Intelligent agent tool input method and device, equipment, storage medium and product

Abstract

The invention discloses an agent tool input method and device, equipment, a storage medium and a product, and relates to the technical field of agent tool management.The method comprises the steps that when a to-be-input tool applied to an agent is received, function description, application scenes and calling parameters of the to-be-input tool are determined; according to the function description, the applicable scene and the calling parameter, judging whether a callable tool library of the intelligent agent contains a target tool conflicting with the tool to be input or not; if yes, differentiating the to-be-input tool and the target tool, and then inputting the to-be-input tool and the target tool into a callable tool library; and if not, inputting the to-be-input tool into the callable tool library. According to the method, redundancy and interference caused by agent tool conflicts are avoided, so that the decision burden and the error probability of the agents in the tool calling process are reduced, and the reliability and the task execution efficiency of the agents are improved.

Description

Technical Field

[0001] This application relates to the field of intelligent agent tool management technology, and in particular to an intelligent agent tool input method, apparatus, device, storage medium and product. Background Technology

[0002] With the rapid development of artificial intelligence technology, the number of tools that intelligent agents can invoke and execute tasks continues to grow, and the tool library is constantly expanding. Against this backdrop, due to the lack of an effective mechanism for identifying and coordinating tool conflicts, newly added tools may overlap in function or have potential logical conflicts with existing tools, leading to redundancy and interference within the tool library. Such conflicts not only increase the decision-making burden on intelligent agents during task invocation but may also cause task execution errors or inconsistent results, seriously affecting the reliability of intelligent agents and task execution efficiency. Therefore, how to promptly identify and handle conflicts between tools during the intelligent agent tool entry phase has become a pressing technical problem in the field of intelligent agent tool management. Summary of the Invention

[0003] The main objective of this application is to provide a method, apparatus, device, storage medium, and product for inputting intelligent agent tools, aiming to solve the technical problem of how to promptly detect and handle conflicts between tools during the input stage of intelligent agent tools.

[0004] To achieve the above objectives, this application provides a method for inputting intelligent agent tools, the method comprising the following steps: When a tool to be input is received for an intelligent agent, the function description, applicable scenarios, and calling parameters of the tool to be input are determined; Based on the function description, the applicable scenario, and the calling parameters, determine whether the callable tool library of the intelligent agent contains a target tool that conflicts with the tool to be entered; If so, the tool to be entered and the target tool are differentiated before being entered into the callable tool library; If not, the tool to be entered will be entered into the callable tool library.

[0005] In one embodiment, the step of determining the function description, applicable scenario, and calling parameters of the input tool to be applied to the intelligent agent when receiving the input tool to be applied to the intelligent agent includes: When a tool to be input is received for use by an intelligent agent, the source code of the tool to be input is parsed to obtain the interface definition, input and output parameters, function method names and code comments of the tool to be input. Based on the interface definition, the input and output parameters, the function method name, and the code comments, determine the function description, applicable scenarios, and calling parameters of the input tool.

[0006] In one embodiment, the step of determining whether the callable tool library of the intelligent agent contains a target tool that conflicts with the tool to be entered, based on the function description, the applicable scenario, and the calling parameters, includes: The functional description, the applicable scenarios, and the calling parameters are semantically encoded to obtain the semantic vector to be entered. The semantic vector to be entered is compared with the stored semantic vectors in the callable tool library of the agent, and the comparison result determines whether the callable tool library contains a target tool that conflicts with the tool to be entered.

[0007] In one embodiment, the step of performing differentiation processing on the tool to be entered and the target tool before entering them into the callable tool library includes: A first prompt word is generated based on the input tool and the target tool, and the first prompt word is input into a first preset large model. The first preset large model is used to refine the semantic vectors of the input tool and the target tool based on the first prompt word. After differentiating the input tool and the target tool based on the first result output by the first preset large model, the input tool is then entered into the callable tool library.

[0008] In one embodiment, the intelligent agent tool input method further includes: In response to the input event of the tool to be inputted, the semantic vector and tool type of the tool to be inputted are determined, and the retrieval index of the callable tool library is updated according to the semantic vector and the tool type; The tool types include basic tools and composite tools. The basic tools are the smallest units that constitute the tools of an intelligent agent, and the composite tools are combinations of multiple basic tools.

[0009] In one embodiment, after the step of updating the retrieval index of the callable tool library based on the semantic vector and the tool type, the method further includes: A second prompt word is generated based on user intent data and tool call logs of the callable tool library, and the second prompt word is input into a second preset large model. The second preset large model is used to perform tool combination mining on the callable tool library based on the second prompt word. Based on the second result output by the second preset large model, the tool combination scheme of the callable tool library is determined, and the tool combination scheme is verified for inclusion in the library. If the tool combination scheme passes the database entry verification, then the tool combination scheme will be entered into the callable tool library.

