cluster control system.

[0008] According to one aspect of the present invention, a task execution method for a multi-agent cluster is provided, the method comprising:

[0009] Receive a task request triggered by the application system, convert the task request into task description data, and infer the type of intelligent agent capability associated with the task description data;

[0010] Based on the agent profile library and the agent's capability type, a target agent is selected from the multi-agent cluster; the agent profile library stores the capability tags and runtime data corresponding to each of the multiple agents.

[0011] Construct a task model corresponding to the task request, and execute each task node in the task model according to the target agent.

[0012] Optionally, the type of agent capability associated with the task description data can be inferred, including:

[0013] Perform semantic parsing on the task description data to obtain the semantic parsing result corresponding to the task description data;

[0014] Based on the semantic parsing results corresponding to the task description data, the type of agent capability required to complete the task request is inferred.

[0015] Optionally, constructing a task model corresponding to the task request includes:

[0016] Based on the agent capability type associated with the task description data, determine the set of task nodes required to complete the task request, and assign role information to each task node in the task node set.

[0017] Generate the dependencies between task nodes based on their role information.

[0018] Based on the set of task nodes and the dependencies, a directed acyclic topology graph is constructed, and based on the directed acyclic topology graph, a task model corresponding to the task request is constructed.

[0019] Optionally, based on the directed acyclic topology graph, a task model corresponding to the task request is constructed, including:

[0020] Perform a validity check on the directed acyclic topology graph;

[0021] If the directed acyclic topology graph is successfully verified, the association between the target intelligent agent and the core task node is established based on the intelligent agent profile data in the intelligent agent profile library.

[0022] The directed acyclic topology graph is serialized, and a task model corresponding to the task request is constructed based on the serialization result and the association relationship.

[0023] Optionally, executing each task node in the task model according to the target agent includes:

[0024] The directed acyclic topology graph in the task model is parsed to obtain multiple task nodes to be executed;

[0025] If the plurality of task nodes includes a preprocessing node, then the preprocessing node is executed first;

[0026] After the preprocessing node is completed, the target agent is invoked to execute the core task node among the multiple task nodes.

[0027] Optionally, after invoking the target agent to execute the core task node among the multiple task nodes, the method further includes:

[0028] After the core task node is completed, the post-processing node among the multiple task nodes is executed;

[0029] After each task node has been executed, the execution status information corresponding to each task node is updated, and the output results of each task node are summarized to obtain the final execution result corresponding to the task request.

[0030] According to another aspect of the present invention, a task execution device for a multi-agent cluster is provided, the device comprising:

[0031] The request receiving module is used to receive task requests triggered by the application system, convert the task requests into task description data, and infer the agent capability type associated with the task description data.

[0032] The agent selection module is used to select a target agent from a multi-agent cluster based on the agent profile library and the agent's capability type; the agent profile library stores the capability tags and runtime data corresponding to each agent.

[0033] The task execution module is used to construct the task model corresponding to the task request and execute each task node in the task model according to the target agent.

[0034] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0035] At least one processor;

[0036] A memory that is communicatively connected to the at least one processor;

[0037] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to execute the multi-agent cluster task execution method according to any embodiment of the present invention.

[0038] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the task execution method of the multi-agent cluster according to any embodiment of the present invention.

[0039] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program, which, when executed by a processor, implements the multi-agent cluster task execution method described in any embodiment of the present invention.

[0040] The technical solution provided by this invention receives a task request triggered by an application system, converts the task request into task description data, infers the agent capability type associated with the task description data, selects a target agent in a multi-agent cluster based on the agent profile library and agent capability type, constructs a task model corresponding to the task request, and executes each task node in the task model based on the target agent. This enables unified control and efficient management of task scheduling and execution processes in a multi-agent cluster, improving the overall controllability, stability, and engineering feasibility of the multi-agent cluster control system. It is applicable to application scenarios such as multi-agent collaboration systems, artificial intelligence workflow platforms, enterprise-level agent orchestration systems, domain large model application platforms, and automated execution pipelines.

