Intelligent agent workflow creation method and device, storage medium and electronic equipment

Abstract

The invention discloses an agent workflow creation method and device, a storage medium and electronic equipment. Relates to the field of artificial intelligence. The method comprises the steps that user demand data and a template file are obtained, and the template file comprises node information of an intelligent agent workflow to be generated and a configuration template used for configuring the intelligent agent workflow to be generated; based on the user demand data and the node information, graph structure information is generated, and the graph structure information comprises information of nodes and edges related to the to-be-generated agent workflow; based on the graph structure information and the template file, a workflow configuration template is generated, and the workflow configuration template comprises an agent workflow of which the parameter value is to be determined; and configuring the parameter values in the workflow configuration template to obtain the target agent workflow. According to the method and the device, the technical problem of poor effect of the intelligent agent workflow constructed by adopting a low-code platform in the related technology is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a method and device for creating an agent workflow, a storage medium and an electronic device. BACKGROUND

[0002] An agent workflow is a technical solution for realizing business process automation and optimization with the help of an AI (Artificial Intelligence) agent, and is a new workflow management method. An agent workflow can be based on a large model, and can decompose a complex task into manageable subtasks through the cooperation of multiple AI agents, and complete the target through iterative optimization.

[0003] In related technologies, an agent workflow can be arranged through a low-code platform. For example, a low-code platform provides a visual interface, and does not require code to be written. A human can drag a large model, a tool plug-in, or other nodes to arrange a workflow. However, the low-code platform in related technologies has the following disadvantages.

[0004] (1) Some low-code platforms can access a knowledge base and use tools through a single large model, and can complete a task through a "one-sentence generation autonomous planning agent" function under the command of a prompt word. However, this solution is difficult to deal with complex tasks, resulting in low quality of the generated workflow. In addition, this construction solution also needs to generate high-quality large model prompt words in advance.

[0005] (2) Some low-code platforms have powerful code generation capabilities, and can directly generate a configuration file (for example, a DSL (Domain-Specific Language) file) according to user requirements. However, the quality of the generated configuration file is uncontrollable, and the generated DSL file often has syntax parameter errors that cannot be parsed and executed by the platform. In addition, black box optimization is difficult, and the generation process of the configuration file is long, resource consumption is large, and the cost is high.

[0006] At present, no effective solution has been proposed to solve the above problems. SUMMARY

[0007] Embodiments of the present application provide a method and device for creating an agent workflow, a storage medium and an electronic device, to at least solve the technical problem of poor effect of an agent workflow constructed by a low-code platform in related technologies.

[0008] According to an aspect of an embodiment of the present application, a method for creating an agent workflow is provided, comprising: obtaining user requirement data and a template file, wherein the template file comprises: node information of an agent workflow to be generated, and a configuration template for configuring the agent workflow to be generated; generating graph structure information based on the user requirement data and the node information, wherein the graph structure information comprises: information of nodes and edges involved in the agent workflow to be generated; generating a workflow configuration template based on the graph structure information and the template file, wherein the workflow configuration template comprises: the agent workflow whose parameter values are to be determined; and configuring parameter values in the workflow configuration template to obtain a target agent workflow.

[0009] Further, generating the workflow configuration template based on the graph structure information and the template file comprises: generating an identifier of each node in the graph structure information; generating an identifier of each edge in the graph structure information; and generating the workflow configuration template based on the identifier of each node, the identifier of each edge, and the template file.

[0010] Further, the node information further comprises: a node type, and node configuration information for configuring a node parameter; the configuration template comprises: edge configuration information for configuring an edge; and generating the workflow configuration template based on the identifier of each node, the identifier of each edge, and the template file comprises: for each node, extracting node configuration information of the node from the template file based on a node type of the node, and adding the identifier of the node to the node configuration information of the node to obtain configuration information of the node; for each edge, extracting edge configuration information of the edge from the template file, and adding the identifier of the edge to the edge configuration information to obtain configuration information of the edge; and adding the configuration information of each node and the configuration information of each edge to the configuration template to obtain the workflow configuration template.

[0011] Further, generating the identifier of each node in the graph structure information comprises: generating an offset of each node, and obtaining a current timestamp in the process of generating the identifier of each node to obtain a timestamp of each node; and generating the identifier of each node based on the offset of each node and the timestamp of the node.

[0012] Further, the configuration template is obtained through the following methods: acquiring multiple historical workflow cases, wherein the historical workflow cases include: agent workflows generated within a historical time period; determining an initial workflow template based on the multiple historical workflows; clustering the parameter values ​​in the multiple historical workflow cases to obtain clustering results; determining a first parameter and a second parameter based on the clustering results, wherein the first parameter includes: parameters whose change frequency is higher than a preset frequency threshold, and the second parameter includes: parameters whose change frequency is lower than the preset frequency threshold; adding a label to the first parameter to obtain a labeled first parameter; updating the labeled first parameter and the second parameter to the initial workflow template to obtain the configuration template.

[0013] Further, the parameters in the workflow configuration template are configured to obtain the target agent workflow, including: determining the target prompt word based on the first parameter after marking, the second parameter and the user demand data; inputting the target prompt word into a large language model to obtain the target agent workflow, wherein the large language model is used to update the parameter values ​​of the first parameter after marking and the second parameter.

[0014] Further, after configuring the parameters in the workflow configuration template to obtain the target agent workflow, the process includes: obtaining the data format of the agent workflow supported by the target platform to obtain the target data format; converting the target agent workflow into the target data format to obtain the converted target agent workflow; and generating a meta workflow based on the converted target agent workflow, wherein the meta workflow includes the workflow of the target agent workflow.