[0010] Furthermore, to achieve the above objectives, this application also proposes an intelligent agent tool input device, which includes: The information acquisition module is used to determine the function description, applicable scenario and calling parameters of the input tool to be applied to the intelligent agent when it receives the input tool to be applied to the intelligent agent. The condition judgment module is used to determine, based on the function description, the applicable scenario and the calling parameters, whether the callable tool library of the intelligent agent contains a target tool that conflicts with the tool to be entered; The first input module is used to perform differential processing on the tool to be input and the target tool, and then input them into the callable tool library if the conditions are met. The second input module is used to input the tool to be input into the callable tool library if no.

[0011] In addition, to achieve the above objectives, this application also proposes an intelligent agent tool input device, the device comprising: a memory, a processor, and an intelligent agent tool input program stored in the memory and executable on the processor, the intelligent agent tool input program being configured to implement the steps of the intelligent agent tool input method as described above.

[0012] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, storing an intelligent agent tool input program, which, when executed by a processor, implements the steps of the intelligent agent tool input method as described above.

[0013] In addition, to achieve the above objectives, the present invention also provides a computer program product, the computer program product including an intelligent agent tool input program, which, when executed by a processor, implements the steps of the intelligent agent tool input method as described above.

[0014] When this application receives a tool to be entered for an intelligent agent, it determines the function description, applicable scenario, and calling parameters of the tool. Based on the function description, applicable scenario, and calling parameters, it determines whether the intelligent agent's callable tool library contains a target tool that conflicts with the tool to be entered. If so, the tool to be entered and the target tool are differentiated before being entered into the callable tool library; otherwise, the tool to be entered into the callable tool library. This method, before the tool to be entered into the callable tool library, determines whether it conflicts with existing tools in the library based on its function description, applicable scenario, and calling parameters. Upon detecting a conflict, it differentiates the conflicting tools before entering the library, avoiding redundancy and interference caused by tool conflicts. This eliminates the need for complex choices between conflicting tools when the intelligent agent performs subsequent tasks, reducing the decision-making burden and error probability during tool invocation, thereby improving the reliability and task execution efficiency of the intelligent agent. Attached Figure Description

[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the first embodiment of the intelligent agent tool input method of this application; Figure 2 This is a flowchart illustrating the second embodiment of the intelligent agent tool input method of this application; Figure 3 This is a flowchart illustrating the third embodiment of the intelligent agent tool input method of this application; Figure 4 This is a structural block diagram of the first embodiment of the intelligent agent tool input device of this application; Figure 5 This is a schematic diagram of the intelligent agent tool input device of this application.

[0018] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of this application and are not intended to limit this application.

[0020] It should be noted that the executing entity in the embodiments of this application can be a computing service device with data processing, network communication, and program execution functions, such as a smart wearable device, a personal computer, or a mobile phone, or an electronic device capable of performing the above functions, such as the aforementioned intelligent agent tool input device. The following embodiments will be described using the intelligent agent tool input device as an example.

[0021] This application provides a method for inputting intelligent agent tools, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the intelligent agent tool input method of this application.

[0022] In this embodiment, the intelligent agent tool input method includes the following steps: Step S10: When a data entry tool to be applied to an intelligent agent is received, determine the function description, applicable scenarios, and calling parameters of the data entry tool.

[0023] It should be noted that the aforementioned intelligent agent refers to an intelligent system with autonomous reasoning and task execution capabilities. It is typically built upon a large model and can understand user intent, plan task steps, invoke external tools, and integrate results to complete a specific task. For example, an AI system for intelligent customer service can understand the user's combined request of "checking order status and recommending related products," autonomously invoke the "order inquiry tool" and "product recommendation tool," execute them in logical order, and return the integrated response. The aforementioned "tools to be entered" refer to newly registered intelligent agent tools that can be invoked by the intelligent agent after successful registration and entry. Tools to be entered can be divided into basic tools and composite tools. Basic tools are indivisible units with a single atomic function, such as an "image compression tool," which only provides the single function of compressing image files. Composite tools are composed of multiple basic tools or other composite tools combined according to preset logic to achieve more complex business functions, such as a "travel planning tool," which is composed of a "map navigation tool," a "hotel booking tool," and a "weather query tool" combined in sequence or parallel logic to complete travel planning tasks in one stop.

[0024] It should be understood that the above functional descriptions represent a textual description of the functions, input / output characteristics, and usage methods of the input tool, typically including the tool name, functional overview, input parameters, output results, and usage examples. For example, for an "image compression tool," the functional description could be: This tool is used to perform lossy or lossless compression on uploaded image files, supports JPEG and PNG format input, and outputs the compressed image file and its size information. The above applicable scenarios indicate the specific task context or business domain in which the input tool is designed or recommended for use, guiding the agent in when to choose to call the tool to improve task matching and execution efficiency. For example, the applicable scenarios for a "PDF parsing tool" could be described as: suitable for extracting text, tables, and image content from PDF documents, commonly used in document analysis, report generation, and knowledge base construction. The above calling parameters represent the input information or configuration items that the agent needs to provide when calling the input tool, typically including required and optional parameters, guiding the tool to execute correctly and return the expected results. For example, for a "weather query tool", the required parameters may include the city or coordinates to be queried, and the optional parameters may include the temperature unit, the number of days in the forecast, etc.