[0041] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1a This is a flowchart of a task execution method for a multi-agent cluster according to an embodiment of the present invention;

[0044] Figure 1bThis is a schematic diagram of the structure of a multi-agent cluster control system according to an embodiment of the present invention;

[0045] Figure 2a This is a flowchart of another task execution method for a multi-agent cluster provided by an embodiment of the present invention;

[0046] Figure 2b This is a schematic diagram of the structure of a multi-agent cluster control system in a containerized deployment environment according to an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of the structure of a multi-agent cluster task execution device according to an embodiment of the present invention;

[0048] Figure 4 This is a schematic diagram of the structure of an electronic device that implements the task execution method of a multi-agent cluster according to an embodiment of the present invention. Detailed Implementation

[0049] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0050] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0051] To address the technical problems in existing technologies, such as severe coupling between scheduling and execution, lack of a unified expression model for task execution processes, difficulty in quantifying and evaluating agent capabilities and operational states, and insufficient stability and scalability of the system under large-scale concurrent scenarios, this embodiment proposes a task execution method for multi-agent clusters. Figure 1a Here is a flowchart of the method. Figure 1bThis is a schematic diagram of the structure of a multi-agent cluster control system to which the method is applicable. This embodiment is applicable to situations where a multi-agent cluster is controlled to execute application system task requests. The method can be executed by a task execution device of the multi-agent cluster, which can be implemented in hardware and / or software and configured in an electronic device, such as... Figure 1a As shown, the method includes:

[0052] Step 110: Receive a task request triggered by the application system, convert the task request into task description data, and infer the type of intelligent agent capability associated with the task description data.

[0053] In this embodiment, as Figure 1b As shown, the multi-agent cluster control system includes an application system, a unified task entry point, a scheduler, an agent profile library, a task executor, and a multi-agent cluster.

[0054] The multi-agent cluster consists of multiple agents providing specific service capabilities, with each agent implementing one or more capability types. The application system refers to the external caller of the multi-agent cluster service, which can be a business system, an artificial intelligence application platform, a workflow platform, a front-end page, or a third-party service platform, etc.

[0055] The unified task entry point serves as a unified access interface for the cluster to provide services externally, receiving task requests from application systems. In this step, the unified task entry point performs unified access processing and format standardization on the task requests, generating task description data that is common within the cluster and independent of specific agent implementations, and forwards this task description data to the scheduler. After task execution is complete, the unified task entry point receives the execution result returned by the scheduler and returns the final result to the application system. By setting a unified task entry point, consistency of external interface semantics and centralized governance of task requests can be achieved.

[0056] In one specific embodiment, the task request includes at least task text, pattern parameters, and optional session information. The task description data includes at least task text, pattern parameters, context information, and identification information for task tracking.

[0057] In this step, after receiving the task description data, the scheduler can parse the task content in the task description data and infer the type of agent capability required to perform the task based on the task text and pattern parameters, so as to determine the target capability type required to complete the task. Optionally, the capability type can be inferred based on keyword rules, template matching, lightweight classification models, or direct parameter specification, etc., and this embodiment does not limit this.

[0058] The advantage of this setup is that, upon receiving a task request, the system infers the type of agent capability required to execute the task, rather than directly binding the task scheduling logic to a specific agent. This allows the task scheduling process to shift from calling specific agents to making scheduling decisions based on capability requirements, thereby reducing the coupling between the upper-layer application system and the agent execution logic at the multi-agent cluster control system architecture level.

[0059] Step 120: Select the target agent from the multi-agent cluster based on the agent profile library and the agent's capability type; the agent profile library stores the capability tags and runtime data corresponding to each agent.

[0060] In this embodiment, after the scheduler infers the agent capability type associated with the task description data, it can send a query request to the agent profile library to obtain candidate agents with the corresponding capability type. Based on the profile data of each agent in the agent profile library, the scheduler can comprehensively evaluate the candidate agents and select the target agent with the best performance.

[0061] The query request includes at least capability type information to retrieve candidate agents with the corresponding capabilities. The agent profile database centrally stores operational data for multiple agents, including capability tags (e.g., text understanding, information extraction, knowledge answering, multi-turn reasoning), historical execution success rates, average response latency, health status, and load conditions.

[0062] After receiving the query request, the intelligent agent profile library can return candidate intelligent agent profile data that meets the capability matching conditions. The profile data includes at least the intelligent agent identifier, capability tag, service address, and profile information reflecting the running status, thereby forming a query and return closed loop between the scheduler and the intelligent agent profile library.