[0015] According to another aspect of the present invention, an apparatus for creating an intelligent agent workflow is also provided, comprising: a first acquisition unit, configured to acquire user requirement data and a template file, wherein the template file includes: node information of the intelligent agent workflow to be generated, and a configuration template for configuring the intelligent agent workflow to be generated; a first generation unit, configured to generate graph structure information based on the user requirement data and the node information, wherein the graph structure information includes: information on the nodes and edges involved in the intelligent agent workflow to be generated; a second generation unit, configured to generate a workflow configuration template based on the graph structure information and the template file, wherein the workflow configuration template includes: an intelligent agent workflow with parameter values ​​to be determined; and a configuration unit, configured to configure the parameter values ​​in the workflow configuration template to obtain a target intelligent agent workflow.

[0016] Furthermore, the second generation unit includes: a first generation subunit, used to generate an identifier for each node in the graph structure information; a second generation subunit, used to generate an identifier for each edge in the graph structure information; and a third generation subunit, used to generate the workflow configuration template based on the identifier of each node, the identifier of each edge, and the template file.

[0017] Furthermore, the node information also includes: node type, node configuration information for configuring node parameters, and the configuration template includes: edge configuration information for configuring edges. The third generation subunit includes: a first processing module, used for extracting the node configuration information of each node from the template file based on the node type of the node, and adding the node's identifier to the node configuration information to obtain the configuration information of each node; a second processing module, used for extracting the edge configuration information of each edge from the template file, and adding the identifier of each edge to the edge configuration information to obtain the configuration information of each edge; and an adding module, used for adding the configuration information of each node and the configuration information of each edge to the configuration template to obtain the workflow configuration template.

[0018] Further, the first generation subunit includes: a third processing module, used to generate the offset of each node, and in the process of generating the identifier of each node, obtain the current timestamp to obtain the timestamp of each node; and a first generation module, used to generate the identifier of the node based on the offset of each node and the timestamp of the node.

[0019] Further, the configuration template is obtained through the following sub-units: an acquisition sub-unit, used to acquire multiple historical workflow cases, wherein the historical workflow cases include: intelligent agent workflows generated within a historical time period; a first determination sub-unit, used to determine an initial workflow template based on the multiple historical workflows; a clustering sub-unit, used to cluster the parameter values ​​in the multiple historical workflow cases to obtain clustering results; a second determination sub-unit, used to determine a first parameter and a second parameter based on the clustering results, wherein the first parameter includes: parameters whose change frequency is higher than a preset frequency threshold, and the second parameter includes: parameters whose change frequency is lower than the preset frequency threshold; a marking sub-unit, used to add a mark to the first parameter to obtain a marked first parameter; and an update sub-unit, used to update the marked first parameter and the second parameter to the initial workflow template to obtain the configuration template.

[0020] Furthermore, the configuration unit includes: a determination unit, used to determine a target prompt word based on the first parameter after labeling, the second parameter, and the user demand data; and an input unit, used to input the target prompt word into a large language model to obtain the target agent workflow, wherein the large language model is used to update the parameter values ​​of the first parameter after labeling and the second parameter.

[0021] Furthermore, the device for creating an intelligent agent workflow also includes: a second acquisition unit, configured to acquire the data format of the intelligent agent workflow supported by the target platform after configuring the parameter values ​​in the workflow configuration template to obtain the target intelligent agent workflow, thereby obtaining the target data format; a conversion unit, configured to convert the target intelligent agent workflow into the target data format to obtain the converted target intelligent agent workflow; and a third generation unit, configured to generate a meta-workflow based on the converted target intelligent agent workflow, wherein the meta-workflow includes the workflow of the target intelligent agent workflow.

[0022] According to another aspect of the present invention, an electronic device is also provided, including: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the method for creating an intelligent agent workflow of any of the above-mentioned methods by executing the executable instructions.

[0023] According to another aspect of the present invention, a computer-readable storage medium is also provided, which stores a computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the method for creating an intelligent agent workflow as described above when the computer program is running.

[0024] In this invention, the following method is adopted: user requirement data and a template file are obtained, wherein the template file includes: node information of the intelligent agent workflow to be generated, and a configuration template for configuring the intelligent agent workflow to be generated; based on the user requirement data and node information, graph structure information is generated, wherein the graph structure information includes: information on the nodes and edges involved in the intelligent agent workflow to be generated; based on the graph structure information and the template file, a workflow configuration template is generated, wherein the workflow configuration template includes: an intelligent agent workflow with parameter values ​​to be determined; the parameter values ​​in the workflow configuration template are configured to obtain the target intelligent agent workflow. This solves the technical problem of poor performance of intelligent agent workflows built using low-code platforms in related technologies. In this invention, by generating graph structure information and generating a workflow configuration template based on the template file and graph structure information, and finally optimizing the parameters in the workflow configuration template, an intelligent agent workflow that meets user requirements is obtained. This avoids the situation in related technologies where building intelligent agent workflows using low-code platforms is difficult, resulting in poor quality of the intelligent agent workflow. Therefore, this invention achieves the technical effects of reducing the difficulty of building intelligent agent workflows and improving the quality of intelligent agent workflows. Attached Figure Description

[0025] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0026] Figure 1 This is a flowchart of an optional method for creating an intelligent agent workflow according to an embodiment of the present invention;

[0027] Figure 2 This is a flowchart illustrating the creation of an optional intelligent agent workflow according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of an optional intelligent agent workflow creation apparatus according to an embodiment of the present invention;

[0029] Figure 4 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0030] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0031] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] For ease of description, some terms or nouns involved in the various embodiments of the present invention are explained below:

[0033] A node is a key component of an agent's workflow. By connecting nodes with different functions, a series of operations in the workflow are executed.