[0025] In the specific implementation, when a newly registered tool to be entered is received, the unified protocol defined by the tool registration module can be used, along with the definitional information or structured description provided by the tool to be entered, to standardize the functional features, usage conditions and interface specifications, thereby obtaining the functional description, applicable scenarios and calling parameters of the tool to be entered.

[0026] Step S20: Based on the function description, the applicable scenario, and the calling parameters, determine whether the callable tool library of the intelligent agent contains a target tool that conflicts with the tool to be entered.

[0027] It should be noted that the aforementioned callable tool library refers to a collection of agent tools that, after being organized, stored, and indexed in a hierarchical structure by the tool management module, can be directly retrieved and invoked by the agent upon receiving a task. The aforementioned target tool refers to a stored tool in the callable tool library that has a high degree of similarity, overlap, or potential logical conflict with the tool to be entered in terms of function, scenario, or parameters.

[0028] It should be understood that, based on the functional description, applicable scenarios, and calling parameters, it can be determined whether the callable tool library of the intelligent agent contains stored tools that satisfy phenomena such as functional overlap with the tool to be entered (i.e., both can complete the same or extremely similar core tasks), scenario confusion (i.e., the applicable scenario descriptions of the two are vague or overlapping, leading to difficulties in the intelligent agent's selection), and logical interference (i.e., when the two are called in parallel or sequentially, errors may occur due to resource contention, state mutual exclusion, etc.). If so, it can be determined that the tool is a target tool that conflicts with the tool to be entered; otherwise, it can be determined that the callable tool library does not contain the target tool.

[0029] Step S30: If so, the tool to be entered and the target tool are differentiated before being entered into the callable tool library.

[0030] Understandably, differentiating the input tool and the target tool involves semantic analysis and reconstruction of their functional descriptions and applicable scenarios to transform them from a conflicting state to a coexisting state of complementary functions or separate scenarios. For example, by adding qualifiers, specifying particular parameters, and defining exclusive scenarios, new, mutually exclusive (or complementary) tool information can be generated. After processing, the differentiated information can be used to overwrite or update the original information of the input tool, while simultaneously updating the new information of the target tool. Finally, the new information for both the input tool and the target tool is stored in a callable tool library to complete the input process.

[0031] Step S40: If not, then input the tool to be entered into the callable tool library.

[0032] It should be understood that if the available tool library does not contain a target tool that conflicts with the tool to be entered, the tool to be entered can be directly entered into the available tool library.

[0033] In this embodiment, when a tool to be entered into is received for an intelligent agent, its functional description, applicable scenario, and calling parameters are determined. Based on the functional description, applicable scenario, and calling parameters, it is determined whether the intelligent agent's callable tool library contains a target tool that conflicts with the tool to be entered. If so, the tool to be entered and the target tool are differentiated before being entered into the callable tool library; otherwise, the tool to be entered into the callable tool library. This method, before the tool to be entered into the callable tool library, determines whether it conflicts with existing tools in the library based on its functional description, applicable scenario, and calling parameters. When a conflict is detected, the conflicting tools are differentiated before being entered into the tool library. This avoids redundancy and interference caused by tool conflicts, allowing the intelligent agent to avoid complex choices between conflicting tools during subsequent task execution. This reduces the decision-making burden and error probability of the intelligent agent during tool invocation, thereby improving the reliability and task execution efficiency of the intelligent agent.

[0034] Reference Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the intelligent agent tool input method of this application.

[0035] In one feasible implementation, step S10 may include: Step S101: When a data entry tool to be applied to an intelligent agent is received, the source code of the data entry tool is parsed to obtain the interface definition, input and output parameters, function method names and code comments of the data entry tool.

[0036] It should be noted that the above interface definition represents the formal specification of the callable entry point exposed by the tool, identified from the source code; the above input and output parameters represent the typed constraints on data exchange further extracted from the interface definition. The input parameters define the variables and their data types that must be provided or are optional when calling the tool, and the output parameters define the data structure returned after the tool is executed; the above function and method names represent the string identifiers used to uniquely identify the callable unit at the code level, identified in the interface definition, and are the direct handles for tool invocation; code comments represent natural language text paragraphs embedded in the source code, marked with specific symbols, used to describe functions, parameters, return values, or usage examples.

[0037] Step S102: Determine the function description, applicable scenarios, and calling parameters of the input tool based on the interface definition, the input / output parameters, the function / method name, and the code comments.

[0038] In practical implementation, function names and code comments (especially the summary describing the overall functionality) can be comprehensively analyzed to extract and reconstruct concise and accurate natural language function statements. For example, the method name `compress_image` and the comment "Compress image files using algorithm X" can be transformed into the function description "This tool is used for lossy compression of image files." Based on the use cases mentioned in the code comments, the data types of input and output parameters (such as images, text, and numbers), and the function description, one or more scenario labels or descriptive statements can be assigned to the tool through classification models or keyword matching. For example, from processing the data type "image," applicable scenarios such as "multimedia processing" and "front-end optimization" can be inferred.

