Multi-agent interactive arrangement method and system and terminal equipment

Abstract

The invention is suitable for the technical field of artificial intelligence, and provides a multi-agent interactive arrangement method and system and terminal equipment, and the method comprises the steps: carrying out the disassembly of a task scene, and obtaining a plurality of stage nodes; based on the task and the execution agent corresponding to each stage node, determining the visibility level of information generated by each stage node; based on the capability constraint information corresponding to the agent sets and the visibility level of the task and information of each stage node, generating an interaction arrangement mode of the agent sets, so that the agent sets execute interaction operation based on the interaction arrangement mode; the set of agents includes execution agents. According to the method, the visibility level of the information generated by the nodes in each stage is determined, so that the information propagation range can be accurately controlled according to actual requirements, sensitive information is prevented from being acquired by unrelated agents, the information leakage risk is reduced, and the data security in the task execution process is guaranteed.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of artificial intelligence, and particularly relates to a multi-agent interaction orchestration method and system and a terminal device. BACKGROUND

[0002] In actual applications, in the process of evolution of artificial intelligence from single-agent independent decision-making to multi-agent collaborative cooperation, multi-agent interaction orchestration becomes a core technical support. Multi-agent interaction orchestration defines rules, allocates roles, and schedules resources, so that multiple agents with different capabilities (such as dialogue robots, data analysis agents, tool calling agents, etc.) form an orderly cooperation to complete complex tasks (such as enterprise-level customer service, complex project management, intelligent supply chain optimization, etc.) that a single agent cannot handle.

[0003] With the rapid growth of demand for multi-agent in collaborative, game and confrontation tasks, the existing technology still stays at the level of dialogue between multiple agents, which leads to the problem that the existing technology is not comprehensive enough and is difficult to meet actual needs. SUMMARY

[0004] The embodiments of the application provide a multi-agent interaction orchestration method and system and a terminal device to solve the problem that the existing technology is not comprehensive enough and is difficult to meet actual needs.

[0005] In a first aspect, the embodiments of the application provide a multi-agent interaction orchestration method, including: dissolving a task scene to obtain multiple stage nodes; determining a visibility level of information generated by each stage node based on a subtask and an execution agent corresponding to each stage node; generating an interaction orchestration mode of an agent set based on capability constraint information corresponding to each agent in the agent set, and the subtasks and the visibility levels of the information of each stage node, so that the agent set performs an interaction operation based on the interaction orchestration mode; the agent set includes multiple agents, and the multiple agents include the execution agent.

[0006] Optionally, before the task scene is dissolved to obtain the multiple stage nodes, the method further includes: when a task request is detected, determining the task scene; performing an image binding operation on each agent based on the task scene to obtain the agent set.

[0007] Optionally, after generating the interaction arrangement mode of the agent set based on the respective capability constraint information of the agent set, and the visibility level of the sub-tasks and information of each stage node, the method further includes: Upon detecting that any one stage node is activated, sending the sub-task corresponding to the stage node to the execution agent corresponding to the stage node; performing capability verification on the first information generated by the execution agent based on the capability constraint information of the execution agent, to obtain a first verification result; the first information is generated by the execution agent based on the sub-task; If the first verification result is a verification pass, sending the first information to the target agent based on the visibility level; the target agent refers to any agent in the agent set that has the right to view the information.

[0008] Optionally, after performing capability verification on the first information generated by the execution agent based on the capability constraint information of the execution agent, to obtain a first verification result, the method further includes: If the first verification result is a verification fail, performing a set operation on the execution agent; After performing the set operation, performing capability verification on the second information generated by the execution agent, to obtain a second verification result; If the second verification result is a verification fail, determining target information of the execution agent based on the first information; the target information is different from the first information; sending the target information to the target agent based on the visibility level.

[0009] Optionally, the method further includes: determining the type of the first information; storing the first information based on the type.

[0010] Optionally, the method further includes: When each agent executes a task scenario, recording events based on three channels of agent interaction channel, environmental state data change channel, and system log channel.

[0011] Optionally, the method further includes: Upon detecting a replay request, determining a scenario to be replayed; reconstructing the three-channel events of the agent interaction channel, the environmental state data change channel, and the system log channel corresponding to the scenario to be replayed based on a sorting rule.

[0012] In a second aspect, an embodiment of the present application provides a multi-agent interaction arrangement device, including: a disassembling unit configured to disassemble a task scenario to obtain a plurality of stage nodes; The first determining unit is configured to determine the visibility level of information generated by each stage node based on the subtask corresponding to each stage node and the execution agent; The generating unit is configured to generate an interaction arrangement manner of the agent set based on the capability constraint information corresponding to each agent in the agent set and the visibility level of the subtask and information of each stage node, so that the agent set performs interaction operation based on the interaction arrangement manner; the agent set includes the execution agent.

[0013] In a third aspect, an embodiment of the present application provides a multi-agent interaction arrangement system, comprising: At least one session scheduler connected with each agent, configured to perform the multi-agent interaction arrangement method according to any one of the first aspect; An event trace recorder connected with the at least one session scheduler and each agent respectively, configured to record events of a task scene execution process corresponding to each session scheduler based on three channels of an agent interaction channel, an environment state data change channel and a system log channel when each agent performs a task scene corresponding to the agent. A replay engine connected with the event trace recorder, configured to reconstruct three-channel events of an agent interaction channel, an environment state data change channel and a system log channel corresponding to a scene to be replayed based on a sorting rule when a replay request is detected.

[0014] In a fourth aspect, an embodiment of the present application provides a terminal device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the multi-agent interaction arrangement method according to any one of the first aspect when executing the computer program.

[0015] In a fifth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executable on a processor to implement the multi-agent interaction arrangement method according to any one of the first aspect.

[0016] In a sixth aspect, an embodiment of the present application provides a computer program product, which, when running on a terminal device, enables the terminal device to execute the multi-agent interaction arrangement method according to any one of the first aspect.