[0034] Agent (intelligent agent): refers to a program or system that can perceive the environment, make autonomous decisions and take actions to achieve specific goals. It can be software, robot, or even virtual character. Its core characteristics are autonomy, perception and action capabilities.

[0035] DSL: Domain-Specific Language, refers to a programming or scripting language specifically designed for a particular domain. It can be used to solve a specific type of problem and possesses efficient, concise, and professional expressive capabilities. In this patent, it mainly refers to workflow configurations described in JSON (JavaScript Object Notation, a lightweight data exchange format) or YAML (YAML Ain't Markup Language, a human-readable data serialization format).

[0036] It should be noted that the user information (including but not limited to user device information, user personal information, etc.), the collected information and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of the relevant data all comply with the relevant laws, regulations, and standards of the relevant regions, have taken necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation portals for users to choose to authorize or refuse.

[0037] Example 1

[0038] According to an embodiment of the present invention, an alternative method for creating an intelligent agent workflow is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0039] Figure 1 This is a flowchart of an optional method for creating an intelligent agent workflow according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0040] Step S101: Obtain user requirement data and template file, wherein the template file includes: node information of the intelligent agent workflow to be generated, and configuration template for configuring the intelligent agent workflow to be generated.

[0041] The configuration template mentioned above can include: node information of the agent workflow to be generated. This template is used to configure the agent workflow (hereinafter referred to as workflow). Node information can include: node type, functional description, configurable parameters, etc. Node information can be obtained in the following ways: for example, by using an automated parsing platform's online documentation to complete a summary description of each node's function (main method: build the workflow, input the node introduction document address, retrieve relevant content, input this content along with the node list into the large model, and output a concise functional description of each node through prompts); or by configuring platform cases to extract specific configurable parameters for each node. It should be noted that the configuration nodes in the agent workflow can include, but are not limited to: platform function nodes, tool nodes, logic nodes, large model nodes, etc. Logic nodes include code nodes, IF / ELSE (conditional statement) nodes, template conversion nodes, iteration nodes, etc., facilitating the construction of automated processes.

[0042] The configuration template mentioned above can be the overall configuration template for the intelligent agent workflow, which can be the basic framework for the intelligent agent workflow configuration, including the basic structure of the workflow and general parameter settings. Specifically, it can include attribute settings, configuration information of workflow graph nodes and edges, etc. The specific node and edge information and parameter values ​​in the configuration template can be blank or preset, and can also be filled and optimized according to specific needs.

[0043] The template files mentioned above can be used as a template reference for selecting and arranging workflow nodes and generating the final workflow configuration during the subsequent process of generating intelligent agents.

[0044] In this embodiment, the template file may also include a workflow case knowledge base, which can provide the most relevant example references for generating workflow diagram structures and generating configuration files for agent workflows.

[0045] In one optional example, the workflow case knowledge base construction method includes: collecting a set of workflow cases supported by the platform, using the case function description as a searchable domain of the knowledge base, and writing the content storage domain into the transformed case graph structure and a complete configuration example, respectively.

[0046] In this embodiment, user demand data input by the user can be collected, and a pre-configured template file can be read.

[0047] Step S102: Based on user requirement data and node information, generate graph structure information, which includes: information on nodes and edges involved in the workflow of the agent to be generated.

[0048] In this embodiment, the logical flow of the agent's workflow can be represented in the form of a graph structure. The graph structure can be a directed acyclic graph (DAG), including nodes and edges connecting connected nodes. The effectiveness of the graph structure plays a crucial role in the success or failure of the overall task. Therefore, in this embodiment, the graph structure generation task can be decomposed separately, allowing the large model to focus on understanding user needs and node functions, and to plan and generate the main business logic flow as a whole while temporarily ignoring complex node parameters and workflow configuration parameters.

[0049] In this embodiment, user requirements and node functions can be parsed to generate workflow graph structure information. For example, based on node information, user requirement data can be decomposed into tasks to obtain workflow node and edge information (i.e., graph structure information).

[0050] In one optional example, descriptive information of all nodes mined from the template file (corresponding to node information) can be incorporated into the prompts generated by the large model. Simultaneously, the most relevant cases can be retrieved from the case knowledge base based on user demand data as few-shot examples of the prompts, thus obtaining the aforementioned graph structure information. Specific prompt examples are as follows:

[0051] [Background] Workflows reduce system complexity and improve the performance of LLM applications for complex tasks by breaking down complex tasks into smaller steps (nodes). To address complex business logic in automation and batch processing scenarios, workflows provide a rich set of logical nodes, such as code nodes, IF / ELSE nodes, template transformations, and iteration nodes, facilitating the construction of automated processes. Nodes are a key component of workflows; by connecting nodes with different functions, a series of workflow operations are executed. The specific supported node types and their functional descriptions are: {{#nodes_info (all node description information)#}}.

[0052] [Instruction] As a workflow orchestration expert, you need to understand the user's requirements, break down the tasks, combine and connect them through workflow nodes for execution, and ultimately meet the user's needs.

[0053] [Output] First, output the process of breaking down the workflow into nodes. Finally, output the orchestrated workflow structure information (corresponding to graph structure information) in JSON format, which may include the set of nodes and the set of edges.

[0054] The node set can include all nodes used in the agent's workflow. Some types of nodes may be used multiple times at different locations in the workflow. Please add a suffix "-number" to nodes that are used again. For example, multiple different nodes of type llm are "llm", "llm-2", and "llm-3". The edge set is the set of edges formed by all source nodes and target nodes to be connected in the agent's workflow.

[0055] For example, an example of the generated workflow structure information:

[0056] {{#graph_demostration (example from related case library)#}};

[0057] User requirement description: {{#requirement(user input requirement)#}}.