[0039] In one feasible implementation, step S20 may include: Step S201: Semantically encode the function description, the applicable scenario, and the calling parameters to obtain the semantic vector to be entered.

[0040] In practical implementation, the function description, applicable scenarios, and calling parameters can be concatenated into a coherent and complete text paragraph according to a predefined template. For example, "Function: [Function Description]. Applicable Scenarios: [Scenario A, Scenario B]. Calling Parameters: Input must include [Parameter 1: Type], [Parameter 2: Type]...". This text paragraph is then input into a pre-defined semantic encoding model. The model performs forward propagation calculations on the input text through its internal neural network weights, and finally extracts a high-dimensional feature vector from its output layer or a specific pooling layer. This feature vector is the semantic vector to be entered.

[0041] Step S202: Compare the semantic vector to be entered with the stored semantic vectors in the callable tool library of the agent, and determine whether the callable tool library contains a target tool that conflicts with the tool to be entered based on the comparison result.

[0042] It should be noted that the aforementioned stored semantic vectors represent the semantic vectors of each stored tool in the callable tool library, which are pre-processed and persistently stored using a preset semantic encoding model (the same model as in step S201).

[0043] In practical implementation, a selected similarity metric function (such as cosine similarity) can be used to calculate the similarity between the semantic vector to be entered and each stored semantic vector. Then, based on the similarity comparison results, it can be determined whether the available tool library contains a target tool that conflicts with the tool to be entered. For example, it can be pre-set that if the similarity between the semantic vector to be entered and the stored semantic vector exceeds a preset threshold (such as 80%), it is considered that there is a conflict between the target tool corresponding to the stored semantic vector and the tool to be entered.

[0044] In this embodiment, when a tool to be input is received for an intelligent agent, the source code of the tool is parsed to obtain its interface definition, input / output parameters, function names, and code comments. Based on the interface definition, input / output parameters, function names, and code comments, the functional description, applicable scenarios, and calling parameters of the tool are determined. The functional description, applicable scenarios, and calling parameters are semantically encoded to obtain a semantic vector to be input. The semantic vector to be input is compared with the stored semantic vectors in the intelligent agent's callable tool library. Based on the comparison result, it is determined whether the callable tool library contains a target tool that conflicts with the tool to be input. This method, by parsing the source code, can extract information such as interface definitions, function names, and code comments, thus providing a more objective data foundation for accurately deriving the tool's functional description, applicable scenarios, and calling parameters. Compared to relying on manual or simple text descriptions, this method reduces subjectivity and ambiguity in the definition process. Subsequently, the semantic vector of the tool to be entered is compared with the semantic vector of the existing tools in the tool library. The potential functional overlap and logical conflict between the tools are deeply identified from the semantic level. This allows for more accurate early detection of conflicts during the tool entry stage, which helps to reduce the problems of agent task execution errors and inefficiency caused by duplicate or mutually exclusive tool functions from the source.

[0045] Reference Figure 3 , Figure 3 This is a flowchart illustrating the third embodiment of the intelligent agent tool input method of this application.

[0046] In one feasible implementation, step S30 may include: Step S301: Generate a first prompt word based on the tool to be entered and the target tool, and input the first prompt word into a first preset large model. The first preset large model is used to refine the semantic vectors of the tool to be entered and the target tool based on the first prompt word.

[0047] It should be noted that the aforementioned first prompt integrates metadata such as the functional descriptions, applicable scenarios, and calling parameters of the tool to be input and the target tool, and contains explicit guiding instructions to guide the large model to perform the task of refining the semantic vectors of the tool to be input and the target tool. The aforementioned first preset large model refers to a model with natural language understanding and generation capabilities that is pre-integrated and invoked, such as a large language model.

[0048] In the specific implementation, the information of the two tools (e.g., the function description of the tool to be entered is X, and the applicable scenario is Y; the function description of the target tool is P, and the applicable scenario is Q) can be filled in according to the preset template, and a clear instruction can be attached (e.g., "Please analyze the descriptions of the following two tools, clarify their functional boundaries, and generate a more specific semantic vector for each tool that can avoid confusion"). The text obtained after the template is filled in is the first prompt word mentioned above.

[0049] It should be understood that after identifying a conflict between the input tool and the target tool, the aforementioned first prompt word is input into the first preset large model. Utilizing the semantic understanding and generation capabilities of the first preset large model, the original semantic vectors of the input tool and the target tool are modified, supplemented, or restated. This refines the semantic vectors of the input tool and the target tool, making them more semantically distinct, thus clarifying their respective functional boundaries and applicable conditions to eliminate conflicts. For example, the newly registered "image compression tool" (i.e., the input tool) and the existing "image optimization tool" (i.e., the target tool) both support JPEG image compression, resulting in functional duplication. When a conflict is detected, the system initiates an optimization process, using the first preset large model to refine the corresponding semantic vectors of the two tools to clarify their differences. For instance, "image compression tool" is refined to "a tool specifically for Web images that supports lossless compression," while "image optimization tool" is refined to "general image processing that supports lossy compression."

