Method and system for supporting use of agent graph

Abstract

According to one aspect of the present invention, provided is a method for supporting the use of an agent graph, the method comprising the steps of: processing a task performance request of a user and deriving a plurality of unit functions necessary for performing the task; and extracting a workflow for performing at least a portion of the plurality of unit functions from an agent graph, wherein the workflow is composed of a plurality of agents for performing each unit function included among the plurality of unit functions.

Description

Method and system for supporting the use of agent graphs

[0001] The present invention relates to a method and system for supporting the use of an agent graph.

[0002] Recently, with the rapid advancement of artificial intelligence technology, massive AI models such as ChatGPT are emerging. However, the reality is that no matter how large an AI model becomes, it can only perform well for specific tasks and has limitations in performing well for all types of tasks.

[0003] For this reason, there is a growing need to build Multi-Agents Systems (MAS) that enable various types of artificial intelligence models, algorithms, and tools to be utilized within a single workflow, and recently, frameworks supporting the convenient implementation of such systems have been introduced.

[0004] However, if multiple entities build their own MAS, it can lead to inefficiencies in terms of time and cost, such as the redundant development of agents (or workflows) performing the same functions or the development of multiple versions with differing quality; therefore, there is a need for a system that allows multiple entities to jointly use agents (or workflows).

[0005] For example, if a specific entity (individual, company, etc.) implements a CRM (Customer Relationship Management) chatbot agent using its own knowledge or database, it is possible to prevent unnecessary waste of resources by allowing other entities to utilize the agent already implemented by the aforementioned entity when they need a chatbot with the same functionality, without having to develop it again. In this case, other entities utilizing the already implemented agent may also customize the agent with their own data (e.g., enhancing the agent as described later). Furthermore, by allowing other entities to use this customized agent as needed, it may be possible to enable multiple users to jointly develop the agent.

[0006] Accordingly, the inventor(s) propose a technology that supports multiple users in jointly using an Agent Graph by processing a user's request to perform a task, deriving a plurality of unit functions necessary for performing the task, and extracting a workflow from the Agent Graph that performs at least a portion of the plurality of unit functions.

[0007] <Prior Art Literature>

[0008] <Patent Literature>

[0009] (Patent Document 1) Korean Published Patent Application No. 10-2020-0048201 (May 8, 2020)

[0010] The present invention aims to solve all the problems of the aforementioned prior art.

[0011] In addition, the present invention has another objective of processing a user's request to perform a task to derive a plurality of unit functions necessary for performing the task, and extracting a workflow from an agent graph that performs at least a portion of the plurality of unit functions, wherein the workflow is composed of a plurality of agents that perform each unit function included in the plurality of unit functions.

[0012] In addition, another objective of the present invention is to support multiple users in jointly using an agent graph.

[0013] A representative configuration of the present invention for achieving the above objective is as follows.

[0014] According to one aspect of the present invention, a method is provided comprising the steps of: processing a user's request to perform a task to derive a plurality of unit functions necessary for performing the task; and extracting a workflow from an agent graph that performs at least some of the plurality of unit functions, wherein the workflow is composed of a plurality of agents that perform each unit function included in the plurality of unit functions.

[0015] According to another aspect of the present invention, a system is provided comprising a request processing unit that processes a user's request to perform a task and derives a plurality of unit functions necessary for the performance of said task, and a workflow extraction unit that extracts a workflow from an agent graph that performs at least some of said unit functions, wherein the workflow is composed of a plurality of agents that perform each unit function included in said plurality of unit functions.

[0016] In addition to this, other methods for implementing the present invention, other systems, and non-transient computer-readable recording media for recording a computer program for executing said methods are further provided.

[0017] According to the present invention, a user's request to perform a task is processed to derive a plurality of unit functions necessary for performing the task, and a workflow that performs at least some of the plurality of unit functions can be extracted from an agent graph.

[0018] In addition, according to the present invention, it is possible to support multiple users in jointly using an agent graph.

[0019] FIG. 1 is a diagram showing the schematic configuration of an overall system for supporting the use of an agent graph according to one embodiment of the present invention.