[0017] Compared with the prior art, the embodiment of the present application has the following beneficial effects: The multi-agent interaction arrangement method provided by the embodiment of the application can clearly define the originally ambiguous task boundary by decomposing the task scene into multiple stage nodes, each of which focuses on a single or a small number of targets; then, the visibility level of information generated by each stage node is determined based on the task and the execution agent corresponding to each stage node, so that the propagation range of the information can be accurately controlled according to actual needs, the sensitive information is prevented from being obtained by irrelevant agents, the information leakage risk is reduced, and the data security in the task execution process is ensured; finally, the interaction arrangement mode of the agent set is generated based on the capability constraint information corresponding to each agent in the agent set and the visibility level of the task and the information of each stage node, so that the agents in the agent set can perform interaction operations based on the interaction arrangement mode, and each agent can participate in the task at a suitable time, thereby realizing reasonable allocation of agent resources, reducing collaboration conflicts, and improving the smoothness and effectiveness of overall collaboration. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0019] Figure 1 is a structural schematic diagram of a multi-agent interaction arrangement system provided by an embodiment of the application; Figure 2 is a multi-session concurrency schematic diagram provided by an embodiment of the application; Figure 3 is an event trace and event replay schematic diagram provided by an embodiment of the application; Figure 4 is a specific structural schematic diagram of a session dispatcher provided by an embodiment of the application; Figure 5 is a broadcast schematic diagram of a visibility broadcaster provided by an embodiment of the application; Figure 6 is a structural schematic diagram of a multi-dimensional memory manager provided by an embodiment of the application; Figure 7 is a workflow schematic diagram of a capability constraint executor provided by an embodiment of the application; Figure 8 is an implementation flowchart of a multi-agent interaction arrangement method provided by an embodiment of the application; Figure 9 is an implementation flowchart of a multi-agent interaction arrangement method provided by another embodiment of the application; Figure 10is an implementation flowchart of a multi-agent interaction arrangement method provided by another embodiment of the present application; Figure 11 is an implementation flowchart of a multi-agent interaction arrangement method provided by another embodiment of the present application; Figure 12 is a structural schematic diagram of a multi-agent interaction arrangement apparatus provided by an embodiment of the present application; Figure 13 is a structural schematic diagram of a terminal device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0020] In the following description, specific details are set forth, such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. However, persons skilled in the art will understand that the present application can be practiced without these specific details. In other instances, well-known structures, devices, circuits, and methods have not been described in detail in order to avoid obscuring the description of the present application.

[0021] It should be understood that the term "comprising" as used in the specification and in the claims indicates the presence of the recited features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0022] It should also be understood that the term "and / or" as used in the specification and in the claims indicates any combination of one or more of the associated listed items and all possible combinations of the items.

[0023] As used in the specification and in the claims, the term "if" can be interpreted as meaning "when" or "upon" or "in response to a determination" or "in response to a detection" depending on the context. Similarly, the phrase "if determined" or "if detected [the described condition or event]" can be interpreted as meaning "upon a determination" or "in response to a determination" or "upon a detection [of the described condition or event]" or "in response to a detection [of the described condition or event]" depending on the context.

[0024] In addition, in the description of the specification and the appended claims, the terms "first", "second", "third", etc. are only used for differentiation in description, and cannot be understood as indicating or implying relative importance.

[0025] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0026] In practical applications, as artificial intelligence evolves from independent decision-making by a single agent to collaborative multi-agent cooperation, multi-agent interaction orchestration has become a core technological support. Multi-agent interaction orchestration, through defining rules, assigning roles, and scheduling resources, enables multiple agents with different capabilities (such as chatbots, data analysis agents, and tool-calling agents) to collaborate in an orderly manner, jointly accomplishing complex tasks that a single agent cannot handle (such as enterprise customer service, complex project management, and intelligent supply chain optimization).

[0027] With the rapid growth in demand for multi-agent systems in collaborative, game-like, and adversarial tasks, existing technologies are still limited to enabling dialogue between multiple agents, leaving gaps in engineering controllability and compliance auditing: for example, information domain segmentation / visibility often relies on cue engineering "soft conventions" and lacks verifiable semantics, which may lead to the leakage of private clues.

[0028] Therefore, this application provides a multi-agent interaction orchestration system and a corresponding multi-agent interaction orchestration method, which can accurately control the scope of information dissemination according to actual needs, prevent sensitive information from being obtained by irrelevant agents, reduce the risk of information leakage, and ensure data security during task execution.

[0029] It should be noted that the following multi-agent interaction orchestration method is applied to multi-agent interaction orchestration systems.

[0030] Please see Figure 1 , Figure 1 This is a schematic diagram of the structure of a multi-agent interaction orchestration system provided in an embodiment of this application. Figure 1 As shown, the multi-agent interaction orchestration system includes at least one session scheduler 10 (three are shown in the figure), an event tracing recorder 20, and a replay engine 30.

[0031] In this embodiment, at least one session scheduler 10 is connected to each agent (not shown in the figure) to implement a multi-agent interaction orchestration method.

[0032] It should be noted that when the multi-agent interaction orchestration system detects any task, it can generate a corresponding session scheduler 10 based on that task. In other words, one task can correspond to one session scheduler 10.

[0033] Therefore, a multi-agent interaction orchestration system can use sessions as the smallest operational unit, providing lifecycle operations such as creation, execution, termination, and cleanup. Each session scheduler shares a persistent layer to store only metadata and event streams, thus achieving isolation and horizontal scaling of individual tasks. Based on this, the multi-agent interaction orchestration system can allow multiple independent task scenarios to run in parallel without interference.

[0034] For example, please refer to Figure 2 , Figure 2 This is a schematic diagram of multi-session concurrency provided in an embodiment of this application. Figure 2 Describes multi-session concurrency and lifecycle: session creation → running → termination → cleanup → deletion, emphasizing instance isolation (i.e., each session is independent of the others) and horizontal scaling (i.e., multiple sessions can be created simultaneously). Figure 2 Only two were created.