[0063] The scheduler can comprehensively evaluate candidate intelligent agents based on the candidate intelligent agent profile data returned by the intelligent agent profile library, filter out intelligent agents that do not meet the availability or health status requirements, and select the optimal target intelligent agent based on operational indicators such as success rate, response latency, and load.

[0064] The advantage of this setup is that by introducing an agent selection mechanism based on runtime data, a cluster of multiple agents can continuously optimize the quality of task execution based on actual operating conditions, avoiding a decline in the overall cluster execution performance due to performance fluctuations or abnormal states of a single agent.

[0065] Step 130: Construct the task model corresponding to the task request, and execute each task node in the task model according to the target agent.

[0066] In this embodiment, after determining the optimal target agent, the scheduler does not directly trigger the agent to execute the task. Instead, it constructs a task model with a directed acyclic topology graph as the core based on the task description data, the inferred capability type, and the target agent information.

[0067] The directed acyclic topology graph is used to describe the dependencies between processing steps during task execution. It breaks down the task execution process into multiple logical nodes, such as preprocessing (for input data processing), core execution (for capability invocation), and post-processing (for result normalization), and clearly defines the execution order and dependency constraints between each node. This approach transforms the task execution flow from an implicit calling relationship into an explicit, structured execution topology, providing a unified model foundation for the expression, monitoring, and control of complex task flows.

[0068] In this step, after the scheduler constructs the task model (including the directed acyclic topology graph and task parameters), it can send the task model and target agent information to the task executor. The task executor then executes each task node according to the topological relationships defined by the directed acyclic topology graph. The task executor, as an execution engine independent of the scheduler, is only responsible for the execution control of specific nodes and agent invocation, and does not participate in the task capability judgment and scheduling decision-making process.

[0069] The advantage of this setup is that by separating scheduling decisions from task execution in the system architecture, it can effectively prevent task execution anomalies from affecting the scheduler, and achieve fault isolation, retry control, and rollback processing during task execution, thereby significantly improving the stability of multi-agent clusters in complex operating environments.

[0070] In this step, specifically, after receiving the task model, the task executor can parse the task model and initialize the execution context. Then, it schedules each node to execute according to the topological relationship defined by the directed acyclic graph, sequentially completing the execution of the preprocessing node, the core execution node, and the postprocessing node, and returns the execution result to the scheduler.

[0071] The core execution nodes are completed by invoking the capabilities of the target intelligent agent. During task execution, the task executor can record the execution status and performance metrics of each task node in real time, and execute corresponding exception handling strategies when anomalies occur.

[0072] The technical solution provided by this invention receives task requests triggered by an application system, converts the task requests into task description data, infers the agent capability type associated with the task description data, selects a target agent from a multi-agent cluster based on an agent profile library and agent capability types, constructs a task model corresponding to the task request, and provides a multi-layered, separated multi-agent scheduling and execution system based on the technical means of the target agent executing each task node in the task model. Compared with existing technologies, by introducing a scheduling and execution separation mechanism at the system structure level, the scheduler no longer directly bears the agent execution load, thereby effectively reducing the risk of execution failure propagation in the system and improving the overall system stability. By adopting a task execution model based on a directed acyclic topology graph, the multi-agent task execution process has an explicit structure and observability, solving the problem that existing agent clusters are difficult to express complex processes and lack execution process control. At the same time, by introducing an agent profile library and making dynamic selection based on runtime indicators, the system can continuously optimize agent scheduling decisions during operation, significantly improving the task execution success rate and the overall system throughput. Therefore, it is suitable for complex application scenarios of large-scale collaborative operation of multiple agents and has significant engineering practical value.

[0073] Figure 2a A flowchart of another task execution method for a multi-agent cluster provided in an embodiment of the present invention is shown below. Figure 2a As shown, the method includes:

[0074] Step 210: Receive a task request triggered by the application system, convert the task request into task description data, perform semantic parsing on the task description data, and obtain the semantic parsing result corresponding to the task description data.

[0075] In this embodiment, after the scheduler receives the task description data forwarded by the unified task entry point, it can initialize the scheduling context of the current scheduling process based on the task description data, and then parse the task text and pattern parameters in the task description data to identify the semantic features and processing targets of the task, thereby obtaining the semantic parsing result corresponding to the task description data.

