Method for quickly creating orchestrable intelligence agent by using fixed rule

Abstract

The invention provides a method for quickly creating an orchestrable intelligence agent by utilizing a fixed rule, which comprises the following steps of: tool configuration: customizing a business logic code by a creator, and then adding a tool name and description on a configuration interface; and flow arrangement: according to the functions of the agent to be realized, after a user selects chain nodes on a system interface and fills in configuration items of related nodes, each node determines an input or output type, nodes of the same type or sub-type stipulate the processing flow direction of the flow, and after the user inputs and outputs, the system stores flow information. And agent creation: a series of visual configurations are compiled into logic codes by automatically generating a framework through the flow arrangement storage in the background, and built-in nodes are dynamically called through node configuration items to realize agent creation. According to the method, the capability of visually arranging the large model application is realized, the method is friendly to users in non-professional fields, the large model application can be rapidly enabled, and the threshold of integrating a digital assistant in an application system is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of comprehensive visualization application of large language models, and relates to a method for quickly creating an intelligence agent that can be programmed using fixed rules. Background Art

[0002] In the comprehensive application field of large language models, with the continuous development of large language model applications, we have witnessed a rapid growth trend in the number of multi-language, multi-modal large models released by various manufacturers. The application fields of these models are also constantly developing in the direction of multi-functionality and specialization, providing users with powerful tools to solve complex problems in actual business. However, due to the diverse sources, types and specifications of large language models, users may need to master some professional knowledge or consult complex operation manuals before they can skillfully apply large models to certain application scenarios, which invisibly increases the user's usage threshold. Therefore, improving the application visualization operation of large language models and realizing the ability to visually orchestrate large models is an effective solution. Summary of the invention

[0003] 1. Technical problems to be solved: Existing large language models in the field of intelligent intelligence are not user-friendly, have high learning costs, and are difficult to implement in specific application scenarios.

[0004] 2. Technical solution: In order to solve the above problems, the present invention provides a method for quickly creating an intelligence agent that can be programmed using fixed rules, comprising the following steps: Step S01: Tool configuration: A tool is a function that an intelligent entity needs to perform. The specific creator can customize the business logic code, and then add the tool name and description in the configuration interface, and the system will have available tool components.

[0005] Step S02: Process orchestration: According to the intelligent agent function to be realized, the user selects the model, tool, prompt word template, and built-in business chain node in the system interface, and fills in the configuration items of the relevant nodes. Each node determines the input or output type, and the nodes of the same type or subclass specify the processing flow of the process. After the user inputs and outputs, the system saves the process information in JSON format to realize the application orchestration of fixed rule processes.

[0006] Step S03: Agent creation: responsible for the generation of the overall functional logic of the orchestrated process. The JSON stored in the process orchestration automatically generates a framework in the background, compiles a series of visual configurations into logic codes, and dynamically calls built-in nodes through node configuration items to realize the creation of the agent.

[0007] Furthermore, it also includes the release of digital assistants: publishing the intelligent application as an API service with one click. Each time an intelligent service is released, the service will be registered in the background maintenance service interface to complete the service release. At the same time, the system calls code scripts in multiple languages ​​for embedding in different platforms and achieving docking with third-party application systems.

[0008] Furthermore, in step S01, the tool set is customized and optimized by visually configuring the tool name and function description.

[0009] Furthermore, the configured tools and functions adopt a unified background interface standard, and the front-end page is configured with clear tool names and function descriptions.

[0010] Furthermore, in step S02, each node component on the corresponding chain is selected in the interface, placed on the canvas, and according to the node input and output type, the output of a node is selected and dragged to connect to the input of the next node. The front-end page automatically draws connections between the input and output nodes to implement the process diagram.

[0011] Furthermore, in step S04, the third-party application system is one of Python, JavaScript, and Shell. The third-party application system adds the calling code to the corresponding application system to achieve one-click docking and publishing, forming a digital assistant for the business system.

[0012] Furthermore, the digital assistant release provides configuration options and customization functions to configure the digital assistant title, avatar, and welcome message.

[0013] 3. Beneficial effects: The present invention provides a method for quickly creating an intelligence agent that can be programmed using fixed rules, which realizes the ability to visually program large model applications. It is user-friendly to non-professional users, can quickly enable large model applications, and lowers the threshold for integrating digital assistants into application systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0015] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0016] An intelligent agent is an application that can use a large model to make autonomous decisions, invoke tools to achieve goals, and evaluate the optimization results over time.

[0017] The present invention provides a method for quickly creating an intelligence agent that can be arranged using fixed rules, which is characterized by comprising the following steps: Step S01: Tool configuration: A tool is a function that an intelligent entity needs to perform. The specific creator can customize the business logic code, and then add the tool name and description in the configuration interface, and the system will have available tool components.

[0018] Step S02: Process orchestration: According to the intelligent agent function to be realized, the user selects the model, tool, prompt word template, and built-in business chain node in the system interface, and fills in the configuration items of the relevant nodes. Each node determines the input or output type, and the nodes of the same type or subclass specify the processing flow of the process. After the user inputs and outputs, the system saves the process information in JSON format to realize the application orchestration of fixed rule processes.

[0019] Step S03: Agent creation: responsible for the generation of the overall functional logic of the orchestrated process. The JSON stored in the process orchestration automatically generates a framework in the background, compiles a series of visual configurations into logic codes, and dynamically calls built-in nodes through node configuration items to realize the creation of the agent.

[0020] Tool configuration adds business logic code and uses the visual configuration module to configure the tool name and description, so that users can create and access built-in tool sets, thereby achieving the use and management of overall capability tools, such as data processing, intelligent tool application, API calls, etc., to meet the requirements of different tasks and needs.