[0058] The workflow diagram structure information output at this stage is described using a simple case as an example:

[0059] User requirement: "Based on the URL (Uniform Resource Locator) entered by the user, crawl and parse the page content, summarize the relevant key points and display them to the user."

[0060] The output graph structure is represented as: {"nodes":["start", "tool-webscraper", "llm", "answer"], "edges":[["start", "tool-webscraper"], ["webscraper", "llm"], ["llm", "answer"]];

[0061] Here, nodes are the selected workflow nodes, and edges represent the edges that consist of connected source and target nodes.

[0062] Step S103: Based on the graph structure information and template file, generate a workflow configuration template, wherein the workflow configuration template includes: agent workflow with parameter values ​​to be determined.

[0063] In this embodiment, the graph structure information can be filled into the configuration template in the template file to obtain the workflow configuration template. For example, a complete and stable workflow configuration template can be generated by a code tool based on the graph structure information and the configuration template file in the preparation stage.

[0064] Specifically, given the numerous configuration items in a complete agent workflow, to avoid the difficulty of ensuring no configuration items are omitted when directly generating the agent workflow from a large model, a workflow configuration template containing node identifiers (IDs), edge identifiers (IDs), and default parameters can be generated using coding tools based on graph structure information and pre-designed configuration templates. For example, based on graph structure information, each node and edge in the workflow graph structure can be precisely mapped to a unique identifier (ID). The identifiers corresponding to the nodes are then added to the node configuration information in the configuration template, and the identifiers corresponding to the edges are added to the edge configuration information in the configuration template, resulting in the workflow configuration template.

[0065] Step S104: Configure the parameter values ​​in the workflow configuration template to obtain the target intelligent agent workflow.

[0066] In this embodiment, the large model (parameter optimization agent) can also optimize and determine the parameter values ​​in the workflow configuration template based on user requirement data and configuration template, and generate a complete and executable workflow configuration (i.e., target agent workflow).

[0067] Through the above steps, in this embodiment, graph structure information is generated, and a workflow configuration template is generated based on the template file and the graph structure information. Finally, the parameters in the workflow configuration template are optimized to obtain an agent workflow that meets user needs. This avoids the situation in related technologies where building agent workflows using low-code platforms is difficult, resulting in poor quality of the agent workflow. Therefore, it achieves the technical effect of reducing the difficulty of building agent workflows and improving their quality. This solves the technical problem of poor performance of agent workflows built using low-code platforms in related technologies.

[0068] Optionally, based on the graph structure information and template file, a workflow configuration template is generated, including: generating an identifier for each node in the graph structure information; generating an identifier for each edge in the graph structure information; and generating a workflow configuration template based on the identifiers of each node, each edge, and the template file.

[0069] In this embodiment, an identifier for each node can be generated based on timestamps and offsets. An identifier for each edge can be generated based on the identifiers of the source and target nodes associated with each edge. The identifiers of each node are added to the node configuration information in the template file, and the identifiers of each edge are added to the edge configuration information in the template file. The node configuration information and edge configuration information with added identifiers are then added to the configuration template to obtain the workflow configuration template.

[0070] An optional workflow configuration template generation process may include: (1) parsing the graph structure information into a JSON object; (2) traversing all nodes in the JSON and generating a unique numeric ID for each node using timestamps and offsets; extracting the node configuration information corresponding to the node type from the template file and filling in the ID according to the node type; (3) traversing all edge information in the JSON, obtaining the source node and target node type and ID of the edge and generating the edge ID; extracting the edge configuration template from the template file and filling in the source / target node type and ID, edge ID, and other information; (4) extracting the configuration template from the template file, filling in all previously generated node and edge configuration information, and obtaining the final complete workflow configuration template.

[0071] In this embodiment, based on the unique ID corresponding to each node and edge, a stable workflow configuration template is generated using a configuration template. This avoids the situation in related technologies where the intelligent agent workflow is directly generated from a large model, making it difficult to ensure that no configuration items are omitted. This achieves the technical effect of improving the accuracy of intelligent agent workflow configuration.

[0072] Optionally, the node information also includes: node type, node configuration information for configuring node parameters, and the configuration template includes: edge configuration information for configuring edges. Based on the identifier of each node, the identifier of each edge, and the template file, a workflow configuration template is generated, including: for each node, based on the node type, extracting the node configuration information of the node from the template file, and adding the node identifier to the node configuration information to obtain the configuration information of each node; for each edge, extracting the edge configuration information of each edge from the template file, and adding the edge identifier to the edge configuration information to obtain the configuration information of each edge; adding the configuration information of each node and the configuration information of each edge to the configuration template to obtain the workflow configuration template.

[0073] The node information mentioned above may include: node type (such as large model node, tool node, etc.) and node configuration information. The node configuration information includes the node's parameter settings, trigger conditions, etc. Each node can undertake specific functions or tasks in the workflow.

[0074] In this embodiment, each edge can connect two nodes, indicating the direction of data and control flow. The edge configuration information can include the identifiers of the source node and the target node, as well as the edge type or attributes, such as data transmission method, delay rules, etc.

[0075] In this embodiment, the corresponding node configuration information can be found from a preset template file based on the type of each node. This template contains placeholders or default parameters indicating how nodes of this type should be configured. Then, the node identifier (ID) can be added to the node configuration information to form the complete configuration information for each node. For example, if the node type is a Large Model (LLM), the configuration template for an LLM node can be read from the template file, and the node ID "node12345" can be added to the configuration.

[0076] Similarly, for each edge, the edge configuration information template can be found in the template file, and the edge identifier (ID), source node ID, and target node ID can be added to the edge configuration information to form the complete configuration information for each edge.