[0020] FIG. 2 is a diagram illustrating in detail the internal configuration of an agent graph management system according to one embodiment of the present invention.

[0021] FIGS. 3, FIGS. 5 and FIGS. 6 are drawings illustrating an agent graph according to an embodiment of the present invention.

[0022] FIG. 4 is a diagram showing a query statement in pseudocode according to an embodiment of the present invention.

[0023] <Explanation of Symbols>

[0024] 100: Communication network

[0025] 200: Agent Graph Management System

[0026] 210: Request Processing Unit

[0027] 220: Workflow Extraction Section

[0028] 230: Graph Management Department

[0029] 240: Communications Department

[0030] 250: Control unit

[0031] 300: Device

[0032] The following detailed description of the invention refers to the accompanying drawings, which illustrate specific embodiments in which the invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that various embodiments of the invention are different but need not be mutually exclusive. For example, specific shapes, structures, and characteristics described herein may be modified from one embodiment to another without departing from the spirit and scope of the invention. It should also be understood that the location or arrangement of individual components within each embodiment may be modified without departing from the spirit and scope of the invention. Accordingly, the following detailed description is not meant to be limiting, and the scope of the invention should be understood to encompass the scope claimed by the claims and all equivalents thereof. Similar reference numerals in the drawings indicate identical or similar components across various aspects.

[0033] Hereinafter, in order to enable a person skilled in the art to easily practice the present invention, various preferred embodiments of the present invention will be described in detail with reference to the attached drawings.

[0034] Configuration of the entire system

[0035] FIG. 1 is a diagram showing the schematic configuration of an overall system for supporting the use of an agent graph according to one embodiment of the present invention.

[0036] As illustrated in FIG. 1, the entire system according to one embodiment of the present invention may include a communication network (100), an agent graph management system (200), and a device (300).

[0037] First, a communication network (100) according to one embodiment of the present invention can be configured regardless of the mode of communication, such as wired communication or wireless communication, and can be configured as various communication networks such as a Local Area Network (LAN), a Metropolitan Area Network (MAN), or a Wide Area Network (WAN). Preferably, the communication network (100) referred to in this specification may be the known Internet or the World Wide Web (WWW). However, the communication network (100) may include at least a known wired / wireless data communication network, a known telephone network, or a known wired / wireless television communication network, without being limited thereto.

[0038] For example, the communication network (100) may be a wireless data communication network and may implement conventional communication methods such as WiFi communication, WiFi-Direct communication, Long Term Evolution (LTE) communication, 5G communication, Bluetooth communication (including Bluetooth Low Energy (BLE) communication), infrared communication, ultrasonic communication, etc., in at least a part thereof. As another example, the communication network (100) may be an optical communication network and may implement conventional communication methods such as Light Fidelity (LiFi), etc., in at least a part thereof.

[0039] Next, an agent graph management system (200) according to one embodiment of the present invention processes a user's request to perform a task, derives a plurality of unit functions required for the performance of the task, and extracts a workflow from the agent graph that performs at least some of the plurality of unit functions, wherein the workflow can perform a function composed of a plurality of agents that perform each unit function included in the plurality of unit functions.

[0040] The configuration and functions of the agent graph management system (200) according to the present invention will be examined in detail through the following detailed description.

[0041] Next, the device (300) according to one embodiment of the present invention is a digital device that includes a function to communicate after connecting to an agent graph management system (200), and any digital device equipped with memory means and equipped with a microprocessor to have computational capabilities, such as a smartphone, tablet, smart watch, smart band, smart glasses, desktop computer, laptop computer, workstation, PDA, web pad, mobile phone, etc., can be adopted as the device (300) according to the present invention.

[0042] In particular, the device (300) may include an application (not shown) that enables a user to receive a service according to the present invention from the agent graph management system (200). Such an application may be downloaded from the agent graph management system (200) or an external application distribution server (not shown). Meanwhile, the nature of such an application may generally be similar to the request processing unit (210), workflow extraction unit (220), graph management unit (230), communication unit (240), and control unit (250) of the agent graph management system (200) as described below. Here, at least a part of the application may be replaced with a hardware device or firmware device capable of performing substantially the same or equivalent functions as needed.