[0050] Step S302: Based on the first result output by the first preset large model, perform differentiation processing on the tool to be entered and the target tool, and then enter them into the callable tool library.

[0051] It should be noted that the aforementioned first result refers to the structured revised text output by the first preset large model after analyzing the conflicting input tools and target tools based on the input prompts. This result includes newly generated, clearly distinguishable functional descriptions and / or new applicable scenarios and / or optimized calling parameter definitions for both.

[0052] In a specific implementation, the original semantic vectors of the tool to be entered and the target tool can be overwritten with new semantic vectors from the first result mentioned above. Then, the tool to be entered, the target tool, and their respective new semantic vectors can be entered into the callable tool library.

[0053] In one feasible implementation, the intelligent agent tool input method may further include: Step S50: In response to the input event of the tool to be inputted, determine the semantic vector and tool type of the tool to be inputted, and update the retrieval index of the callable tool library according to the semantic vector and the tool type.

[0054] The tool types include basic tools and composite tools. The basic tools are the smallest units that constitute the tools of an intelligent agent, and the composite tools are combinations of multiple basic tools.

[0055] It's important to note that basic tools refer to intelligent agent tools with single, atomic functions. They are the fundamental modules that constitute complex capabilities, and are indivisible. Their characteristics include clearly defined functions, simple interfaces, strong independence, and fast execution speed. Composite tools are composed of two or more basic tools (or even other composite tools) arranged and combined through specific logic to accomplish a more complex and higher-level task objective. Their characteristics include functional compositeness, business-scenario orientation, and encapsulation of internal calling logic. The core of composite tools is their internal "workflow orchestration," and the composite methods include at least the following four: Sequential chaining: The simplest method, where the output of tool A serves as the input of tool B, and they are executed sequentially. Example: User intent analysis → calling a search engine → generating a result summary. Parallel execution: Calling multiple tools simultaneously and then summarizing all results. Example: Simultaneously calling get_stock_price(symbol_A) and get_stock_price(symbol_B), then returning a comparison report containing the prices of both. Conditional branching: Determining which tool to call next based on the execution result of the previous tool or a certain condition. Iterative looping: Repeatedly calling a tool until a specific condition is met. Example: The `check_price(item_id)` function is called repeatedly until the price falls below a user-defined threshold, at which point the `send_notification` tool is triggered. When a new composite tool is added to the library, it will be vectorized and added to the callable tool library, just like the base tools.

[0056] In practice, the semantic vector of the tool to be entered can be added to the vector index database, so that the tool can be retrieved through subsequent semantic similarity search (such as cosine similarity calculation); the tool type of the tool to be entered can also be added to the inverted list, so that the tool can be retrieved through conditional retrieval methods such as type filtering and keyword matching.

[0057] In one feasible implementation, after step S50, the following may also be included: Step S60: Generate a second prompt word based on the user intent data and the tool call log of the callable tool library, and input the second prompt word into the second preset large model. The second preset large model is used to perform tool combination mining on the callable tool library based on the second prompt word.

[0058] It should be noted that the aforementioned user intent data represents data that is abstracted from the interaction history between the agent and the user, and represents the user's task goals. It typically includes high-frequency task goals (such as itinerary planning and report generation) and core query phrases. The aforementioned tool call log represents the tool call sequence of the agent's historical tasks, which can reveal the tool combination patterns in the real world. In addition, the tool call log can also contain standardized information on all basic and composite tools in the callable tool library.

[0059] It should be understood that the aforementioned second prompt word refers to the instructional text constructed according to a predefined mining task template based on user intent data and tool call logs, guiding the large model to perform tool combination mining. For example, the model may be required to output each potential tool combination scheme in a specified structured JSON format, including the scheme name, description, constituent tools, workflow logic, and efficiency gain estimation. The aforementioned second preset large model refers to a pre-integrated and invoked basic large language model with natural language understanding and generation capabilities. The second preset large model can be the same model as the first preset large model, or it can be a different model; this embodiment does not impose any restrictions on this.

[0060] It should be understood that during the iterative optimization of the tool library, a second prompt word containing user intent data and tool call logs can be input into the second preset large model. Based on its semantic understanding and pattern discovery capabilities, the second preset large model analyzes and infers the high-frequency task objectives, tool usage sequences, and tool function semantics reflected by the second prompt word, thereby identifying potential efficient tool combination patterns and outputting specific tool combination schemes.

[0061] In practice, the above process can be carried out in the tool combination recommendation engine module. This module regularly (e.g., once a day) obtains the latest information of the callable tool library from the tool management module, uses a large model to analyze basic tools and composite tools, and explores potential efficient tool combination schemes (e.g., "map navigation tool + hotel booking tool" can be combined into "travel planning composite tool").