[0077] The complete configuration information of each node and each edge is integrated into the configuration template to obtain the workflow configuration template. At this point, the workflow configuration template can contain all the structural information and basic parameter settings of the intelligent workflow.

[0078] By separating the structural design and parameter settings of the agent workflow, the technical effect of improving the efficiency and accuracy of agent workflow configuration is achieved.

[0079] Optionally, generating an identifier for each node in the graph structure information includes: generating the offset of each node, and obtaining the current timestamp during the generation of the identifier for each node to obtain the timestamp of each node; generating the identifier of the node based on the offset of each node and the timestamp of the node.

[0080] For each node, before generating the node identifier, an offset can be generated for each node. This offset can be a random number or an increasing sequence. The offset can add extra variation to the timestamp, preventing the situation where the timestamps are the same when multiple nodes are created in a short period of time, and ensuring the uniqueness of the identifier of each node.

[0081] When creating a node identifier, the current timestamp can be obtained to ensure that each generated identifier has time characteristics. Even if the offset is the same, different timestamps can distinguish different nodes.

[0082] For each node, the node's offset can be concatenated with the current timestamp when the node's identifier was generated, and the concatenation result can be converted into a string to obtain the node's identifier.

[0083] Node identifiers are generated based on timestamps and offsets, ensuring that each node identifier is unique. For example, even if multiple nodes are created within the same second, each node's identifier remains unique due to the different offsets.

[0084] Optionally, the configuration template is obtained through the following methods: acquiring multiple historical workflow cases, wherein the historical workflow cases include: intelligent agent workflows generated within a historical time period; determining an initial workflow template based on multiple historical workflows; clustering the parameter values ​​in the multiple historical workflow cases to obtain clustering results; determining a first parameter and a second parameter based on the clustering results, wherein the first parameter includes: parameters whose change frequency is higher than a preset frequency threshold, and the second parameter includes: parameters whose change frequency is lower than a preset frequency threshold; adding a label to the first parameter to obtain the labeled first parameter; updating the labeled first parameter and the second parameter to the initial workflow template to obtain the configuration template.

[0085] In this embodiment, intelligent agent workflow cases generated over a period of time can be collected (including the workflow structure, the types of nodes used, and specific parameter configurations, etc.).

[0086] Based on collected historical workflow cases, a general workflow template can be analyzed and summarized. This template can include the basic structure and framework of the workflow, such as common node types, edge connection methods, and possible configuration items. The initial template serves as the basis for subsequent configuration template optimization, reflecting the most common and fundamental patterns in workflows.

[0087] In this embodiment, cluster analysis can also be performed on the parameter values ​​in historical workflow cases. For example, statistical methods can be used to identify and classify similar parameters. Specifically, cluster analysis can be performed on the parameters in historical workflow cases to determine parameters with high frequency of change and parameters that are relatively fixed or have low frequency of change.

[0088] In one optional example, parameters can be divided into two categories based on clustering results: a first parameter and a second parameter. The first parameter refers to parameters whose frequency of change in historical cases exceeds a preset threshold. This first parameter is usually closely related to specific business scenarios and needs to be customized according to different workflow requirements. The second parameter refers to parameters whose frequency of change in historical cases is lower than a preset threshold. In most cases, the second parameter can use the same default value.

[0089] To quickly identify which parameters require further customization by the user or system when generating new workflow configurations, the first parameter can be marked in the configuration template. This marking can use placeholders (such as...). Workflow Name (This can be a template-based annotation, or a formatted comment that allows for quick identification of parameter values ​​that need updating or optimization when a new workflow is generated based on the template.)

[0090] In this embodiment, the tagged first parameter and the untagged second parameter can be updated together in the initial workflow template to generate an optimized configuration template. Using the configuration template, a workflow configuration template can be quickly generated based on the user's new workflow requirement. Subsequently, only the tagged first parameter needs to be adjusted, without having to configure the entire workflow from scratch, saving time and effort and improving efficiency.

[0091] In one optional example, the configuration template can be obtained as follows: Extract the overall workflow template from platform case configurations. For instance, for numerous configuration parameters, automatically cluster the parameter values ​​from multiple cases to identify frequently changing parameters and relatively fixed default parameters. Frequently changing parameters are generally highly relevant to business logic; parameter descriptions and guiding tags (such as...) can be added when building the configuration template. Workflow Name This allows for easy updates and optimizations of large models to meet business needs; default parameters are directly retained in the template.

[0092] By setting configuration templates, it is possible to adapt to different types of agent workflow requirements, while ensuring the stability and accuracy of agent workflow configuration and reducing the probability of errors in the generation process.

[0093] Optionally, the parameters in the workflow configuration template are configured to obtain the target agent workflow, including: determining the target prompt word based on the labeled first parameter, second parameter and user demand data; inputting the target prompt word into the large language model to obtain the target agent workflow, wherein the large language model is used to update the parameter values ​​of the labeled first parameter and second parameter.

[0094] The aforementioned target prompts may include instructions designed for the large language model to guide it on how to update and configure parameters in the workflow template. In this embodiment, these prompts can be based on the tagged first and second parameters and user requirements. For example, user requirement data can be integrated into the prompts so that the large model can accurately understand the context and objectives of the task. Furthermore, the large language model can be instructed to adjust the first and second parameters based on the user requirement data.

[0095] Large language models can optimize the labeled first and second parameters and update parameter values ​​based on an understanding of user needs, ensuring that the agent's workflow can effectively solve problems or perform tasks. For the first parameter, large language models can provide personalized and precise configuration; for the second parameter, they can provide reasonable default values ​​or minor adjustments based on the model's judgment or analysis of historical data.