[0043] Configuration of the Agent Graph Management System

[0044] Below, we will examine the internal configuration of the agent graph management system (200) that performs important functions for the implementation of the present invention and the functions of each component.

[0045] FIG. 2 is a drawing illustrating in detail the internal configuration of an agent graph management system (200) according to one embodiment of the present invention.

[0046] As illustrated in FIG. 2, an agent graph management system (200) according to one embodiment of the present invention may be configured to include a request processing unit (210), a workflow extraction unit (220), a communication unit (240), and a control unit (250), and may further include a graph management unit (230) as needed. According to one embodiment of the present invention, the request processing unit (210), the workflow extraction unit (220), the graph management unit (230), the communication unit (240), and the control unit (250) may be program modules in which at least some of them communicate with an external system (not shown). Such program modules may be included in the agent graph management system (200) in the form of an operating system, an application program module, or other program modules, and may be physically stored in various known storage devices. Additionally, such program modules may be stored in a remote storage device capable of communicating with the agent graph management system (200). Meanwhile, such program modules include, but are not limited to, routines, subroutines, programs, objects, components, data structures, etc. that perform specific tasks or execute specific abstract data types as described below according to the present invention.

[0047] Meanwhile, although the agent graph management system (200) has been described as above, this description is exemplary, and it is obvious to those skilled in the art that at least some of the components or functions of the agent graph management system (200) may be realized within a device (300) or server (not shown) or included within an external system (not shown) as needed.

[0048] First, a request processing unit (210) according to one embodiment of the present invention can perform the function of deriving a plurality of unit functions necessary for performing the task by processing a user's request to perform the task.

[0049] Specifically, a request processing unit (210) according to one embodiment of the present invention can receive a request to perform a task input by a user via a web or app using a device (300). According to one embodiment of the present invention, such a request to perform a task can be input in a natural language format, such as voice or text.

[0050] A request processing unit (210) according to one embodiment of the present invention can perform the function of deriving a plurality of unit functions necessary for the execution of the task by processing a task execution request received in this manner. Here, a unit function may refer to each of the partial tasks when multiple partial tasks are required to perform the task requested by the user. According to one embodiment of the present invention, these partial tasks may be performed sequentially and / or in parallel.

[0051] For example, when a user inputs a request to perform a task such as, "I want to list the competitor detail pages of a specific product, extract the text, and analyze the marketing phrases," the request processing unit (210) according to one embodiment of the present invention processes the above request to perform the task and can derive the following functions as a plurality of unit functions necessary for performing the task:

[0052] 1. Extraction of Product Features

[0053] 2. Search for similar products or competitors

[0054] 3. Collection of Detail Page Information

[0055] 4. OCR

[0056] 5. Analysis of Marketing Copy

[0057] A request processing unit (210) according to one embodiment of the present invention may use a fine-tuned Large Language Model (LLM) that processes natural language input by a user (i.e., a request to perform a task) to derive a plurality of unit functions as described above, analyzes the user's intent (e.g., research on similar products or competitors, extraction of detailed page text and analysis of marketing phrases), and divides it into atomic actions.

[0058] However, the method of receiving a user's request to perform a task or deriving a unit function according to one embodiment of the present invention is not limited to the contents described above, and may be varied within the scope of achieving the purpose of the present invention.

[0059] Next, the workflow extraction unit (220) according to one embodiment of the present invention may perform the function of extracting from the agent graph a workflow that performs at least some of the plurality of unit functions derived by the request processing unit (210) as described above. According to one embodiment of the present invention, the workflow may be composed of a plurality of agents that perform each unit function included in the plurality of unit functions. Here, it should be understood that the workflow does not necessarily need to be composed of a plurality of agents that perform all of the plurality of unit functions. That is, according to one embodiment of the present invention, the workflow may be composed of a plurality of agents that perform some of the unit functions among the plurality of unit functions derived by the request processing unit (210).

[0060] FIG. 3 is a simplified illustration of an agent graph to aid in understanding the present invention. An actual agent graph may contain many more nodes and edges, similar to a large ontology graph. Hereinafter, nodes, edges, workflows, and agent graphs will be described in detail with reference to FIG. 3. However, it should be noted in advance that the description below corresponds only to one embodiment of the present invention, and therefore the meaning of nodes, edges, workflows, and / or agent graphs is not limited by the description below.