[0021] This flexible configuration method can help users quickly build the capabilities they need in their workflow. By following internal tool interface standards, users only need to implement specific tool logic codes.

[0022] In one embodiment, the customization and optimization of the toolset is accomplished by visually configuring the tool name and function description.

[0023] In one embodiment, the tool configuration supports the configuration of tools and functions, adopts a unified background interface standard, and the front-end page uses clear tool names and function descriptions for configuration, and the operation is clear and convenient for quick configuration and use of the tool; Process orchestration is based on the intelligent function that the user wants to realize, and follows the preset business processing chain. For example, the definition of uploading files, file training, model prompt words, model calling, model output control, etc. is a business processing chain. By selecting each node component on the corresponding chain in the interface, placing it on the canvas, and selecting the output of a node and dragging it to the input of the next node according to the node input and output type, the front-end page will automatically draw lines between the input and output nodes to realize the process diagram. This method allows users to complete the processing node configuration without writing process code, and the system will automatically generate the JSON corresponding to the process, which is convenient for completing the process orchestration of fixed rules and realizing the effective management and scheduling of large-scale model applications. Users can freely combine various configured processing nodes according to their needs, and design processes that meet their business needs according to the business processing chain, so as to achieve more efficient and flexible intelligent applications.

[0024] In one embodiment, an intelligent agent of "Smart PDF Reading" is constructed. Users can combine the "File Loader" node, "File Embedding" node, "Data Storage" node, and "Large Model Calling" node in order in the interface, connect all nodes according to input and output, and link them to the smart reading business chain to complete the overall process arrangement. This visual arrangement method greatly simplifies the complexity of process design and improves the work efficiency and productivity of users.

[0025] It also includes digital assistant publishing: the intelligent application can be published as an API service with one click, and the background maintains the service interface. Each time an intelligent service is published, the service will be registered in the service interface to complete the service publishing. At the same time, the system has calling code scripts in multiple languages ​​for embedding in different platforms and achieving docking with third-party application systems.

[0026] The digital assistant releases the service interface maintained by the background. Every time an intelligent agent is created and released, a unique identifier is assigned to the service interface and the service is registered to complete the service release. When calling the digital assistant, the system can locate a certain intelligent agent API service in the service interface through the API unique identifier and complete the service call.

[0027] In one embodiment, the system provides calling code scripts in multiple languages ​​for embedding calls in different platforms, such as Python, JavaScript, Shell, etc. The third-party application system only needs to add the calling code to the corresponding application system to achieve one-click docking and release, thereby forming a digital assistant for the business system.

[0028] In one embodiment, the system has programmed an intelligent agent for "text emotion recognition". After publishing the API interface service, the third-party application system only needs to pass the text to be recognized to the publishing address through the interface. The intelligent agent generates text emotion information according to the programmed process and returns it to the third-party application system, thus completing the docking call of the text emotion recognition assistant.

[0029] In one embodiment, the digital assistant release also provides a wealth of configuration options and customization functions, such as configuring the digital assistant title, avatar, welcome message, etc., so that users can customize it according to their needs to meet the needs of different business scenarios. Such a digital assistant system can effectively improve the intelligence level of the business system and provide users with a more convenient large-scale model application service experience.

Claims

1. A method for quickly creating an intelligence agent that can be programmed using fixed rules, characterized in that: The following steps are involved: Step S01: Tool configuration: A tool is a function that an agent needs to perform. The specific creator can customize the business logic code, and then add the tool name and description in the configuration interface, and the system will have available tool components; Step S02: Process orchestration: According to the intelligent agent function to be implemented, the user selects the model, tool, prompt word template, and built-in business chain node in the system interface, and fills in the configuration items of the relevant nodes. Each node determines the input or output type, and the nodes of the same type or subclass specify the processing direction of the process. After the user inputs and outputs, the system saves the process information in JSON format to realize the application orchestration of fixed rule processes; Step S03: Agent creation: responsible for the generation of the overall functional logic of the orchestrated process. The JSON stored in the process orchestration automatically generates a framework in the background, compiles a series of visual configurations into logic codes, and dynamically calls built-in nodes through node configuration items to realize the creation of the agent.

2. The method for quickly creating an intelligence agent that can be arranged using fixed rules as claimed in claim 1, characterized in that: It also includes the release of digital assistants: publishing intelligent applications as API services with one click. Each time an intelligent service is released, it will be registered in the background maintenance service interface to complete the service release. At the same time, the system calls code scripts in multiple languages ​​for embedding in different platforms and achieving docking with third-party application systems.

3. The method for quickly creating an intelligence agent that can be programmed using fixed rules as described in claim 1 or 2, characterized in that: In step S01, the tool set is customized and optimized by visually configuring the tool name and function description.

4. The method for quickly creating an intelligence agent programmable by using fixed rules as claimed in claim 3, characterized in that: The configured tools and functions adopt a unified background interface standard, and the front-end page is configured with clear tool names and function descriptions.

5. The method for quickly creating an intelligence agent that can be programmed using fixed rules as described in claim 1 or 2, characterized in that: In step S02, select each node component on the corresponding chain in the interface, place it on the canvas, and select the output of a node and drag it to the input of the next node according to the node input and output type. The front-end page automatically draws connections between the input and output nodes to implement the process diagram.

6. The method for rapidly creating an intelligence agent programmable by using fixed rules as claimed in claim 2, characterized in that: In step S04, the third-party application system is one of Python, JavaScript, and Shell. The third-party application system adds the calling code to the corresponding application system to achieve one-click docking and publishing, forming a digital assistant for the business system.

7. The method for rapidly creating an intelligence agent programmable by using fixed rules as claimed in claim 2, characterized in that: Digital assistant publishing provides configuration options and customization functions to configure the digital assistant title, avatar, and welcome message.

Images