[0076] Step 220: Based on the semantic parsing results corresponding to the task description data, infer the type of intelligent agent capability required to complete the task request, and select the target intelligent agent in the multi-agent cluster based on the intelligent agent profile library and the intelligent agent capability type.

[0077] Step 230: Based on the agent's capability type, determine the set of task nodes required to complete the task request, and assign role information to each task node in the set of task nodes.

[0078] In this embodiment, after the scheduler infers the type of agent capability required to complete the task request, it can determine the set of task nodes required for this task execution based on predefined execution model rules or task type characteristics.

[0079] Optionally, the task node set includes at least a preprocessing node for input processing, a core task node for capability invocation, and a post-processing node for result normalization. The scheduler can assign corresponding node roles and node attribute information to each task node to clarify its functional position in the task execution chain.

[0080] Step 240: Generate the dependency relationships between each task node based on the role information of each task node, construct a directed acyclic topology graph based on the set of task nodes and the dependency relationships, and construct the task model corresponding to the task request based on the directed acyclic topology graph.

[0081] In this step, the scheduler can generate dependencies between task nodes based on node role information and task execution logic. These dependencies describe the sequential execution constraints between task nodes, clarify the preconditions for each task node's execution, and transform the originally implicit execution order into an explicit structured topological relationship.

[0082] In one embodiment of this example, constructing a task model corresponding to the task request based on the directed acyclic topology graph includes: performing a validity check on the directed acyclic topology graph; if the directed acyclic topology graph is successfully checked, establishing an association between the target intelligent agent and the core task node based on the intelligent agent profile data in the intelligent agent profile library; performing serialization processing on the directed acyclic topology graph, and constructing the task model corresponding to the task request based on the serialization processing result and the association.

[0083] In one specific embodiment, the scheduler can construct candidate directed acyclic topologies based on the task node set and dependencies. Then, it performs a validity check on these candidate topologies to confirm that there are no loops, missing dependencies, or unreachable nodes. After the validity check of the directed acyclic topology passes, the scheduler binds core task nodes to target agents based on agent profile data returned from the agent profile library. Specifically, this can be achieved by binding the target agent's identification information and capability interface information to the core task nodes, thereby establishing an association between the target agent and the core task nodes.

[0084] After binding core task nodes with target agents, the scheduler can serialize the constructed directed acyclic topology graph and build a task model based on the serialization result and the association between core task nodes and target agents. The serialization result includes at least the set of execution nodes, node dependencies, node role information, node parameters, and information about the target agents bound to the core task nodes.

[0085] Step 250: Analyze the directed acyclic topology graph in the task model to obtain multiple task nodes to be executed.

[0086] In this embodiment, after receiving the task model and target agent information, the task executor can create an independent execution context to uniformly maintain the execution status, node output results, and exception handling status of each execution node during task execution. The execution context remains valid throughout the task execution cycle to support multi-node execution control, state backtracking, and result aggregation.

[0087] Simultaneously, the task executor can also parse the received directed acyclic topology graph to obtain the set of task nodes and the dependencies between each node. Based on the node dependencies, it identifies nodes that currently have no preceding dependencies or whose preceding nodes have already been executed, and uses these as task nodes to be executed in the current stage. Through this parsing process, the task executor transforms the task execution flow from a linear call method into a node scheduling process based on topology relationships.

[0088] Step 260: If the plurality of task nodes includes a preprocessing node, then the preprocessing node is executed first; after the preprocessing node is executed, the target agent is invoked to execute the core task node among the plurality of task nodes.

[0089] In this embodiment, if a preprocessing node exists among the multiple task nodes to be executed, the task executor will prioritize scheduling the preprocessing node for execution. The preprocessing node processes the task input data to form an input format that meets the requirements of subsequent execution. The execution result of the preprocessing node is written into the execution context and serves as the source of input data for subsequent execution nodes.

[0090] After the preprocessing nodes complete their execution, the task executor schedules the core task nodes for execution. The core task nodes are task nodes in the directed acyclic topology graph directly bound to the capabilities of the target agent, and their execution is completed by invoking the target agent. Based on the target agent information bound to the core task nodes, the task executor initiates a capability invocation request to the target agent and passes the required task parameters. After the target agent completes its capability processing, it returns the processing result to the task executor, which then writes the processing result as the output of the core task node into the execution context.