[0061] As shown in FIG. 3, the agent graph may be composed of a plurality of nodes (illustrated as a circle, such as N1, N2, N3, N4, N5, etc.) and a plurality of edges (indicated as lines connecting two nodes, such as E12, E23, E34, E45, etc.) that connect the nodes according to the relationship between the nodes.

[0062] According to one embodiment of the present invention, the types of nodes may include (i) agent nodes and (ii) branch nodes, and, if necessary, (iii) workflow nodes. Of course, the types of nodes are not limited to (i) through (iii) above.

[0063] First, according to one embodiment of the present invention, (i) an agent node may refer to an artificial intelligence (or machine learning) model, API, tool, LLM agent, OCR module, etc., that performs a specific function within a workflow. Such an agent node may have the following attributes:

[0064] - name: Agent name

[0065] - capabilities: Key functions provided by the agent (e.g., document summarization, OCR, etc.)

[0066] - I / O type: Input / output data format processed by the agent

[0067] - Agent-related metrics such as cost, latency, and accuracy

[0068] - Metadata such as version, apiEndpoint, securityLevel, etc.

[0069] - Docker-related information such as dockerImage, dockerTag, registry, etc.

[0070] Next, according to one embodiment of the present invention, (ii) a branch node may mean a node that connects a specific node within a workflow to two or more other nodes according to a defined condition. Such a branch node may have the following attributes:

[0071] - conditionalExpression: If conditional expression (e.g., if inputLang == 'ko' then → Agent B else → Agent C)

[0072] - branches: Node information for each branch result (e.g., 'ko' → Agent B, 'en' → Agent C, etc.)

[0073] Next, (iii) a workflow node may refer to a workflow to be described below, and according to one embodiment of the present invention, the workflow itself may be defined as a node as necessary. Therefore, it should be understood that, above and below, a node may refer to a workflow node depending on the context.

[0074] According to one embodiment of the present invention, an edge connects two nodes and generally has directionality (unidirectional or bidirectional), but in some cases, a non-directional edge may also exist. Such an edge may include information regarding the format of data (text, image, video, etc.) exchanged between the two nodes, and taking the case of a directional edge as an example, if the output data of the node in the direction from which the edge emerges matches the input data of the node in the direction in which the edge enters, the two nodes may be connected by an edge. According to one embodiment of the present invention, the following types of edges may exist, but are not limited thereto:

[0075] (i) (Agent) - [:NEXT] → (Agent): Agent B consumes the result of Agent A

[0076] (ii) (Agent) - [:CONDITION] → (Branch): Connects the agent's processing result to the next stage branch node

[0077] (iii) (Agent) - [:HAS_Agent] → (Agent): Indicates that the agent belongs to a specific workflow

[0078] (iv) (Agent) - [:USES_DB] → (DB Node): Refers to the external DB referenced by the agent

[0079] (v) (Agent) - [:USES_TOOL] → (TOOL Node): Refers to the external TOOL referenced by the agent

[0080] According to one embodiment of the present invention, an agent graph may mean a graph including all nodes and all edges, and a workflow may mean a graph composed of some nodes and some edges, that is, a subgraph that can be extracted from the agent graph.

[0081] Referring to FIG. 3, the entire graph including all nodes and all edges corresponds to the agent graph, and a partial graph (WF1) consisting of some nodes (N1, N2, N3, N4 and N5) and some edges (E12, E23, E34 and E45) may correspond to the workflow.

[0082] A workflow extraction unit (220) according to one embodiment of the present invention can generate a query statement for searching for a workflow in an agent graph based on a plurality of unit functions derived by the request processing unit (210) as described above, and can extract a workflow from an agent graph using the query statement.