[0096] The output of the large language model can be optimized into a workflow configuration file with optimized parameter values, and can be run directly on the intelligent agent platform to meet the user's requirements.

[0097] In one optional example, the parameters in the workflow configuration template are configured to obtain the target agent workflow, which includes: a large model (acting as a parameter optimization agent) updates and optimizes the parameter values ​​of configuration items based on the generated complete workflow configuration template and user requirement data. For example, for parameter values ​​that are highly relevant to the business (i.e., double-valued parameters in the configuration template)... The marked items should be emphasized in the prompt text as needing to be updated or replaced in a timely manner; other parameters can be updated as needed.

[0098] The specific prompts are as follows:

[0099] As a workflow orchestration expert, you need to understand the user's requirements, break down tasks, combine and connect them through workflow nodes for execution, and ultimately meet the user's needs.

[0100] The currently supported nodes and tools are described below:

[0101] {{#nodes_info (description information for all nodes)#}};

[0102] The initial workflow orchestration has been completed, and the corresponding configuration items are as follows:

[0103] {{#config_template (workflow configuration template)#}};

[0104] Based on specific requirements and the descriptions of the configuration file parameters, complete the filling and optimization of relevant configuration parameters in the workflow. This includes two pairs... The content between the symbols describes the parameter item (e.g., ...). Workflow Name (These parameters are closely related to business operations and need to be updated or replaced according to specific business needs. Other parameters should be updated and optimized as needed.)

[0105] User requirements description and corresponding complete workflow configuration example:

[0106] {{#workflow_demostration (related case library examples)#}};

[0107] User requirement description: {{#requirement(user input requirement)#}};

[0108] In this embodiment, automated parameter configuration via a large model significantly reduces manual intervention and accelerates the creation of the intelligent agent workflow. Updating parameters based on a deep understanding of user needs is more accurate than manual configuration and avoids configuration errors.

[0109] Optionally, after configuring the parameters in the workflow configuration template to obtain the target agent workflow, the process includes: obtaining the data format of the agent workflow supported by the target platform to obtain the target data format; converting the target agent workflow into the target data format to obtain the converted target agent workflow; and generating a meta workflow based on the converted target agent workflow, wherein the meta workflow includes the workflow of the target agent workflow.

[0110] Since different platforms have different data exchange and storage formats, such as JSON and YAML, before deploying the workflow configuration file to a specific platform, it is necessary to first determine the intelligent agent workflow data format supported by the platform and then convert the target intelligent agent workflow configuration file to the correct format. For example, if the target platform supports YAML format and the current workflow configuration file is in JSON format, a conversion tool or function can be used to convert JSON to YAML. In one optional example, the conversion process may include: (1) Parsing: Parsing the current JSON format configuration file into an operable data structure. (2) Reconstruction: Reconstructing the structure and syntax of the configuration file according to the data format specifications of the target platform. (3) Generation: Outputting the reconstructed data structure in a data format supported by the target platform, such as a YAML string or file.

[0111] The aforementioned meta-workflow refers to a workflow used to generate, manage, and optimize agent workflows. After format conversion, a meta-workflow can be constructed based on the converted target agent workflow. This meta-workflow contains all information about the target workflow, including its structure, configuration, and execution logic. The generation of meta-workflows is part of an automated process and can further improve the efficiency of workflow creation and management.

[0112] The components of a meta-workflow may include: (1) Workflow generation: describing how to start from user requirements, generate configuration templates, update parameters, convert data formats, and finally create an executable target agent workflow; (2) Workflow management: functions such as starting, stopping, and monitoring the status of the agent workflow; (3) Workflow optimization: by collecting feedback on the execution of the target agent workflow, automatically adjusting and optimizing workflow parameters or structure to improve execution efficiency or effectiveness.

[0113] Currently, mainstream platforms support workflow configurations in YAML files, while in this embodiment, the workflow configuration generation process mainly uses JSON format for data transmission (because JSON format has simple syntax and clear structure). Therefore, in this embodiment, the JSON format of the target agent's workflow configuration can be automatically converted into a YAML file supported by the platform using code tools.

[0114] Ultimately, the entire process can be fully automated by building a meta-workflow (a workflow that generates workflows). That is, when the user's requirement description is input, the meta-workflow automatically generates the corresponding business workflow YAML configuration file.

[0115] In an alternative example, the meta-workflow can be further extended. For instance, after the target agent workflow is generated, the execution of the generated workflow can be run and monitored. Based on the feedback during the workflow's execution, the generation of the workflow can be automatically adjusted and optimized. After several iterations, this helps users generate more efficient workflows.

[0116] Figure 2 This is a flowchart illustrating the creation of an optional intelligent agent workflow according to an embodiment of the present invention, such as... Figure 2 As shown, the creation of the intelligent agent workflow includes three stages: graph structure generation, workflow configuration template generation, and complete workflow configuration generation. Specifically, (1) First, by understanding user requirements and node functions, the graph structure information of the workflow is generated. (2) Second, based on the graph structure and the pre-designed configuration template, a workflow configuration template containing node IDs, edge IDs, and default parameters is generated using code tools. (3) Finally, through a large model (parameter optimization agent), the parameter values ​​are optimized and determined according to user requirements and the configuration template to generate a complete and executable workflow configuration (corresponding to the target intelligent agent workflow).

[0117] In this embodiment, for the complex task of generating workflow configurations, it can be broken down into a stable and controllable phased process (first generating the graph structure, then assembling the complex configuration template, and finally filling in and optimizing the configuration parameters), which improves the success rate of workflow generation. Through the configuration template parameter optimization mechanism, it not only ensures the optimization of constantly changing parameters that fit the business, but also ensures the stability of the better default parameters, thereby improving the overall effect of the workflow. It can also find relevant cases in the case library through semantic relevance and integrate them into the prompts in different forms at different stages to improve the effectiveness of example learning. Furthermore, it can automatically adjust and optimize the workflow (including graph structure and configuration parameter value optimization) based on the running feedback of the generated workflow, and generate a more efficient workflow through several rounds of iterative optimization.