[0091] Optionally, if a timeout, unavailability, or execution exception occurs during the process of a core task node invoking the target agent, the task executor will trigger an exception handling process. Specifically, the exception handling process is used to perform retry, rollback, or degradation processing on the abnormal node to avoid the failure of the entire task execution due to the failure of a single node. The usable output generated by the exception handling process is also written into the execution context to support the continued execution of subsequent nodes.

[0092] Step 270: After the core task node is completed, execute the post-processing node among the multiple task nodes.

[0093] In this embodiment, after the core task node completes its execution, the task executor schedules a post-processing node for execution. The post-processing node processes the output of the core task node according to a standardized format. The post-processing node does not alter the processing logic of the agent's capabilities; its execution result is written into the execution context as the final node-level output.

[0094] Step 280: After each task node has been executed, update the execution status information corresponding to each task node, and summarize the output results of each task node to obtain the final execution result corresponding to the task request.

[0095] In this embodiment, after each task node has been executed, the task executor updates the execution status information corresponding to each task node. The execution status information includes a node execution completion marker, execution time, and execution result status.

[0096] The technical solution provided in this invention involves receiving a task request triggered by an application system, converting the task request into task description data, performing semantic parsing on the task description data, inferring the agent's capability type based on the semantic parsing results, selecting a target agent from a multi-agent cluster based on an agent profile library and agent capability types, determining a task node set based on the agent's capability type, allocating role information for each task node in the task node set, generating dependencies based on the role information of each task node, constructing a directed acyclic topology graph based on the task node set and dependencies, and constructing a task model based on the directed acyclic topology graph. The solution then assigns roles to the tasks within the task model. This technical approach involves parsing an acyclic topology graph to obtain multiple task nodes to be executed. After the preprocessing nodes have completed their execution, the target agent is invoked to execute the core task nodes. After the core task nodes have completed their execution, the postprocessing nodes are executed. After each task node has completed its execution, the execution status information is updated. The output results of each task node are then summarized to obtain the final execution result. This approach supports the structured orchestration and execution of complex task flows. As the scale of agents continues to expand, the decoupling of scheduling and execution enables independent expansion of system resources and high concurrency processing capabilities, meeting the technical requirements of high reliability and high scalability for multi-agent clusters in production environments.

[0097] Based on the above implementation methods, in order to better explain the technical solution provided in this embodiment, Figure 2b This is a schematic diagram of the structure of a multi-agent cluster control system in a containerized deployment environment, as provided in this embodiment. Figure 2b As shown:

[0098] The application system or client acts as an external caller to the multi-agent cluster control system, deployed outside of a containerized environment or in an independent business runtime environment. The application system submits task requests to the multi-agent cluster control system through a network interface and receives the execution results returned by the control system. The application system does not directly perceive the deployment location, number of instances, or running status of the functional modules within the control system, thus maintaining a decoupled relationship with the internal structure of the control system.

[0099] The unified task entry point is deployed as an independent container instance, exposing a unified task access interface to the outside world. The unified task entry point receives task requests from application systems and forwards them to the scheduler. In a containerized deployment environment, the unified task entry point can be deployed as multiple instances to achieve load balancing, thereby improving the concurrent processing capability of the central control system for external requests. After task execution is complete, the unified task entry point receives the execution result and returns the final result to the application system, forming a closed loop for external calls.

[0100] The scheduler runs as an independent container instance, serving as the core scheduling and decision-making component of the overall control system. During operation, the scheduler receives task descriptions from a unified task entry point and performs scheduling decision operations such as capability inference, agent selection, and execution model construction. The scheduler initiates query requests to the agent profile library via a network interface and receives agent profile data returned by the library. After completing the scheduling decision, the scheduler distributes the constructed task execution model and related parameters to the task executors. The scheduler itself does not directly invoke agents, thus avoiding the impact of execution load on the scheduling decision logic.

[0101] The agent profile library is deployed as an independent container instance, used to centrally store and manage the capability information and runtime profile data of each agent in the system. During operation, the agent profile library receives query requests from the scheduler and returns candidate agent profile data that meet the capability matching criteria. The agent profile library only provides data support for scheduling decisions and does not participate in the actual task execution process. By deploying the agent profile library independently, agent capability information management and task scheduling logic are separated, facilitating independent maintenance and expansion.