[0083] In the example described above, a request processing unit (210) according to one embodiment of the present invention processes a user's request to perform a task, such as "I want to list competitor detail pages of a specific product, extract text, and analyze marketing phrases," and can derive a plurality of unit functions, such as "1. extracting product features, 2. searching for similar products or competitors, 3. collecting detail page information, 4. OCR, and 5. analyzing marketing phrases." FIG. 4 is a diagram showing a query statement that can be generated by the workflow extraction unit (220) in this case, expressed in pseudo code.

[0084] As illustrated in FIG. 4, a query statement according to one embodiment of the present invention may include information regarding the functions or capabilities of the agent to be searched, such as "product feature extraction," "competitor search," "scraping" (corresponding to 3. collection of detail page information), "OCR," and "phrase analysis," as information for searching for an agent capable of performing each unit function. Of course, information regarding the functions or capabilities of the agent may be expressed as vector (embedding)-based data for searching by semantic similarity instead of this method and included in the query statement. In addition, a query statement according to one embodiment of the present invention may also include information regarding the input / output data formats (inputType and outputType) of each agent. To generate such a query statement, a workflow extraction unit (220) according to one embodiment of the present invention may use a query statement generator having prior knowledge of the agent graph.

[0085] Continuing, a workflow extraction unit (220) according to one embodiment of the present invention can extract a workflow from an agent graph using a query statement generated as above. For example, with reference to FIG. 3, node N1 may correspond to a first agent (a1) that performs a "product feature extraction" function, node N2 to a second agent (a2) that performs a "competitor search" function, node N3 to a third agent (a3) ​​that performs a "scraping" function, node N4 to a fourth agent (a4) that performs an "OCR" function, and node N5 to a fifth agent (a5) that performs a "phrase analysis" function. Then, a workflow (WF1) including nodes N1 to N5 and edges (E12, E23, E34, and E45) connecting the nodes may be extracted from the agent graph.

[0086] A workflow extraction unit (220) according to one embodiment of the present invention can support a user who has entered a request to perform a task to use the extracted workflow (WF1).

[0087] Specifically, the workflow extraction unit (220) according to one embodiment of the present invention configures an execution environment (e.g., an inference environment) of the workflow (WF1) before providing the extracted workflow (WF1) to the user, and can test whether the user's intended result is obtained when a task execution request is processed according to the workflow (WF1). Through this, the user can use an already configured workflow as is or with partial modifications (e.g., enhancement of the agent described later) without the need to individually call each agent (node) or create a workflow to directly configure the execution environment.

[0088] In addition, if necessary, frequently used (called) workflows may be prepared in advance on a server (not shown), for example, by configuring the execution environment and / or testing the frequently used (called) workflows, so that a user's request for task execution can be quickly responded to simply by extracting the workflow without the need to configure the environment anew.

[0089] Continuing, the workflow extraction unit (220) according to one embodiment of the present invention may support the user in downloading and executing the extracted workflow on a PC or executing it via a server or cloud. In some cases, the user may perform the task using the agent graph itself without separately downloading the workflow environment configuration. When using the agent graph itself, a UI / UX or API for performing input / output to the agent graph may be provided to the user.

[0090] According to one embodiment of the present invention, orchestration may be supported in the execution environment to enable the smooth operation of a multi-agent system (MAS). Additionally, according to one embodiment of the present invention, a user-customized UI / UX may be supported in the execution environment for users to request various operations and perform input / output, and various tools, databases, etc., for storing / retrieving user data or logging / monitoring user activities may be supported.

[0091] Meanwhile, a workflow extraction unit (220) according to one embodiment of the present invention may determine a workflow included in a pair data set as a workflow extracted from an agent graph in response to the existence of a pair data set containing a workflow that performs a plurality of unit functions.

[0092] Specifically, according to one embodiment of the present invention, in the process where a request processing unit (210) processes a user's request for task execution to derive a plurality of unit functions and a workflow extraction unit (220) extracts a workflow that performs at least a portion of the plurality of unit functions, a pair data set in a format similar to [task execution request (or a plurality of unit functions derived therefrom), workflow] may be constructed. By learning such a pair data set in a manner such as embedding, the workflow extraction unit (220) according to one embodiment of the present invention can quickly search (extract) a workflow that matches the user's request for task execution (or a plurality of unit functions derived therefrom) without the need to search the agent graph every time.