[0035] It should be noted that the specific implementation process of the above-mentioned multi-agent interaction orchestration method can be found in [reference needed]. Figures 8-11 The corresponding implementation examples will not be described in detail here.

[0036] In this embodiment of the application, the event tracing recorder 20 is connected to at least one session scheduler 10 and each intelligent agent, and is used to record events on the execution process of the task scenario corresponding to each session scheduler based on three channels: intelligent agent interaction channel, environmental state data change channel and system log channel, when each intelligent agent executes its corresponding task scenario.

[0037] The replay engine 30, connected to the event tracing recorder 20, is used to reconstruct the three-channel events—the agent interaction channel, the environmental state data change channel, and the system log channel—corresponding to the scene to be replayed, based on a sorting rule when a replay request is detected. The sorting rule can be based on timestamps from earliest to latest.

[0038] For example, please refer to Figure 3 , Figure 3 This is a schematic diagram of event tracing and event replay provided in an embodiment of this application. Figure 3The process of time axis overlay and state reduction of three channels of events—the agent interaction channel (SPEAK), the environment state data change channel (STATE_CHANGE), and the system log channel (SYSTEM)—is described.

[0039] In one embodiment of this application, in order to improve the rational allocation of agent resources in multi-agent interaction orchestration, reduce cooperation conflicts, and improve the smoothness and effectiveness of overall cooperation, please refer to... Figure 4 , Figure 4 This is a schematic diagram of the specific structure of a session scheduler provided in an embodiment of this application. Each session scheduler 10 may specifically include: a profile binder 11, a visibility broadcaster 12, a multidimensional memory manager 13, and a capability constraint enforcer 14.

[0040] It should be noted that, Figure 4 This describes the data flow between the image binder 11, the visibility broadcaster 12, the multidimensional memory manager 13, the capability constraint enforcer 14, the event sourcing recorder 20, and the replay engine 30.

[0041] In this embodiment, the portrait binding device 11 is used to perform portrait binding operations on each intelligent agent based on the task scenario to obtain a set of intelligent agents.

[0042] The visibility broadcaster 12 is connected to the set of agents and is used to determine the target agent corresponding to each agent based on the visibility level of the information generated by each agent, and to send the information generated by each agent to the corresponding target agent based on the visibility level of the information generated by each agent.

[0043] It should be noted that the executing agent refers to the agent used to execute the sub-tasks corresponding to each stage node. Each stage node is obtained by decomposing the task scenario. The target agent refers to any agent among the agents that has the right to view information.

[0044] For example, please refer to Figure 5 , Figure 5 This is a broadcast schematic diagram of a visibility broadcaster provided in an embodiment of this application. Figure 5 This describes how a visibility broadcaster distributes information to authorized agents based on the message visibility level (SAME_TEAM), thereby achieving information domain-based distribution.

[0045] It should be noted that visibility levels can include five categories: visible only to oneself (PRIVATE), visible to the same character (SAME_ROLE), visible to the same group (SAME_TEAM), publicly visible (PUBLIC), and visible to a custom range (CUSTOM).

[0046] In this embodiment, the multidimensional memory manager 13 is connected to each intelligent agent and is used to store the information generated by each intelligent agent.

[0047] For example, please refer to Figure 6 , Figure 6 This is a schematic diagram of the structure of a multidimensional memory manager provided in an embodiment of this application. Figure 6 This describes the relationship between agent memory and group memory in the scope dimension, as well as the time dimension migration from short-term memory to long-term memory. The up and down directions represent the migration and decay from short-term to long-term memory.

[0048] Combination Figure 6 The multidimensional memory manager 13 can include two layers: a single agent memory (i.e., agent-private memory) and multiple group memories (i.e., group-shared memory). Here, a group refers to a combination consisting of at least two agents.

[0049] It should be noted that, in terms of scope, both agents and groups possess their own memory systems; agent memory is private, while group memory is shared. In terms of time, both agents and groups distinguish between short-term and long-term memory, thereby improving information retention and retrieval capabilities in complex scenarios.

[0050] In some possible embodiments, the conversion relationship between short-term memory (STM) and long-term memory (LTM) is defined as follows, for any memory unit m: ; Where m represents a single memory item / memory unit, and t represents the current time. This represents the memory score, used to determine transfer and decay. This represents the normalized frequency of reads / writes of memory item m within the most recent window W (e.g., the past 24 hours or the most recent K interactions). This indicates the time interval since the last access or write (the unit is the same as λ, such as seconds / minute). The term represents time-sensitive terms, which decay exponentially over time. λ>0, and α,β≥0 represent weights.

[0051] It should be noted that if s(m,t)≥ (e.g., 0.7) will trigger an STM→LTM boost; If s(m,t) < (e.g., 0.3) then decay will be performed; Pick > Create hysteresis to avoid frequent jitter.

[0052] In this embodiment, the capability constraint actuator 14 is used to perform capability verification on the information generated by each intelligent agent and obtain the corresponding verification result.

[0053] For example, please refer to Figure 7 , Figure 7 This is a schematic diagram of the workflow of a capability constraint actuator provided in an embodiment of this application. Figure 7 The information used to describe the intelligent agent follows a structured protocol and may trigger capability verification. If the intelligent agent has the ability to modify environmental state data, it is considered a compliant behavior and can modify the environmental state data. Otherwise, it will enter the correction chain, resulting in rollback or resampling.

[0054] Based on this, the aforementioned capability verification process can be specifically as follows: Agent outputs information → JSON Schema verification (i.e., protocol verification) → Capability verification → Compliant write / otherwise rollback or resampling → If still non-compliant, then write the first minimum compliant subset of JSON as a fallback. Specifically, writing the first minimum compliant subset of JSON as a fallback refers to a predictable default data structure that conforms to the basic JSON Schema specification, has the fewest fields, and is automatically generated by the capability constraint executor 14 after all verifications (protocol, capability, etc.) of the agent's output information have failed, to ensure the stable operation of subsequent processes.

