CN118819617A - Configuration method and device of large model application intelligent agent - Google Patents

Abstract

The application relates to a configuration method and a device of a large model application agent, wherein the method comprises the steps of receiving an agent configuration instruction; the intelligent agent configuration instruction is obtained based on web front end configuration; configuring corresponding agent metadata information according to the agent configuration instruction, and storing the information in metadata service; and the agent metadata information includes knowledge base metadata information, plug-in metadata information, workflow metadata information and database metadata information; and calling the large model agent service to create an agent framework, and binding the agent metadata information with the agent framework to create the agent. The application can greatly reduce the difficulty of developing the intelligent agent in the configuration process through the web visual configuration mechanism, shield the technologies such as a bottom large language model, a knowledge base and the like, and enable business personnel who do not know the technology to conveniently develop the intelligent agent.

Description

Configuration method and device of large model application intelligent agent

Technical Field

The present disclosure relates to the field of computer technology, and in particular to a configuration method and device for a large model application intelligent agent.

Background Art

Large model application agents refer to agents that rely on current advanced technologies such as large language models and multimodal models, and guide agents to produce text or image outputs in specific fields through prompts (instructions) combined with the construction of vertical knowledge systems. The above instructions can be in the form of questions, task descriptions, conditions, etc. When the agent receives the instruction, it will use its pre-trained model and the corresponding vertical domain knowledge base data for reasoning and generation to produce output with specific domain expertise.

As big model technology is increasingly used in enterprises, and more and more enterprises are not satisfied with the output of daily general knowledge, building their own big model intelligent agents based on their internal knowledge systems is becoming an increasingly urgent need and challenge. For example, in fields such as healthcare, law, or finance, intelligent agents are required to understand and answer questions in these specific fields more accurately.

At present, the mainstream way to develop application intelligent agents is to use Python technology, which is relatively inefficient and complex. For people who do not understand technology, it is impossible to achieve the purpose of intelligent agent development.

Summary of the invention

In view of this, the present application proposes a configuration method and device for a large model application agent to solve the above-mentioned problems.

In one aspect, the present application proposes a configuration method for a large model application agent, comprising the following steps:

Receiving agent configuration instructions; wherein the agent configuration instructions are obtained based on the web front-end configuration;

According to the agent configuration instruction, corresponding agent metadata information is configured and stored in the metadata service; and the agent metadata information includes knowledge base metadata information, plug-in metadata information, workflow metadata information and database metadata information;

The large model agent service is called to create an agent framework, and the agent metadata information is bound to the agent framework to create an agent.

As an optional implementation scheme of the present application, optionally, configuring corresponding agent metadata information according to the agent configuration instruction and storing it in the metadata service includes:

Calling the knowledge base service according to the agent configuration instruction, and storing the knowledge document in the knowledge base service in a preset manner;

The metadata service is called, and the knowledge base metadata information corresponding to the knowledge document is stored in the metadata service.

As an optional implementation scheme of the present application, optionally, the preset storage method includes:

After the knowledge document is vectorized through a preset model, it is stored in a corresponding vector knowledge base.

As an optional implementation scheme of the present application, optionally, the preset storage method further includes:

Keywords of the knowledge in the knowledge document are extracted using text frequency, classification and clustering algorithms, and the knowledge and the keywords are stored in a corresponding search engine.

As an optional implementation scheme of the present application, optionally, configuring corresponding agent metadata information according to the agent configuration instruction and storing it in the metadata service also includes:

Call the plug-in service according to the agent configuration instruction, configure the corresponding plug-in, and persist it in the plug-in service;

The metadata service is called, and plug-in metadata information corresponding to the plug-in persistence is stored in the metadata service.

As an optional implementation scheme of the present application, optionally, configuring corresponding agent metadata information according to the agent configuration instruction and storing it in the metadata service also includes:

Calling database services according to the agent configuration instructions to create databases and data tables;

The metadata service is called, and database metadata information corresponding to the database and the data table is stored in the metadata service.

As an optional implementation scheme of the present application, optionally, configuring corresponding agent metadata information according to the agent configuration instruction and storing it in the metadata service also includes:

Calling the workflow service according to the agent configuration instruction to configure the workflow;

The metadata service is called, and the workflow metadata information corresponding to the workflow is stored in the metadata service.