[0093] According to one embodiment of the present invention, as described above, there may be multiple workflows extracted by the workflow extraction unit (220). In this case, the workflow extraction unit (220) according to one embodiment of the present invention may determine a recommended workflow among the multiple workflows based on scores for the multiple workflows.

[0094] For example, with reference to FIG. 5, a situation may be assumed in which a workflow extraction unit (220) extracts a workflow containing unit functions necessary for performing a task requested by a user from an agent graph, and as a result, a first workflow (WF1) composed of nodes (N1, N2, N3, N4 and N5) and edges (E12, E23, E34 and E45) and a second workflow (WF2) composed of nodes (N1, N2, N3, N4 and N6) and edges (E12, E23, E34 and E46) are extracted. In this case, the workflow extraction unit (220) according to an embodiment of the invention may determine the recommended workflow between the two based on scores for each of the first workflow (WF1) and the second workflow (WF2). According to an embodiment of the present invention, the score for each workflow may be calculated based on factors such as the performance (accuracy, response speed, etc.), cost, user rating, and usage frequency of the corresponding workflow, but is not limited thereto.

[0095] In addition, according to one embodiment of the present invention, at least some of the plurality of agents (nodes) constituting the workflow may have a version suitable for the workflow extracted by the workflow extraction unit (220) among the plurality of versions applied.

[0096] For example, with reference to FIG. 6, among the nodes included in the agent graph, there may be nodes included in multiple workflows (WF1 and WF3), such as node (N4). This node (N4) may exist in multiple versions with different attribute values ​​so that different versions can be applied to each workflow to which it belongs, as needed. For example, when the first workflow (WF1) is extracted by the workflow extraction unit (220), the first version may be applied to node (N4) and the node (N4) of the first version may be included in the first workflow (WF1), and when the third workflow (WF3) is extracted, the third version may be applied to node (N4) and the node (N4) of the third version may be included in the third workflow (WF3).

[0097] Meanwhile, according to one embodiment of the present invention, at least some of the plurality of agents (nodes) constituting the workflow can be scaled based on usage frequency.

[0098] Specifically, according to one embodiment of the present invention, among the agents (nodes) included in the agent graph, there may be some that are used more frequently than other nodes, and conversely, there may be some that are used less frequently than other nodes. For example, an agent that is included in a plurality of workflows (WF1 and WF3), such as node (N4) in FIG. 6, or an agent that is included in a frequently used workflow, may be used more frequently than other nodes. The graph management unit (230) according to one embodiment of the present invention described below can operate the agent graph efficiently by scaling agents with high usage frequency by increasing their number within the agent graph, and by scaling agents with low usage frequency by reducing their number within the agent graph or leaving only brief information within the graph so that they can be added to the agent graph later if needed.

[0099] Next, a graph management unit (230) according to one embodiment of the present invention can perform the function of providing a user with an interface that allows a new agent to be added to the workflow.

[0100] Specifically, according to one embodiment of the present invention, the user may add agents to a workflow by inserting a new node in the middle of the workflow extracted by the workflow extraction unit (220) as needed, or by attaching a new node to the front or back of the workflow. Additionally, according to one embodiment of the present invention, the user may add a new agent (i.e., an enhanced agent) to the workflow by enhancing an agent already existing in the workflow, and a detailed explanation thereof will be provided later. According to one embodiment of the present invention, the user may simply set up a new LLM-based agent and add it to the workflow through the above interface, or may add an agent by linking an external API or a directly developed AI model to the agent.

[0101] A graph management unit (230) according to one embodiment of the present invention may provide a user with an interface that allows a new agent to be added to the workflow in response to the fact that the workflow extracted by the workflow extraction unit (220) cannot perform some of the multiple unit functions derived by the request processing unit (210). At this time, the graph management unit (230) according to one embodiment of the present invention may also provide the user with information regarding the unit functions that cannot be performed.