[0055] Please see Figure 8 , Figure 8 This is a flowchart illustrating the implementation of a multi-agent interaction orchestration method according to an embodiment of this application. In this embodiment, the executing entity of the multi-agent interaction orchestration method is a terminal device. The terminal device includes, but is not limited to, devices such as laptops, desktop computers, and computers.

[0056] It should be noted that terminal devices include, for example, Figure 1 and Figure 4 The corresponding multi-agent interaction system. like Figure 8 As shown, a multi-agent interaction orchestration method provided in one embodiment of this application may include S101~S103, which are described in detail below: In S101, the task scenario is broken down into multiple stage nodes.

[0057] In this embodiment of the application, when the terminal device detects a task request, it can determine the task scenario based on the task request. The task request carries a specific task description that requires multi-agent collaboration to complete.

[0058] Therefore, the terminal device can determine the task scenario based on the specific task description.

[0059] In this embodiment, after determining the task scenario, the terminal device can clarify the overall goal of the task and input the overall goal into a trained task decomposition model for processing, thereby decomposing the task scenario into multiple stage nodes. The task decomposition model can be trained from a pre-built neural network model.

[0060] It should be noted that the relationships between the various stage nodes can be sequential, conditional branching, and convergence. Each stage node defines its sub-goals and boundaries, entry and exit conditions, and environmental state data to be exposed to the agent. The sub-goals and boundaries, entry and exit conditions, and environmental state data to be exposed to the agent for each stage node can all be set according to actual needs; no restrictions are imposed here.

[0061] The activation and transition of each stage node are based solely on the relationships between the stage nodes and the stage's own entry / exit conditions.

[0062] For example, taking a simplified Werewolf game scenario, the scenario arrangement, i.e., the decomposed stage nodes, can include: the nighttime werewolf voting stage, the seer stage, the daytime speaking stage, and the daytime voting stage. Among them, the daytime speaking stage and the daytime voting stage are sequential; if the seer survives, the seer stage will begin after the werewolf voting stage; otherwise, the daytime speaking stage will begin directly. This is a branching relationship.

[0063] In one embodiment of this application, to ensure the consistency and substitutability of model selection, role constraints, and parameter control within a session, and to enhance the collaborative stability of multi-agent systems, the terminal device may specifically execute the following before decomposing the task scenario: Figure 9 Steps S201 to S202 shown are described in detail below: In S201, when a task request is detected, the task scenario is determined.

[0064] In S202, a profile binding operation is performed on each intelligent agent based on the task scenario to obtain a set of intelligent agents.

[0065] In this embodiment, combined with Figure 4 After determining the task scenario, the terminal device can create multiple intelligent agents through the image binding device, and any intelligent agent can satisfy the following formula: ; in, Create a profile of the i-th intelligent agent. This is the inference backend identifier for the i-th agent. For the semantic description of the role / target / background of the i-th agent, This is the set of decoding and dialogue behavior parameters for the i-th agent, such as temperature, top-p, maximum output, etc.

[0066] It should be noted that the semantic description of the agent's role / goal / background mentioned above is a natural language or structured semantic description of the agent's position, mission, and context information within the agent set.

[0067] Specifically, the semantic description of a role is used to define the functional positioning and responsibility boundaries of an agent in the system, and to answer what the agent does.

[0068] The semantic description of the objective is used to clarify the agent's mission and expected output, and to answer what result the agent wants to achieve.

[0069] The semantic description of the background is used to explain the agent's operating environment, dependencies, and historical constraints, and to answer the question of under what conditions the agent works.

[0070] Therefore, the semantic description of the agent's role / goal / background can clearly define the agent's capabilities and environmental state data.

[0071] It should be noted that the set of decoding and dialogue behavior parameters of an agent is used to describe the combination of parameters of the core features of the agent in the process of information decoding (analysis and processing of received information) and dialogue interaction (communication behavior with other agents or users). It is a key attribute for characterizing the interaction mode of an agent. Its essence is a quantitative and structured description of how the agent understands and expresses information.

[0072] The set of decoding and dialogue behavior parameters of the agent can include the technical characteristics of the agent's information processing (such as decoding rules and format preferences) or the style characteristics of its interaction behavior (such as dialogue rhythm and sentence selection), in order to ensure that communication between multiple agents or between an agent and a user is understandable, efficient and unambiguous.

[0073] By combining the above formulas, the terminal device can obtain a set of intelligent agents.

[0074] It should be noted that the profile binder can inject task scenario context at runtime to ensure intra-session consistency and replaceability of model selection, role constraints, and parameter control. Simultaneously, during the same session, the profiles used by each agent... The image binder remains unchanged to use the same version; outside of a session, the image binder can simply be replaced. Without changing other configurations.

[0075] In S102, the visibility level of the information generated by each stage node is determined based on the sub-tasks and executing agents corresponding to each stage node.

[0076] It should be noted that each stage node carries environmental state data that needs to be exposed to the agent. This environmental state data, which describes the environmental information that the stage node actively provides to the agent for executing the corresponding subtask, can include the output of the previous stage node (e.g., the intent label output from the user intent recognition stage), the basic environmental parameters required for the subtask of this stage node (e.g., current time, user ID, business context identifier), constraint data (e.g., time threshold, accuracy requirements, resource limits for subtask execution), and external environment snapshots (e.g., current system load, third-party interface status).

[0077] In some possible embodiments, the terminal device can generate corresponding subtasks according to the definition of each stage node, and determine the execution agent of each stage node according to the image binding operation performed by the image binder.