[0118] Example 2

[0119] Embodiment 2 of the present invention provides an optional intelligent agent workflow creation device, wherein each embodiment of the creation device corresponds to each implementation step in Embodiment 1.

[0120] Figure 3 This is a schematic diagram of an optional intelligent agent workflow creation apparatus according to an embodiment of the present invention, such as... Figure 3 As shown, it includes: a first acquisition unit 31, a first generation unit 32, a second generation unit 33, and a configuration unit 34.

[0121] The first acquisition unit 31 is used to acquire user requirement data and template files. The template files include: node information of the intelligent agent workflow to be generated, and configuration templates for configuring the intelligent agent workflow to be generated.

[0122] The first generation unit 32 is used to generate graph structure information based on user demand data and node information, wherein the graph structure information includes: information on the nodes and edges involved in the workflow of the intelligent agent to be generated;

[0123] The second generation unit 33 is used to generate a workflow configuration template based on graph structure information and template file, wherein the workflow configuration template includes: an agent workflow with parameter values ​​to be determined;

[0124] Configuration unit 34 is used to configure the parameter values ​​in the workflow configuration template to obtain the target intelligent agent workflow.

[0125] In the intelligent agent workflow creation device provided in this embodiment, user requirement data and template files can be acquired through the first acquisition unit 31. The template file includes node information of the intelligent agent workflow to be generated and a configuration template for configuring the intelligent agent workflow. Based on the user requirement data and node information, the first generation unit 32 generates graph structure information, which includes information on the nodes and edges involved in the intelligent agent workflow to be generated. Based on the graph structure information and the template file, the second generation unit 33 generates a workflow configuration template, which includes an intelligent agent workflow with parameter values ​​to be determined. The configuration unit 34 configures the parameter values ​​in the workflow configuration template to obtain the target intelligent agent workflow. This solves the technical problem of poor performance of intelligent agent workflows built using low-code platforms in related technologies. In this embodiment, by generating graph structure information and generating a workflow configuration template based on the template file and graph structure information, and finally optimizing the parameters in the workflow configuration template, an intelligent agent workflow that meets user needs is obtained. This avoids the situation in related technologies where it is difficult to build intelligent agent workflows using low-code platforms, resulting in poor quality of intelligent agent workflows. Thus, the technical effect of reducing the difficulty of building intelligent agent workflows and improving the quality of intelligent agent workflows is achieved.

[0126] Optionally, in the intelligent agent workflow creation device provided in this embodiment, the second generation unit includes: a first generation subunit, used to generate the identifier of each node in the graph structure information; a second generation subunit, used to generate the identifier of each edge in the graph structure information; and a third generation subunit, used to generate a workflow configuration template based on the identifier of each node, the identifier of each edge, and the template file.

[0127] Optionally, in the intelligent agent workflow creation device provided in this embodiment, the node information further includes: node type, node configuration information for configuring node parameters, and the configuration template includes: edge configuration information for configuring edges. The third generation subunit includes: a first processing module, used for extracting the node configuration information of each node from the template file based on the node type of the node, and adding the node's identifier to the node configuration information to obtain the configuration information of each node; a second processing module, used for extracting the edge configuration information of each edge from the template file, and adding the identifier of each edge to the edge configuration information to obtain the configuration information of each edge; and an adding module, used for adding the configuration information of each node and the configuration information of each edge to the configuration template to obtain the workflow configuration template.

[0128] Optionally, in the intelligent agent workflow creation device provided in this embodiment, the first generation subunit includes: a third processing module, used to generate the offset of each node, and obtain the current timestamp during the generation of the identifier of each node to obtain the timestamp of each node; and a first generation module, used to generate the identifier of the node based on the offset of each node and the timestamp of the node.

[0129] Optionally, in the intelligent agent workflow creation device provided in this embodiment, the configuration template is obtained through the following sub-units: an acquisition sub-unit, used to acquire multiple historical workflow cases, wherein the historical workflow cases include: intelligent agent workflows generated within a historical time period; a first determination sub-unit, used to determine an initial workflow template based on multiple historical workflows; a clustering sub-unit, used to cluster the parameter values ​​in multiple historical workflow cases to obtain clustering results; a second determination sub-unit, used to determine a first parameter and a second parameter based on the clustering results, wherein the first parameter includes: parameters whose change frequency is higher than a preset frequency threshold, and the second parameter includes: parameters whose change frequency is lower than a preset frequency threshold; a marking sub-unit, used to add a mark to the first parameter to obtain the marked first parameter; and an update sub-unit, used to update the marked first parameter and the second parameter to the initial workflow template to obtain the configuration template.

[0130] Optionally, in the intelligent agent workflow creation device provided in this embodiment, the configuration unit includes: a determination unit, used to determine the target prompt word based on the labeled first parameter, the second parameter and user requirement data; and an input unit, used to input the target prompt word into a large language model to obtain the target intelligent agent workflow, wherein the large language model is used to update the parameter values ​​of the labeled first parameter and the second parameter.

[0131] Optionally, in the intelligent agent workflow creation device provided in this embodiment, the intelligent agent workflow creation device further includes: a second acquisition unit, used to acquire the data format of the intelligent agent workflow supported by the target platform after configuring the parameter values ​​in the workflow configuration template to obtain the target intelligent agent workflow, and obtain the target data format; a conversion unit, used to convert the target intelligent agent workflow into the target data format, and obtain the converted target intelligent agent workflow; and a third generation unit, used to generate a meta-workflow based on the converted target intelligent agent workflow, wherein the meta-workflow includes: the workflow of the target intelligent agent workflow.