[0102] The task executor is deployed as an independent container instance, serving as the system's task execution engine, and is used to actually execute tasks issued by the scheduler. During operation, the task executor receives the directed acyclic topology graph, task parameters, and target agent information from the scheduler, and schedules and executes each execution node according to the execution topology relationships defined by the directed acyclic topology graph. Within the core task nodes, the task executor calls the corresponding agent through a network interface and receives the execution results returned by the agent. After task execution is complete, the task executor sends the final execution result back to the scheduler.

[0103] A multi-agent cluster consists of multiple agents providing specific capabilities, each running as an independent container instance. Each agent exposes a capability processing interface to receive capability call requests from the task executor and return the execution result to the task executor after processing the capability. The multi-agent cluster can be horizontally scaled up or down according to actual load without affecting the operation of the scheduler and task executor. Containerized deployment allows agents to be replaced or upgraded without interrupting system operation.

[0104] The observability and monitoring unit is deployed as a standalone container or a bypass service to collect log information, operational metrics, and call chain data during system operation. The scheduler, task executor, and agent profiling library can report operational data to the observability and monitoring unit during operation. The observability and monitoring unit does not participate in the system's core business processes; it is only used for system operational status monitoring, analysis, and problem localization.

[0105] Figure 3 This is a schematic diagram of a task execution device for a multi-agent cluster provided in an embodiment of the present invention. The device is applied in electronic devices, such as... Figure 3 As shown, the device includes: a request receiving module 310, an agent selection module 320, and a task execution module 330.

[0106] The request receiving module 310 is used to receive a task request triggered by the application system, convert the task request into task description data, and infer the type of intelligent agent capability associated with the task description data.

[0107] The agent selection module 320 is used to select a target agent from a multi-agent cluster based on the agent profile library and the agent's capability type; the agent profile library stores the capability tags and runtime data corresponding to each of the multiple agents.

[0108] The task execution module 330 is used to construct a task model corresponding to the task request and execute each task node in the task model according to the target intelligent agent.

[0109] The technical solution provided by this invention receives a task request triggered by an application system, converts the task request into task description data, infers the agent capability type associated with the task description data, selects a target agent in a multi-agent cluster based on the agent profile library and agent capability type, constructs a task model corresponding to the task request, and executes each task node in the task model based on the target agent. This enables unified control and efficient management of task scheduling and execution processes in a multi-agent cluster, improving the overall controllability, stability, and engineering feasibility of the system. It is applicable to application scenarios such as multi-agent collaboration systems, artificial intelligence workflow platforms, enterprise-level agent orchestration systems, domain large model application platforms, and automated execution pipelines.

[0110] Based on the above embodiments, the request receiving module 310 includes:

[0111] The capability type inference unit is used to perform semantic parsing on the task description data to obtain the semantic parsing result corresponding to the task description data; and to infer the agent capability type required to complete the task request based on the semantic parsing result corresponding to the task description data.

[0112] Task execution module 330 includes:

[0113] The node set determination unit is used to determine the set of task nodes required to complete the task request based on the agent capability type associated with the task description data, and to allocate the role information of each task node in the task node set.

[0114] The dependency generation unit is used to generate the dependency relationships between task nodes based on the role information of each task node.

[0115] The topology graph construction unit is used to construct a directed acyclic topology graph based on the task node set and the dependency relationship, and to construct the task model corresponding to the task request based on the directed acyclic topology graph.

[0116] A topology graph verification unit is used to verify the legality of the directed acyclic topology graph.

[0117] The agent binding unit is used to establish an association between the target agent and the core task node based on the agent profile data in the agent profile library if the directed acyclic topology graph verification is successful.

[0118] The topology graph processing unit is used to perform serialization processing on the directed acyclic topology graph and construct the task model corresponding to the task request based on the serialization processing result and the association relationship.

[0119] The topology graph parsing unit is used to parse the directed acyclic topology graph in the task model to obtain multiple task nodes to be executed.

[0120] A preprocessing node execution unit is configured to execute the preprocessing node first if the plurality of task nodes include a preprocessing node.