[0062] Step S70: Based on the second result output by the second preset large model, determine the tool combination scheme of the callable tool library, and verify the tool combination scheme for inclusion in the library.

[0063] For example, in a tool combination mining operation, the second prompt word generated based on user intent data and tool call logs is: "You are a top AI system architect. Your task is to mine efficient combination solutions that can be created as 'composite tools' based on the following tool library to improve the efficiency of the intelligent agent in performing tasks." Upon receiving the second prompt word, the second preset large model automatically generates the following mining instructions: 1. Identify high-frequency task patterns: Analyze the functions and scenarios of tools, imagine a typical user goal (such as "planning a business trip" or "writing a market analysis report"), and see which basic tools are usually called sequentially or in parallel to achieve these goals? 2. Identify efficiency bottlenecks: Consider which tool combinations are frequently invoked, and encapsulate these invocations into a single composite tool. This can reduce the decision-making steps of the agent, lower network latency, and simplify user interaction.

[0064] 3. Propose Combinations: For each potential combination you discover, output a solution in JSON format. The solution must include: - `proposed_name`: Give the new composite tool a concise and clear name.

[0065] - `description`: Provide a detailed description of the functionality and value of this composite tool.

[0066] - `constituent_tools`: Lists the names of the underlying tools it contains.

[0067] - `workflow_logic`: Briefly describe the calling logic between tools (such as "sequential call", "parallel call followed by aggregation").

[0068] - `efficiency_gain`: Estimates the efficiency improvement this combination will bring (e.g., "reduce 3 API calls" or "reduce task completion time from 5 seconds to 2 seconds").

[0069] Based on the mining instructions in the example above, the second result output by the second preset large model is: { "proposed_name": "BusinessTripPlanner", "description": "A one-stop business travel planning tool that automatically integrates flight, hotel, and weather information." "constituent_tools": ["find_flights", "book_hotel", "get_weather"], "workflow_logic": "In parallel calls to find_flights and book_hotel to obtain the destination, then call get_weather, and finally integrate the information to generate a report." "efficiency_gain": "Encapsulates a task that originally required three independent calls into one, reducing the inference burden on the agent." }

[0070] Furthermore, after determining the tool combination scheme of the callable tool library based on the second result above, two verifications can be performed on the tool combination scheme: functional integrity verification, that is, verifying whether the tool combination scheme covers the target task requirements; and efficiency superiority verification, that is, verifying whether the tool combination scheme is more efficient than calling the basic tools alone.

[0071] Step S80: If the tool combination scheme passes the database entry verification, then the tool combination scheme is entered into the callable tool library.

[0072] It should be understood that the tool combination scheme that passes the entry verification can be packaged into a new composite tool. The new composite tool can be regarded as a tool to be entered. Following the process of steps 10 to 40 above, it is re-entered into the callable tool library through the tool registration, conflict detection and other processes, thereby completing the iteration of the callable tool library.

[0073] Based on the above embodiments, when the intelligent agent receives a task instruction, it can provide relevant guidance in the prompt words according to the functional requirements and complexity of the task, so as to retrieve the appropriate tool (basic tool or composite tool) from the tool library of the tool management layer and initiate a call request. The intelligent agent can then use the called tool to complete the task execution.

[0074] This embodiment generates a first prompt word based on the tool to be entered and the target tool, and inputs the first prompt word into a first preset large model. The first preset large model is used to refine the semantic vectors of the tool to be entered and the target tool based on the first prompt word. After differentiating the tool to be entered and the target tool based on the first result output by the first preset large model, the tools are then entered into the callable tool library. In response to the entry event of the tool to be entered, the semantic vector and tool type of the tool to be entered are determined, and the retrieval index of the callable tool library is updated according to the semantic vector and the tool type. The tool type includes... The method employs basic tools and composite tools. Basic tools are the smallest units constituting the tools of an intelligent agent, while composite tools are combinations of multiple basic tools. A second prompt word is generated based on user intent data and the tool call logs of the callable tool library. This second prompt word is input into a second preset large model, which is used to mine tool combinations from the callable tool library based on the second prompt word. A tool combination scheme is determined based on the second result output by the second preset large model, and the scheme is validated for inclusion in the library. If the scheme passes the validation, it is entered into the callable tool library. This embodiment utilizes a preset large model to differentiate the semantic vectors of conflicting tools, resolving the functional overlap between tools and enhancing the coordination of the tool library. After tool entry, the retrieval index is updated by combining its semantic vector with the tool type, constructing a more efficient hierarchical retrieval system. This allows the intelligent agent to quickly locate the most suitable tool based on task complexity, improving task invocation efficiency. Furthermore, this embodiment can also perform tool combination mining on the callable tool library to discover potential efficient composite solutions, and form new composite tools after verification in the library, thereby realizing the self-iteration of the tool library and reducing the reasoning and calling costs of intelligent agents when handling complex tasks.

[0075] Reference Figure 4 , Figure 4 This is a structural block diagram of the first embodiment of the intelligent agent tool input device of this application.