As an optional implementation scheme of the present application, optionally, the workflow includes a plurality of work nodes for connecting to corresponding third-party business systems.

As an optional implementation scheme of the present application, optionally, when calling the large model agent service to create an agent framework, guide words are configured based on the web front end, and the guide words are matched to the corresponding large language model in a preset manner.

In another aspect, the present application provides a device for implementing the configuration method of the large model application agent described in any one of the above items, comprising:

An instruction receiving module is configured to receive an agent configuration instruction; and the agent configuration instruction is obtained based on the web front-end configuration;

A metadata information configuration module is configured to configure corresponding agent metadata information according to the agent configuration instruction and store it in the metadata service; and the agent metadata information includes knowledge base metadata information, plug-in metadata information, workflow metadata information and database metadata information;

The intelligent agent creation module is configured to call the large model intelligent agent service to create an intelligent agent framework, and bind the intelligent agent metadata information with the intelligent agent framework to create an intelligent agent.

Technical effects of the present invention:

The present application proposes a configuration method for a large model application agent, including receiving agent configuration instructions; and the agent configuration instructions are obtained based on the web front-end configuration; configuring the corresponding agent metadata information according to the agent configuration instructions, and storing it in the metadata service; and the agent metadata information includes knowledge base metadata information, plug-in metadata information, workflow metadata information and database metadata information; calling the large model agent service to create an agent framework, and binding the agent metadata information to the agent framework to create an agent. The present application can greatly reduce the difficulty of developing agents during the configuration process through a web visual configuration mechanism, shielding the underlying large language model, knowledge base and other technologies, so that business personnel who do not understand technology can also easily develop agents.

Further features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments with reference to the attached drawings.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.

FIG1 is a flowchart showing a configuration method of a large model application agent according to an embodiment of the present application;

FIG2 is a schematic diagram showing an implementation flow of a configuration method for a large model application agent according to an embodiment of the present application;

FIG3 is a schematic diagram showing a flow chart of a method for using a large model application agent according to an embodiment of the present application;

FIG4 is a schematic diagram showing an implementation flow of a method for using a large model application agent according to an embodiment of the present application;

FIG5 shows an overall topology diagram of a large model application agent according to an embodiment of the present application.

DETAILED DESCRIPTION

Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.

The configuration method of the large model application agent of the present application adopts a web front-end and a back-end server to form an agent development platform, and the back-end server includes metadata services, large model agent services, knowledge base services, plug-in services, workflow services and database services. In the process of interaction with the agent development platform, the user can efficiently and accurately build the desired agent based on the knowledge base construction capabilities, workflow configuration capabilities and agent framework configuration capabilities of the agent development platform. Among them, when configuring the development agent, the relevant management personnel call the interface of the back-end server through the web front-end, save the agent information to the metadata service, and display the metadata of the data in the form of a web, so that business personnel who do not understand technology can also quickly develop model application agents that meet business needs, effectively reducing the development difficulty of users.

Example 1

As shown in FIG. 1 and FIG. 2 , on the one hand, the present application proposes a configuration method for a large model application agent, comprising the following steps:

S1, receiving agent configuration instructions; and the agent configuration instructions are obtained based on the web front-end configuration;

In this step, in order to develop a complete intelligent agent, you first need to receive the corresponding intelligent agent configuration instructions based on the web front-end configuration and call the corresponding background server, so that the background service can perform knowledge base configuration, database configuration, workflow configuration, and plug-in configuration according to the intelligent agent configuration instructions.

S2. According to the agent configuration instruction, the corresponding agent metadata information is configured and stored in the metadata service; and the agent metadata information includes knowledge base metadata information, plug-in metadata information, workflow metadata information and database metadata information;

In this step, the platform administrator sends the corresponding intelligent agent configuration instructions to the background service through the web front end. After receiving the intelligent agent configuration instructions sent by the web front end, the background service configures the corresponding intelligent agent metadata information, and stores the intelligent agent metadata information in the metadata service. When subsequent users query, they can query the metadata information configured by calling the metadata service, and call the corresponding background service according to the configured metadata information to query the required information.

S3. Call the large model agent service to create an agent framework, and bind the agent metadata information to the agent framework to create an agent.