[0078] In this embodiment of the application, the terminal device can determine the visibility level of the information generated by each stage node based on the sub-tasks of each stage node and the profile of the executing intelligent agent.

[0079] Specifically, for any given stage node, the terminal device can determine the information elements required to complete the subtask based on the goal of the subtask at that stage node; then, the terminal device can determine the capabilities of the executing agent based on the profile of the executing agent at that stage node; and then, the terminal device can determine the visibility level of the information generated by that stage node based on the information elements and the capabilities of the executing agent.

[0080] In this embodiment of the application, the executing agent can generate reasonable information for each stage node based on the environmental state data that needs to be exposed to the agent and its own local view carried by each stage node.

[0081] The local view includes the set of information accessible to the executing agent at time t, including private memory, shared memory of the group to which the executing agent belongs, and visible environmental state data.

[0082] It should be noted that the environmental state data visible to the executing agent is used to describe the environmental information that the executing agent actually perceives and processes when executing the corresponding sub-task. This can include: exposure data received from the corresponding stage node, the internal state of the executing agent itself (such as current computing load, cached historical data, execution progress), environmental changes actively detected by the executing agent (such as real-time data obtained through sensors / interfaces), and interaction information from other agents (such as collaboration requests, status notifications), etc.

[0083] In S103, based on the capability constraint information corresponding to each intelligent agent set, as well as the visibility level of subtasks and information of each stage node, an interaction orchestration method for the intelligent agent set is generated so that the intelligent agent set can perform interactive operations based on the interaction orchestration method; the intelligent agent set includes the executing intelligent agent.

[0084] It should be noted that the capability constraint information corresponding to each agent in the set of agents is used to describe the capabilities of each agent in the set. The set of agents includes the executing agents.

[0085] In this embodiment of the application, the terminal device can generate an interaction orchestration method for the intelligent agent set based on the capability constraint information corresponding to each intelligent agent set, as well as the visibility level of the sub-tasks and information of each stage node, so that the intelligent agent set can perform interactive operations based on the interaction orchestration method.

[0086] Specifically, the interaction operation performed by the intelligent agent set based on the above-mentioned interaction orchestration method can be as follows: When a certain stage node is activated, the terminal device can generate a corresponding subtask according to the definition of that stage node, and dispatch the subtask to the executing intelligent agent according to the binding rules within that stage node. The subtask carries the environmental state data exposed by that stage node, used to constrain and guide the intelligent agent to generate information. Subsequently, if the information contains an intention to modify the environmental state data, it is submitted as a state change suggestion. Whether it takes effect is then determined, and the visibility level of the information is determined (e.g., step S102), and the capability is verified (e.g., ...). Figure 7 The diagram shows the workflow of the capability constraint actuator and memory writing (such as...). Figure 6 (e.g., the multidimensional memory manager shown). When the exit condition of a stage node is met, the terminal device can proceed to the next stage node according to the preset transfer rules, until the termination condition of the task scenario is reached, thereby completing the task. The preset transfer rules and termination conditions can be determined according to actual needs and are not limited here.

[0087] As can be seen from the above, the multi-agent interaction orchestration method provided in this application decomposes the task scenario into multiple stage nodes, each focusing on a single or a few objectives, thus clarifying the originally ambiguous task boundaries. Then, based on the tasks and executing agents corresponding to each stage node, the visibility level of the information generated by each stage node is determined. This allows for precise control of the information propagation range according to actual needs, preventing sensitive information from being acquired by irrelevant agents, reducing the risk of information leakage, and ensuring data security during task execution. Finally, based on the capability constraints of each agent set and the visibility levels of the tasks and information at each stage node, an interaction orchestration method for the agent set is generated. This enables the agent set to perform interactive operations based on the interaction orchestration method, allowing each agent to participate in the task at the appropriate time. This achieves reasonable allocation of agent resources, reduces collaboration conflicts, and improves the smoothness and effectiveness of overall collaboration.

[0088] Please see Figure 10 , Figure 10 This is a flowchart illustrating the implementation of a multi-agent interaction orchestration method provided in another embodiment of this application. Compared to... Figure 8 In a corresponding embodiment, this embodiment may further include S301~S303 after S103, as detailed below: In S301, when any stage node is detected to be activated, the subtask corresponding to the stage node is sent to the execution agent corresponding to the stage node.

[0089] In S302, the capability of the first information generated by the executing agent is verified based on the capability constraint information of the executing agent, and a first verification result is obtained; the first information is generated by the executing agent based on the subtask.

[0090] In this embodiment, combined with Figure 4 and Figure 7 When the terminal device detects that any stage node is activated, it sends the subtask corresponding to the stage node to the execution agent corresponding to the stage node. Then, the terminal device can perform capability verification on the initial information generated by the execution agent (i.e., the first information) based on the execution agent's capability constraint information. Figure 7 The flowchart of the capability constraint executor shown is used to obtain the first verification result. The first information is generated by the executing agent based on subtasks.

[0091] In some possible embodiments, a set of capabilities C(R) is preset for the executing agent R, and the terminal device can process the first information generated by the executing agent. Execute the following formula: ; in, This represents the first verification result, op( ) represents the action that the first information intends to perform, and Schema(R) represents the set of structured protocol constraints for the executing agent.

[0092] Combination Figure 7 If Valid=0, proceed to the correction chain: resampling or rollback. If it still fails, extract the smallest compliant subset from the first output. This serves as a fallback write to ensure the process can proceed smoothly and that disk writes are stable.

[0093] It should be noted that the first verification result includes verification pass and verification fail. Verification pass indicates that the ability of the intelligent agent to modify environmental state data is a compliant behavior, while verification fail indicates that the ability of the intelligent agent to modify environmental state data is a non-compliant behavior.

[0094] Taking a simple Werewolf game scenario as an example, the werewolf agent has the ability to modify the data of the voting targets for tonight in the environmental state data, which is a compliant behavior. However, if it tries to modify the voting data of the villagers, it is illegal, which is an unqualified behavior.