[0076] like Figure 4 As shown, the intelligent agent tool input device proposed in this application includes: Information acquisition module 401 is used to determine the function description, applicable scenario and calling parameters of the input tool to be used by the intelligent agent when it receives the input tool to be used by the intelligent agent. The condition judgment module 402 is used to determine, based on the function description, the applicable scenario and the calling parameters, whether the callable tool library of the intelligent agent contains a target tool that conflicts with the tool to be entered; The first input module 403 is used to perform differential processing on the tool to be input and the target tool, and then input them into the callable tool library if the conditions are met. The second input module 404 is used to input the tool to be input into the callable tool library if no.

[0077] In this embodiment, when a tool to be entered into is received for an intelligent agent, its functional description, applicable scenario, and calling parameters are determined. Based on the functional description, applicable scenario, and calling parameters, it is determined whether the intelligent agent's callable tool library contains a target tool that conflicts with the tool to be entered. If so, the tool to be entered and the target tool are differentiated before being entered into the callable tool library; otherwise, the tool to be entered into the callable tool library. This method, before the tool to be entered into the callable tool library, determines whether it conflicts with existing tools in the library based on its functional description, applicable scenario, and calling parameters. When a conflict is detected, the conflicting tools are differentiated before being entered into the tool library. This avoids redundancy and interference caused by tool conflicts, allowing the intelligent agent to avoid complex choices between conflicting tools during subsequent task execution. This reduces the decision-making burden and error probability of the intelligent agent during tool invocation, thereby improving the reliability and task execution efficiency of the intelligent agent.

[0078] Based on the first embodiment of the intelligent agent tool input device described in this application, a second embodiment of the intelligent agent tool input device of this application is proposed.

[0079] In this embodiment, the information acquisition module 401 is further configured to, when receiving a tool to be entered applied to an intelligent agent, parse the source code of the tool to be entered to obtain the interface definition, input and output parameters, function method names and code comments of the tool to be entered; and determine the functional description, applicable scenarios and calling parameters of the tool to be entered based on the interface definition, the input and output parameters, the function method names and the code comments.

[0080] Furthermore, the condition judgment module 402 is also used to perform semantic encoding on the function description, the applicable scenario and the calling parameters to obtain a semantic vector to be entered; compare the semantic vector to be entered with the stored semantic vectors in the callable tool library of the intelligent agent, and determine whether the callable tool library contains a target tool that conflicts with the tool to be entered based on the comparison result.

[0081] Furthermore, the first input module 403 is also used to generate a first prompt word based on the tool to be input and the target tool, and input the first prompt word into a first preset large model. The first preset large model is used to refine the semantic vectors of the tool to be input and the target tool based on the first prompt word. After differentiating the tool to be input and the target tool based on the first result output by the first preset large model, the tool is then input into the callable tool library.

[0082] Furthermore, the second input module 404 is also used to respond to the input event of the tool to be input, determine the semantic vector and tool type of the tool to be input, and update the retrieval index of the callable tool library according to the semantic vector and the tool type; wherein, the tool type includes basic tools and composite tools, the basic tools are the smallest units constituting intelligent agent tools, and the composite tools are a combination of multiple basic tools.

[0083] Furthermore, the second input module 404 is also used to generate a second prompt word based on user intent data and the tool call log of the callable tool library, and input the second prompt word into a second preset large model. The second preset large model is used to perform tool combination mining on the callable tool library based on the second prompt word; determine the tool combination scheme of the callable tool library based on the second result output by the second preset large model, and perform database entry verification on the tool combination scheme; if the tool combination scheme passes the database entry verification, then the tool combination scheme is entered into the callable tool library.

[0084] Other embodiments or specific implementations of the intelligent agent tool input device of this application can be found in the above-described method embodiments, and will not be repeated here.

[0085] This application provides an intelligent agent tool input device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the intelligent agent tool input method in the above embodiment 1.

[0086] The following reference Figure 5The diagram illustrates a structural schematic of an intelligent agent tool input device suitable for implementing embodiments of this application. The intelligent agent tool input device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 5 The intelligent agent tool input device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0087] like Figure 5 As shown, the intelligent agent tool input device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the intelligent agent tool input device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. The communication device 1009 allows the intelligent agent tool input device to communicate wirelessly or wiredly with other devices to exchange data. Although the figures show intelligent agent tool input devices with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0088] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0089] The intelligent agent tool input device provided in this application, employing the intelligent agent tool input method in the above embodiments, can solve the technical problem of how to promptly detect and handle conflicts between tools during the intelligent agent tool input stage. Compared with the prior art, the beneficial effects of the intelligent agent tool input device provided in this application are the same as those of the intelligent agent tool input method provided in the above embodiments, and other technical features in this intelligent agent tool input device are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0090] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0091] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0092] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the intelligent agent tool input method in the above embodiments.

[0093] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0094] The aforementioned computer-readable storage medium may be included in the intelligent agent tool input device; or it may exist independently and not be assembled into the intelligent agent tool input device.