[0102] Specifically, according to one embodiment of the present invention, it may be assumed that there is no agent capable of performing a specific unit function among a plurality of unit functions derived by the request processing unit (210) in the agent graph. In such a case, the graph management unit (230) according to one embodiment of the present invention may provide the user with an interface that allows a new agent capable of performing the specific unit function to be added to a workflow (in this case, the workflow may include an empty node corresponding to the new agent to be added, or may include agents (nodes) that can be connected when the new agent (node) is added), while providing the user with information regarding the specific unit function (e.g., the function, the required input / output data format, etc.).

[0103] A graph management unit (230) according to one embodiment of the present invention may add a newly added agent to an agent graph in response to a user adding a new agent to a workflow using the interface provided as above. To this end, the graph management unit (230) according to one embodiment of the present invention may perform a test to check whether there are any risk factors when a new agent is added to the agent graph, and may add the new agent to the agent graph only if there are no risk factors. The new agent added by the user in this way may be included in the agent graph along with a new edge, taking into account other agents (nodes) that can be connected to the agent, the format of data exchanged, etc.

[0104] Furthermore, the graph management unit (230) according to one embodiment of the present invention can allow a new agent added to the agent graph as described above to be registered in a marketplace. That is, a user can add a new agent to the agent graph, but other users can use the agent only if they purchase the agent from the marketplace.

[0105] As described above, according to one embodiment of the present invention, a new agent added to a workflow (or agent graph) by a user may be an enhanced version of an agent already existing in the workflow (or agent graph). In this case, a graph management unit (230) according to one embodiment of the present invention may encrypt information regarding the enhancement of the enhanced agent and associate the enhanced agent with the encrypted information regarding the enhancement.

[0106] Specifically, according to one embodiment of the present invention, a user can enhance an agent that already exists in a workflow (or agent graph) by utilizing their knowledge or know-how, such as by linking an external API, a self-developed AI model, etc., to the agent, applying a specific prompt, or linking a specific DB. Here, according to one embodiment of the present invention, enhancement may mean processing such as making the target agent produce better results, making the response speed of the target agent faster, or reducing the computing resources required to use the target agent (e.g., lightweighting the agent).

[0107] A graph management unit (230) according to one embodiment of the present invention can encrypt information regarding the enhancement of an enhanced agent, that is, information regarding actions taken by a user to enhance an agent as described above, and store it in a format such as a Vector DB, Graph DB, or RDB (Relational Database), and associate it with the enhanced agent. By doing so, when another user uses an agent enhanced by a specific user or a workflow containing such an agent, the information regarding the enhancement of the agent is not known to that other user, thereby protecting the specific user's knowledge or know-how. According to one embodiment of the present invention, such enhanced agents can also be traded through the aforementioned marketplace.

[0108] As described above, if multiple users utilizing the agent graph according to various embodiments of the present invention continuously develop the agent graph by enhancing existing agents and adding new agents, a large-scale ecosystem of various interconnected agents similar to a large ontology graph is formed, and it becomes possible to effectively respond to various task execution requests from users solely through combinations of agents (i.e., workflows) included in the agent graph.

[0109] Next, a communication unit (240) according to one embodiment of the present invention can perform the function of enabling data transmission and reception from / to a request processing unit (210), a workflow extraction unit (220), and a graph management unit (230).

[0110] Finally, the control unit (250) according to one embodiment of the present invention can perform the function of controlling the flow of data between the request processing unit (210), the workflow extraction unit (220), the graph management unit (230), and the communication unit (240). That is, the control unit (250) according to one embodiment of the present invention can control the request processing unit (210), the workflow extraction unit (220), the graph management unit (230), and the communication unit (240) to perform their respective unique functions by controlling the flow of data from / to / from the outside of the agent graph management system (200) or the flow of data between each component of the agent graph management system (200).

[0111] The embodiments according to the present invention described above may be implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc., either individually or in combination. The program instructions recorded on the computer-readable recording medium may be those specifically designed and configured for the present invention or those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. Hardware devices may be modified into one or more software modules to perform processing according to the present invention, and vice versa.

[0112] Although the present invention has been described above with reference to specific details such as specific components, limited embodiments, and drawings, this is provided only to aid in a more comprehensive understanding of the invention, and the invention is not limited to the above embodiments, and a person skilled in the art to which the invention belongs can make various modifications and changes from this description.