[0095] In one embodiment of this application, when the terminal device detects that the first verification result is a verification failure, it can perform the following... Figure 11 The steps S401 to S404 are shown.

[0096] In another embodiment of this application, when the terminal device detects that the first verification result is successful, it may execute step S303.

[0097] In S303, if the first verification result is successful, the first information is sent to the target intelligent agent based on the visibility level; the target intelligent agent refers to any intelligent agent in the set of intelligent agents that has the right to view the information.

[0098] In this embodiment, when the terminal device detects that the first verification result is successful, it indicates that the ability of the intelligent agent to modify certain environmental state data is a compliant behavior. Therefore, the terminal device can send the first information to the target intelligent agent based on the visibility level. The target intelligent agent refers to any intelligent agent in the set of intelligent agents that has the right to view this information.

[0099] It should be noted that the first piece of information includes modifications to the target environment state data. The target environment state data can be determined according to actual needs and is not restricted here.

[0100] As can be seen from the above, the multi-agent interaction orchestration method provided in this embodiment, when detecting the activation of any stage node, sends the subtask corresponding to the stage node to the executing agent corresponding to the stage node; performs capability verification on the first information generated by the executing agent based on the capability constraint information of the executing agent, and obtains a first verification result; the first information is generated by the executing agent based on the subtask; if the first verification result is successful, the first information is sent to the target agent based on the visibility level; the target agent refers to any agent in the agent set that has the right to view the information. This method constitutes a closed loop for the execution and monitoring of multi-agent interaction orchestration. Only when the capability verification result of the agent is successful will the information generated by the agent be dispatched to other agents with the right to view it, thereby ensuring the reliable progress of the task and the secure flow of information.

[0101] In one embodiment of this application, when the capability verification result of the intelligent agent is that the verification fails, in order to ensure the forward flow and stable disk write, the terminal device may specifically execute the following: Figure 11 The steps S401 to S404 are shown.

[0102] Therefore, please refer to Figure 11 , Figure 11 This is a flowchart illustrating the implementation of a multi-agent interaction orchestration method provided in another embodiment of this application. Compared to... Figure 10 In a corresponding embodiment, this embodiment may further include S401 to S404 after S302, as detailed below: In S401, if the first verification result is that the verification fails, the setting operation is performed on the executing agent.

[0103] In this embodiment, when the terminal device detects that the first verification result is a verification failure, it indicates that the ability of the executing agent to modify certain environmental state data is an irregular behavior. Therefore, the terminal device can perform setting operations on the executing agent. These setting operations include, but are not limited to, resampling and rollback.

[0104] In S402, after the setting operation is performed, the capability of the second information generated by the executing agent is verified to obtain the second verification result.

[0105] In this embodiment, after the terminal device performs the set operation, the executing agent can continue to generate second information. Then, the terminal device can continue to perform capability verification on this second information to obtain a second verification result. The second verification result includes, but is not limited to, verification passed and verification failed.

[0106] It should be noted that the specific implementation process of the above-mentioned capability verification of the second information can be found in the specific implementation process of step S302, and will not be repeated here.

[0107] In one embodiment of this application, when the terminal device detects that the second verification result is that the verification failed, it may execute steps S403 to S404.

[0108] In another embodiment of this application, when the terminal device detects that the second verification result is successful, it indicates that the ability of the intelligent agent to modify environmental state data is a compliant behavior. Therefore, the terminal device can send the second information to the target intelligent agent based on the visibility level.

[0109] In S403, if the second verification result is that the verification fails, the target information of the executing agent is determined based on the first information; the target information is different from the first information.

[0110] In S404, target information is sent to the target agent based on visibility level.

[0111] In this embodiment, when the terminal device detects that the second verification result is a verification failure, it indicates that the ability of the executing agent to modify environmental state data is still an illegal behavior. Therefore, the terminal device determines the target information of the executing agent based on the first information and sends the target information to the target agent based on the visibility level. The target information is different from the first information.

[0112] In particular, in conjunction with step S302, the target information can be the smallest compliance subset that exists when performing capability verification based on the first information. .

[0113] As can be seen from the above, the multi-agent interaction orchestration method provided in this embodiment, if the first verification result is a verification failure, then a setting operation is performed on the executing agent; after performing the setting operation, the capability of the second information generated by the executing agent is verified to obtain a second verification result; if the second verification result is a verification failure, then the target information of the executing agent is determined based on the first information; the target information is different from the first information; the target information is sent to the target agent based on the visibility level. This method constructs a multi-level fault tolerance and information repair mechanism by performing setting operations and determining the target information of the executing agent based on the first information, thereby ensuring the forward flow and stable disk storage.

[0114] In another embodiment of this application, when multiple agents perform interactive operations based on an interactive orchestration method, the terminal device may specifically execute the following steps, detailed below: Determine the type of the first piece of information; The first piece of information is stored based on its type.

[0115] It should be noted that the types include, but are not limited to: single agent memory and multiple group memory.

[0116] In this embodiment, the terminal device can determine the type of the first information based on the visibility level of the first information.

[0117] For example, if the visibility level of the first information is visible only to itself, the terminal device can determine that the first information is a single intelligent agent memory; if the visibility level of the first information is visible to the same role, visible to the same group, publicly visible, or visible within a custom range, the terminal device can determine that the first information is multiple group memories.

[0118] In another embodiment of this application, when multiple agents perform interactive operations based on an interactive orchestration method, the terminal device can specifically record events (e.g., when each agent performs its own task scenario, i.e., when it performs its own sub-task) based on three channels: the agent interaction channel, the environmental state data change channel, and the system log channel. Figure 3 (As shown).

[0119] In another embodiment of this application, after multiple agents have completed their interactive operations based on an interactive orchestration method, the terminal device, upon detecting a replay request, can determine the scene to be replayed based on the replay request. The replay request carries information about certain events visible to a specific agent at a certain time.