[0121] The core node execution unit is used to call the target agent to execute the core task node among the multiple task nodes after the preprocessing node has been executed;

[0122] The post-processing node execution unit is used to execute the post-processing nodes among the plurality of task nodes after the core task node has been executed;

[0123] The status update unit is used to update the execution status information of each task node after each task node has been executed, and to summarize the output results of each task node to obtain the final execution result corresponding to the task request.

[0124] The above-described apparatus can execute the methods provided in all the foregoing embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the above methods. Technical details not described in detail in the embodiments of the present invention can be found in the methods provided in all the foregoing embodiments of the present invention.

[0125] Figure 4A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0126] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from the storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0127] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0128] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processing (DSP) processors, and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as task execution methods for multi-agent clusters.

[0129] In some embodiments, the task execution method for a multi-agent cluster can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the task execution method for a multi-agent cluster described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to execute the task execution method for a multi-agent cluster by any other suitable means (e.g., by means of firmware).

[0130] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chips (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0131] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0132] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0133] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a cathode ray tube (CRT) or a liquid crystal display (LCD)) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0134] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0135] A computing system can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system. It addresses the shortcomings of traditional physical hosts and Virtual Private Servers (VPS) in terms of management difficulty and weak business scalability.

[0136] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and no limitation is imposed herein.

[0137] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A task execution method for a multi-agent cluster, characterized in that, The method includes: Receive a task request triggered by the application system, convert the task request into task description data, and infer the type of intelligent agent capability associated with the task description data; Based on the agent profile library and the agent's capability type, a target agent is selected from the multi-agent cluster; the agent profile library stores the capability tags and runtime data corresponding to each of the multiple agents. Construct a task model corresponding to the task request, and execute each task node in the task model according to the target agent.

2. The method according to claim 1, characterized in that, The types of agent capabilities associated with the task description data can be inferred, including: Perform semantic parsing on the task description data to obtain the semantic parsing result corresponding to the task description data; Based on the semantic parsing results corresponding to the task description data, the type of agent capability required to complete the task request is inferred.

3. The method according to claim 1, characterized in that, Constructing the task model corresponding to the task request includes: Based on the agent capability type associated with the task description data, determine the set of task nodes required to complete the task request, and assign role information to each task node in the task node set. Generate the dependencies between task nodes based on their role information. Based on the set of task nodes and the dependencies, a directed acyclic topology graph is constructed, and based on the directed acyclic topology graph, a task model corresponding to the task request is constructed.

4. The method according to claim 3, characterized in that, Based on the directed acyclic topology graph, a task model corresponding to the task request is constructed, including: Perform a validity check on the directed acyclic topology graph; If the directed acyclic topology graph is successfully verified, the association between the target intelligent agent and the core task node is established based on the intelligent agent profile data in the intelligent agent profile library. The directed acyclic topology graph is serialized, and a task model corresponding to the task request is constructed based on the serialization result and the association relationship.

5. The method according to claim 1, characterized in that, Executing each task node in the task model according to the target agent includes: The directed acyclic topology graph in the task model is parsed to obtain multiple task nodes to be executed; If the plurality of task nodes includes a preprocessing node, then the preprocessing node is executed first; After the preprocessing node is completed, the target agent is invoked to execute the core task node among the multiple task nodes.

6. The method according to claim 5, characterized in that, After invoking the target agent to execute the core task node among the multiple task nodes, the process also includes: After the core task node is completed, the post-processing node among the multiple task nodes is executed; After each task node has been executed, the execution status information corresponding to each task node is updated, and the output results of each task node are summarized to obtain the final execution result corresponding to the task request.

7. A task execution device for a multi-agent cluster, characterized in that, The device includes: The request receiving module is used to receive task requests triggered by the application system, convert the task requests into task description data, and infer the agent capability type associated with the task description data. The agent selection module is used to select a target agent from a multi-agent cluster based on the agent profile library and the agent's capability type; the agent profile library stores the capability tags and runtime data corresponding to each agent. The task execution module is used to construct the task model corresponding to the task request and execute each task node in the task model according to the target agent.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the task execution method for a multi-agent cluster as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, which, when executed by a processor, implement the task execution method for a multi-agent cluster as described in any one of claims 1-6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the task execution method for a multi-agent cluster as described in any one of claims 1-6.

Images