[0132] The aforementioned intelligent agent workflow creation device may further include a processor and a memory. The first acquisition unit 31, the first generation unit 32, the second generation unit 33, and the configuration unit 34 are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0133] The aforementioned processor contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured. By adjusting kernel parameters, graph structure information is generated, and a workflow configuration template is generated based on the template file and the graph structure information. Finally, the parameters in the workflow configuration template are optimized to obtain an agent workflow that meets user requirements. This avoids the high difficulty of building agent workflows using low-code platforms in related technologies, which often results in poor quality of agent workflows. Therefore, it achieves the technical effect of reducing the difficulty of building agent workflows and improving their quality.

[0134] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0135] According to another aspect of the present invention, an electronic device is also provided, including: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the method for creating an intelligent agent workflow of any of the above-mentioned methods by executing the executable instructions.

[0136] According to another aspect of the present invention, a computer-readable storage medium is also provided, which stores a computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the method for creating an intelligent agent workflow as described above when the computer program is running.

[0137] Figure 4 This is a schematic diagram of an electronic device according to an embodiment of the present invention, such as... Figure 4As shown, an embodiment of the present invention provides an electronic device 40, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the method for creating an intelligent agent workflow as described above.

[0138] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0139] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0140] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0141] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0142] Furthermore, the functional units in the various embodiments of the present invention 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.

[0143] 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, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0144] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for creating an intelligent agent workflow, characterized in that, include: Obtain user requirement data and template files, wherein the template files include: node information of the intelligent agent workflow to be generated, and configuration templates for configuring the intelligent agent workflow to be generated; Based on the user demand data and the node information, graph structure information is generated, wherein the graph structure information includes: information on the nodes and edges involved in the intelligent agent workflow to be generated; Based on the graph structure information and the template file, a workflow configuration template is generated, wherein the workflow configuration template includes: an agent workflow with parameter values ​​to be determined; Configure the parameter values ​​in the workflow configuration template to obtain the target intelligent agent workflow.

2. The creation method according to claim 1, characterized in that, Based on the graph structure information and the template file, a workflow configuration template is generated, including: Generate an identifier for each node in the graph structure information; Generate an identifier for each edge in the graph structure information; The workflow configuration template is generated based on the identifier of each node, the identifier of each edge, and the template file.

3. The creation method according to claim 2, characterized in that, The node information also includes: node type, node configuration information for configuring node parameters, and the configuration template includes: edge configuration information for configuring edges. Based on the identifier of each node, the identifier of each edge, and the template file, the workflow configuration template is generated, including: For each node, based on the node type, the node configuration information of the node is extracted from the template file, and the node's identifier is added to the node configuration information to obtain the configuration information of each node; For each edge, extract the edge configuration information of each edge from the template file, and add the identifier of each edge to the edge configuration information to obtain the configuration information of each edge; The configuration information of each node and the configuration information of each edge are added to the configuration template to obtain the workflow configuration template.

4. The creation method according to claim 2, characterized in that, Generate the identifier for each node in the graph structure information, including: Generate the offset of each node, and obtain the current timestamp during the generation of the identifier of each node to obtain the timestamp of each node; An identifier for each node is generated based on its offset and timestamp.

5. The creation method according to claim 1, characterized in that, The configuration template is obtained in the following way: Obtain multiple historical workflow cases, wherein the historical workflow cases include: agent workflows generated within a historical time period; Based on multiple historical workflows, an initial workflow template is determined; Cluster the parameter values ​​from multiple historical workflow cases to obtain clustering results; Based on the clustering results, a first parameter and a second parameter are determined, wherein the first parameter includes a parameter whose change frequency is higher than a preset frequency threshold, and the second parameter includes a parameter whose change frequency is lower than the preset frequency threshold. Add a label to the first parameter to obtain the labeled first parameter; The first and second parameters after the marking are updated in the initial workflow template to obtain the configuration template.

6. The creation method according to claim 5, characterized in that, Configure the parameter values ​​in the workflow configuration template to obtain the target agent workflow, including: Based on the first parameter after the tagging, the second parameter, and the user demand data, the target prompt words are determined; The target prompt word is input into a large language model to obtain the target agent workflow, wherein the large language model is used to update the parameter values ​​of the first and second parameters after the labeling.

7. The creation method according to claim 3, characterized in that, After configuring the parameter values ​​in the workflow configuration template to obtain the target agent workflow, the process includes: Obtain the data format of the intelligent agent workflow supported by the target platform to obtain the target data format; The target agent workflow is converted into the target data format to obtain the converted target agent workflow. Based on the transformed target agent workflow, a meta-workflow is generated, wherein the meta-workflow includes the workflow of the target agent workflow.

8. A device for creating an intelligent agent workflow, characterized in that, include: The first acquisition unit is used to acquire user requirement data and template files, wherein the template files include: node information of the intelligent agent workflow to be generated, and a configuration template for configuring the intelligent agent workflow to be generated; The first generation unit is used to generate graph structure information based on the user demand data and the node information, wherein the graph structure information includes: information on the nodes and edges involved in the intelligent agent workflow to be generated; The second generation unit is used to generate a workflow configuration template based on the graph structure information and the template file, wherein the workflow configuration template includes: an agent workflow with parameter values ​​to be determined; The configuration unit is used to configure the parameter values ​​in the workflow configuration template to obtain the target intelligent agent workflow.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the method for creating an intelligent agent workflow as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method for creating an intelligent agent workflow as described in any one of claims 1 to 7.

Images