[0120] Therefore, the terminal device can determine the scene to be replayed based on the fact that certain events are visible to the intelligent agent that needs to be replayed at a certain time.

[0121] Afterwards, the terminal device can reconstruct the three-channel events corresponding to the scene to be replayed—the agent interaction channel, the environmental state data change channel, and the system log channel—based on a sorting rule. The sorting rule can be based on timestamps from earliest to latest.

[0122] It should be noted that the terminal device can also obtain the timestamp, time source agent, event type, visibility level, and event information carrier of the above-mentioned scene to be replayed.

[0123] In some possible embodiments, the terminal device can implement event replay through a deterministic reducer.

[0124] In this embodiment, the reduction is performed sequentially in partial order over time, which must satisfy the following formula: ; Reduce is a deterministic reducer that processes events sequentially according to a sorting rule (usually timestamps). This represents the scene state at time t. Let be the event at time i.

[0125] For any intelligent agent The set of visible events at time t as follows: ; in, This indicates task assignment.

[0126] Based on this, terminal devices can reproduce events with event log data as a single trusted source, which can be directly used for auditing and review.

[0127] The following will further illustrate the effectiveness of the multi-agent interaction orchestration method provided in the embodiments of this application using experimental data: Experimental setup: (1) Operating system: Ubuntu 20.04.6 LTS; (2) CPU: AMD EPYC 7763 64-Core Processor with 1.0TB DRAM; (3) GPU: NVIDIA A800-SXM4-80GB; (4) Inter-GPU interconnect: NVLink, peak bandwidth 50GB / s; (5) GPU CPU interconnect: PCIe Gen4 x16, bandwidth 16GT / s; Model configuration: (1) Model: DeepSeek-R1-0528 685B; (2) Number of scene rounds: 60 rounds; Final test results: Compared to the baseline simulation scenario, without distinguishing message visibility, using a single-dimensional memory system, and without capability constraints, in a simulated 60-round scenario, through online statistics and replay engine statistics, the multi-agent interaction orchestration method and multi-agent interaction orchestration system provided in this embodiment reduced the actual unauthorized access rate of each agent by 82%, the information leakage rate by 96%, and the replay consistency increased by 11% due to better constraints and specifications.

[0128] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0129] Corresponding to the multi-agent interaction orchestration method described in the above embodiments, Figure 12 A schematic diagram of a multi-agent interaction orchestration device according to an embodiment of this application is shown. For ease of explanation, only the parts related to the embodiment of this application are shown. (Refer to...) Figure 12The multi-agent interaction orchestration device 500 includes: a disassembly unit 51, a first determining unit 52, and a generating unit 53. Wherein: The disassembly unit 51 is used to disassemble the task scenario to obtain multiple stage nodes.

[0130] The first determining unit 52 is used to determine the visibility level of the information generated by each stage node based on the sub-tasks and executing agents corresponding to each stage node.

[0131] The generation unit 53 is used to generate an interaction orchestration method for the intelligent agent set based on the capability constraint information corresponding to each intelligent agent set, as well as the visibility level of the sub-tasks and information of each stage node, so that the intelligent agent set can perform interactive operations based on the interaction orchestration method; the intelligent agent set includes the executing intelligent agent.

[0132] In one embodiment of this application, the multi-agent interaction orchestration device 500 further includes: a second determining unit and a first execution unit. Wherein: The second determining unit is used to determine the task scenario when a task request is detected.

[0133] The first execution unit is used to perform profile binding operations on each intelligent agent based on the task scenario, thereby obtaining a set of intelligent agents.

[0134] In one embodiment of this application, the multi-agent interaction orchestration device 500 further includes: a first sending unit, a first verification unit, and a second sending unit. Wherein: The first sending unit is used to send the subtask corresponding to any stage node to the execution agent corresponding to the stage node when any stage node is detected to be activated.

[0135] The first verification unit is used to perform capability verification on the first information generated by the executing agent based on the capability constraint information of the executing agent, and obtain the first verification result; the first information is generated by the executing agent based on the sub-task.

[0136] The second sending unit is used to send the first information to the target intelligent agent based on the visibility level if the first verification result is successful; the target intelligent agent refers to any intelligent agent in the set of intelligent agents that has the right to view the information.

[0137] In one embodiment of this application, the multi-agent interaction orchestration device 500 further includes: a second execution unit, a second verification unit, a third determination unit, and a third sending unit. Wherein: The second execution unit is used to perform setting operations on the execution agent if the first verification result is that the verification fails.

[0138] The second verification unit is used to perform capability verification on the second information generated by the executing agent after the set operation is performed, and obtain the second verification result.

[0139] The third determining unit is used to determine the target information of the executing agent based on the first information if the second verification result is that the verification fails; the target information is different from the first information.

[0140] The third sending unit is used to send target information to the target agent based on the visibility level.

[0141] In one embodiment of this application, the multi-agent interaction orchestration device 500 further includes: a fourth determining unit and a storage unit. Wherein: The fourth determining unit is used to determine the type of the first information.

[0142] The storage unit is used to store the first piece of information based on its type.

[0143] In one embodiment of this application, the multi-agent interaction orchestration device 500 further includes a recording unit.

[0144] The recording unit is used to record events based on three channels: the agent interaction channel, the environmental state data change channel, and the system log channel, when each agent performs a task.

[0145] In one embodiment of this application, the multi-agent interaction orchestration device 500 further includes: a fifth determining unit and a reconstruction unit. Wherein: The fifth determining unit is used to determine the scene to be replayed when a replay request is detected.

[0146] The reconstruction unit is used to reconstruct the three-channel events corresponding to the scene to be replayed—the agent interaction channel, the environmental state data change channel, and the system log channel—based on sorting rules.