[0113] Accordingly, the scope of the present invention should not be limited to the embodiments described above, and all scopes equivalent to or equivalently modified from the claims set forth below, as well as the claims set forth below, shall be considered to fall within the scope of the concept of the present invention.

Claims

1. As a method to support the use of an agent graph, A step of deriving a plurality of unit functions necessary for the execution of the above task by processing a user's request for task execution, and The method includes the step of extracting a workflow from an agent graph that performs at least some of the above-mentioned plurality of unit functions, The above workflow is composed of a plurality of agents that perform each unit function included in the plurality of unit functions. method.

2. In Paragraph 1, In the extraction step above, a query statement for searching the workflow in the agent graph based on the derived plurality of unit functions is generated, and the workflow is extracted using the query statement. method.

3. In Paragraph 1, In the extraction step above, in response to the existence of a pair data set including a workflow that performs the plurality of unit functions, a workflow included in the pair data set is determined as the extracted workflow. method.

4. In Paragraph 1, In response to the fact that there are multiple workflows extracted above, the method further includes the step of determining a recommended workflow among the multiple workflows based on a score for the multiple workflows. method.

5. In Paragraph 1, At least some of the above multiple agents have a version suitable for the workflow applied among the multiple versions. method.

6. In Paragraph 1, At least some of the above plurality of agents are scaled based on usage frequency. method.

7. In Paragraph 1, The method further includes the step of providing the user with an interface that enables the addition of a new agent to the workflow. method.

8. In Paragraph 7, In the above providing step, the interface is provided to the user in response to the workflow being unable to perform some of the plurality of unit functions, and information regarding the unit functions that cannot be performed is also provided to the user. method.

9. In Paragraph 7, The method further includes the step of adding the new agent to the agent graph in response to the user adding the new agent to the workflow using the interface. method.

10. In Paragraph 9, In the above additional step, to register the new agent in the marketplace method.

11. In Paragraph 9, The new agent added to the workflow by the user is an enhanced version of an agent already existing in the workflow, and In the above additional step, information regarding the enhancement of the enhanced agent is encrypted, and the enhanced agent is associated with the encrypted information regarding the enhancement. method.

12. A non-transient computer-readable recording medium for recording a computer program for executing the method according to paragraph 1.

13. As a system to support the use of an agent graph, A request processing unit that processes a user's request to perform a task and derives a plurality of unit functions necessary for the performance of the above task, and It includes a workflow extraction unit that extracts a workflow from an agent graph that performs at least some of the above-mentioned plurality of unit functions, The above workflow is composed of a plurality of agents that perform each unit function included in the plurality of unit functions. System.

14. In Paragraph 13, The workflow extraction unit generates a query statement for searching for the workflow in the agent graph based on the derived plurality of unit functions, and extracts the workflow using the query statement. System.

15. In Paragraph 13, The workflow extraction unit determines a workflow included in the pair data set as the extracted workflow in response to the existence of a pair data set containing workflows that perform the plurality of unit functions. System.

16. In Paragraph 13, The workflow extraction unit determines a recommended workflow among the plurality of workflows based on a score for the plurality of workflows in response to the fact that the extracted workflows are plurality. System.

17. In Paragraph 13, At least some of the above multiple agents have a version suitable for the workflow applied among the multiple versions. System.

18. In Paragraph 13, At least some of the above plurality of agents are scaled based on usage frequency. System.

19. In Paragraph 13, A graph management unit further comprising providing the user with an interface to add a new agent to the workflow. System.

20. In Paragraph 19, The graph management unit provides the interface to the user in response to the workflow being unable to perform some of the plurality of unit functions, and together provides information regarding the unit functions that cannot be performed to the user. System.

21. In Paragraph 19, The graph management unit adds the new agent to the agent graph in response to the user adding the new agent to the workflow using the interface. System.

22. In Paragraph 21, The above graph management unit enables the above new agent to be registered in the marketplace. System.

23. In Paragraph 21, The new agent added to the workflow by the user is an enhanced version of an agent already existing in the workflow, and The graph management unit above encrypts information regarding the enhancement of the enhanced agent and associates the encrypted information regarding the enhancement with the enhanced agent. System.

Images