[0095] The aforementioned computer-readable storage medium carries one or more programs that, when executed by the intelligent agent tool input device, enable the intelligent agent tool input device to write computer program code for performing the operations of this application in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages ​​such as Java, Smalltalk, and C++; and also conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., via the Internet using an Internet service provider).

[0096] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0097] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0098] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the above-described intelligent agent tool input method. This solves the technical problem of how to promptly detect and handle conflicts between tools during the intelligent agent tool input stage. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the intelligent agent tool input method provided in the above embodiments, and will not be repeated here.

[0099] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the intelligent agent tool input method described above.

[0100] The computer program product provided in this application can solve the technical problem of intelligent agent tool input. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as the beneficial effects of the intelligent agent tool input method provided in the above embodiments, and will not be repeated here.

[0101] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.

Claims

1. An agent tool entry method, characterized by, The method includes the following steps: When a tool to be input is received for an intelligent agent, the function description, applicable scenario, and calling parameters of the tool to be input are determined; Based on the function description, the applicable scenario, and the calling parameters, determine whether the callable tool library of the intelligent agent contains a target tool that conflicts with the tool to be entered; If so, the tool to be entered and the target tool are differentiated before being entered into the callable tool library; If not, the tool to be entered will be entered into the callable tool library.

2. The agent tool entry method of claim 1, wherein, The step of determining the function description, applicable scenarios, and calling parameters of the input tool to be applied to the intelligent agent when the input tool is received includes: When a tool to be input is received for use by an intelligent agent, the source code of the tool to be input is parsed to obtain the interface definition, input and output parameters, function method names and code comments of the tool to be input. Based on the interface definition, the input and output parameters, the function method name, and the code comments, determine the function description, applicable scenarios, and calling parameters of the input tool.

3. The agent tool entry method of claim 1, wherein, The step of determining whether the callable tool library of the intelligent agent contains a target tool that conflicts with the tool to be entered, based on the function description, the applicable scenario, and the calling parameters, includes: The functional description, the applicable scenarios, and the calling parameters are semantically encoded to obtain the semantic vector to be entered. The semantic vector to be entered is compared with the stored semantic vectors in the callable tool library of the agent, and the comparison result determines whether the callable tool library contains a target tool that conflicts with the tool to be entered.

4. The agent tool entry method of claim 1, wherein, The step of performing differentiation processing on the tool to be entered and the target tool before entering them into the callable tool library includes: A first prompt word is generated based on the tool to be entered and the target tool, and the first prompt word is input into a first preset large model. The first preset large model is used to refine the semantic vectors of the tool to be entered and the target tool based on the first prompt word. After differentiating the input tool and the target tool based on the first result output by the first preset large model, the input tool is then entered into the callable tool library.

5. The agent tool entry method of claim 1, wherein, The method further includes: In response to the input event of the tool to be inputted, the semantic vector and tool type of the tool to be inputted are determined, and the retrieval index of the callable tool library is updated according to the semantic vector and the tool type; The tool types include basic tools and composite tools. The basic tools are the smallest units that constitute the tools of an intelligent agent, and the composite tools are combinations of multiple basic tools.

6. The intelligent agent tool logging method of claim 5, wherein, After the step of updating the retrieval index of the callable tool library based on the semantic vector and the tool type, the method further includes: A second prompt word is generated based on user intent data and tool call logs of the callable tool library, and the second prompt word is input into a second preset large model. The second preset large model is used to perform tool combination mining on the callable tool library based on the second prompt word. Based on the second result output by the second preset large model, the tool combination scheme of the callable tool library is determined, and the tool combination scheme is verified for inclusion in the library. If the tool combination scheme passes the database entry verification, then the tool combination scheme will be entered into the callable tool library.

7. An agent tool entry device, comprising: The intelligent agent tool input device includes: The information acquisition module is used to determine the function description, applicable scenario and calling parameters of the input tool to be applied to the intelligent agent when it receives the input tool to be applied to the intelligent agent. The condition judgment module is used to determine, based on the function description, the applicable scenario and the calling parameters, whether the callable tool library of the intelligent agent contains a target tool that conflicts with the tool to be entered; The first input module is used to perform differential processing on the tool to be input and the target tool, and then input them into the callable tool library if the conditions are met. The second input module is used to input the tool to be input into the callable tool library if no.

8. An agent tool entry device, comprising: The device includes: a memory, a processor, and an intelligent agent tool input program stored in the memory and executable on the processor, the intelligent agent tool input program being configured to implement the steps of the intelligent agent tool input method as described in any one of claims 1 to 6.

9. A storage medium, characterized by The storage medium is a computer-readable storage medium, and the storage medium stores an intelligent agent tool input program. When the intelligent agent tool input program is executed by a processor, it implements the steps of the intelligent agent tool input method as described in any one of claims 1 to 6.

10. A computer program product, characterised in that, The computer program product includes an intelligent agent tool input program, which, when executed by a processor, implements the steps of the intelligent agent tool input method as described in any one of claims 1 to 6.

Images