[0147] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0148] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0149] Figure 13 This is a schematic diagram of the structure of a terminal device provided in an embodiment of this application. Figure 13 As shown, the terminal device 6 in this embodiment includes: at least one processor 60 ( Figure 13 (Only one is shown in the diagram), memory 61, and computer program 62 stored in the memory 61 and executable on the at least one processor 60, wherein the processor 60 executes the computer program 62 to implement the steps in any of the above embodiments of the multi-agent interaction orchestration method.

[0150] The terminal device 6 may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will understand that... Figure 13 This is merely an example of terminal device 6 and does not constitute a limitation on terminal device 6. It may include more or fewer components than shown in the figure, or combine certain components, or different components, such as input / output devices, network access devices, etc.

[0151] The processor 60 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0152] In some embodiments, the memory 61 may be an internal storage unit of the terminal device 6, such as the RAM of the terminal device 6. In other embodiments, the memory 61 may be an external storage device of the terminal device 6, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the terminal device 6. Furthermore, the memory 61 may include both internal and external storage units of the terminal device 6. The memory 61 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 61 can also be used to temporarily store data that has been output or will be output.

[0153] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps described in the various method embodiments above.

[0154] This application provides a computer program product that, when run on a terminal device, enables the terminal device to implement the steps described in the various method embodiments above.

[0155] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a terminal device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium. Examples include USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0156] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0157] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for orchestrating multi-agent interactions, characterized in that, include: The task scenario is broken down into multiple stage nodes; Based on the sub-tasks and executing agents corresponding to each stage node, the visibility level of the information generated by each stage node is determined. Based on the capability constraint information corresponding to each of the intelligent agents and the visibility level of the subtasks and information of each stage node, an interaction orchestration method is generated for the intelligent agent set, so that the intelligent agent set can perform interactive operations based on the interaction orchestration method; the intelligent agent set includes multiple intelligent agents, and the multiple intelligent agents include the execution intelligent agent.

2. The multi-agent interaction orchestration method as described in claim 1, characterized in that, Before breaking down the task scenario into multiple stage nodes, the process also includes: Upon detecting a task request, the task scenario is determined; Based on the task scenario, a profile binding operation is performed on each intelligent agent to obtain the intelligent agent set.

3. The multi-agent interaction orchestration method as described in claim 1, characterized in that, After generating the interaction orchestration method for the intelligent agent set based on the capability constraint information corresponding to each intelligent agent set, and the visibility levels of subtasks and information of each stage node, the method further includes: When any stage node is detected to be activated, the subtask corresponding to the stage node is sent to the execution agent corresponding to the stage node. Based on the capability constraint information of the executing agent, the first information generated by the executing agent is verified to obtain a first verification result; the first information is generated by the executing agent based on the subtask. If the first verification result is successful, the first information is sent to the target agent based on the visibility level; the target agent refers to any agent in the set of agents that has the right to view the information.

4. The multi-agent interaction orchestration method as described in claim 3, characterized in that, After performing capability verification on the first information generated by the execution agent based on the capability constraint information of the execution agent to obtain the first verification result, the method further includes: If the first verification result is that the verification fails, then the setting operation is performed on the executing agent; After the setting operation is performed, the capability of the second information generated by the executing agent is verified to obtain the second verification result; If the second verification result is a verification failure, then the target information of the executing agent is determined based on the first information; the target information is different from the first information. The target information is sent to the target agent based on the visibility level.

5. The multi-agent interaction orchestration method as described in claim 3, characterized in that, The method further includes: Determine the type of the first information; The first information is stored based on the type.

6. The multi-agent interaction orchestration method as described in any one of claims 1-5, characterized in that, The method further includes: When each of the aforementioned intelligent agents executes the aforementioned task scenario, events are recorded based on three channels: the intelligent agent interaction channel, the environmental state data change channel, and the system log channel.

7. The multi-agent interaction orchestration method as described in claim 6, characterized in that, The method further includes: When a replay request is detected, the scene to be replayed is determined; Based on the sorting rules, the three channels of events corresponding to the scenario to be replayed—the agent interaction channel, the environmental state data change channel, and the system log channel—are reconstructed.

8. A multi-agent interactive orchestration system, characterized in that, include: At least one session scheduler, connected to each agent, is used to execute the multi-agent interaction orchestration method as described in any one of claims 1-5; An event tracing recorder is connected to at least one of the session schedulers and each of the intelligent agents, and is used to record events on the execution process of the task scenario corresponding to each of the session schedulers based on three channels: intelligent agent interaction channel, environmental state data change channel and system log channel, when each of the intelligent agents executes its corresponding task scenario. The replay engine, connected to the event tracing recorder, is used to reconstruct the three-channel events—the agent interaction channel, the environmental state data change channel, and the system log channel—corresponding to the scene to be replayed, based on sorting rules when a replay request is detected.

9. The multi-agent interactive orchestration system as described in claim 8, characterized in that, Each of the session schedulers specifically includes: A profile binding device is used to perform profile binding operations on each of the intelligent agents based on a task scenario to obtain a set of intelligent agents; the set of intelligent agents includes multiple intelligent agents. A visibility broadcaster, connected to the set of intelligent agents, is used to determine the corresponding target intelligent agent based on the visibility level of the information generated by each executing intelligent agent, and to send the information generated by each executing intelligent agent to its corresponding target intelligent agent based on the visibility level of the information generated by each executing intelligent agent; the executing intelligent agent refers to the intelligent agent used to execute the sub-tasks corresponding to each stage node, and each stage node is obtained by decomposing the task scenario; the target intelligent agent refers to any intelligent agent among the intelligent agents that has the right to view the information. A multidimensional memory manager, connected to each of the aforementioned intelligent agents, is used to store the information generated by each of the aforementioned intelligent agents; A capability constraint actuator is used to perform capability verification on the information generated by each of the intelligent agents and obtain the corresponding verification results.

10. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the multi-agent interaction orchestration method as described in any one of claims 1 to 7.

Images