In this step, the platform administrator calls the large model intelligent agent service, i.e., the large model agent service, through the web front end, creates an intelligent agent framework, and configures various configurations required to create the intelligent agent. Moreover, after the large model intelligent agent service creates the intelligent agent framework, it calls the metadata service to obtain the intelligent agent metadata information stored in the metadata service in the above step S200, binds it to the intelligent agent framework, and finally realizes the development of a complete intelligent agent, so that the intelligent agent has multiple capabilities such as knowledge base, plug-in, workflow, and database query.

As an optional implementation scheme of the present application, optionally, corresponding intelligent agent metadata information is configured according to the intelligent agent configuration instructions and stored in the metadata service, including: calling the knowledge base service according to the intelligent agent configuration instructions, and storing the knowledge document in the knowledge base service in a preset manner; calling the metadata service, and storing the knowledge base metadata information corresponding to the knowledge document in the metadata service.

It should be noted that the knowledge base is configured to expand the ability of the intelligent body's vertical business domain knowledge. In configuring the knowledge base metadata information, the platform administrator calls the knowledge base service through the web front end and uploads the knowledge file to the knowledge base service. Among them, after the knowledge base service receives the uploaded knowledge document, it saves it according to the preset storage method. Furthermore, after the knowledge base saves the knowledge document, it calls the metadata service and sends the metadata information stored in the document to the metadata service, which stores the corresponding knowledge base metadata information.

As an optional implementation scheme of the present application, optionally, the preset storage method includes: vectorizing the knowledge document through a preset model and storing it in a corresponding vector knowledge base.

It should be noted that in one scenario, after the knowledge base service receives the uploaded knowledge document, it vectorizes the knowledge through the preset model and then saves it to the corresponding vector library. Among them, the preset model is preferably the text2vec-bge-large-chinese model, which can effectively improve the recall rate of Chinese knowledge. After the knowledge is vectorized by the text2vec-bge-large-chinese model, it is saved in the Milvus vector knowledge base.

As an optional implementation scheme of the present application, optionally, the preset storage method also includes: using text frequency, classification and clustering algorithms to extract keywords of knowledge in knowledge documents, and storing the knowledge and keywords in a corresponding search engine.

It should be noted that in another scenario, text frequency, classification and clustering algorithms are used to extract knowledge keywords and save them in the es search engine, ultimately achieving a hybrid recall method of knowledge that combines semantics with keywords, greatly improving the recall rate.

As an optional implementation scheme of the present application, optionally, corresponding intelligent agent metadata information is configured according to the intelligent agent configuration instructions and stored in the metadata service, and also includes: calling the plug-in service according to the intelligent agent configuration instructions, configuring the corresponding plug-in, and persisting it in the plug-in service; calling the metadata service, and storing the plug-in metadata information corresponding to the plug-in persistence in the metadata service.

It should be noted that the plug-in configuration is used to expand the ability of the intelligent agent to obtain instant information. In the process of configuring the plug-in metadata information, the platform administrator calls the plug-in service through the web front end. The configuration of the plug-in can enable the intelligent agent to have specific information query, image recognition, and analysis capabilities. Among them, the plug-in service persists the corresponding plug-in into the service memory through the class load technology. After the plug-in service persists the plug-in, it calls the metadata service and sends the plug-in persistent metadata information to the metadata service, which stores the metadata.

As an optional implementation scheme of the present application, optionally, corresponding intelligent agent metadata information is configured according to the intelligent agent configuration instructions and stored in the metadata service, and also includes: calling the database service according to the intelligent agent configuration instructions to create a database and data tables; calling the metadata service and storing the database metadata information corresponding to the database and data tables in the metadata service.

It should be noted that in the process of configuring database metadata information, the platform administrator calls the database service through the web front end to create the database and data table. After receiving the creation information, the database service calls the MySQL database to create the database and data table. After the database service creates the database and data table, it calls the metadata service and sends the metadata information of the created database and data table to the metadata service, which then performs the corresponding metadata storage.

As an optional implementation scheme of the present application, optionally, corresponding intelligent agent metadata information is configured according to the intelligent agent configuration instructions and stored in the metadata service, and also includes: calling the workflow service according to the intelligent agent configuration instructions to configure the workflow; calling the metadata service and storing the workflow metadata information corresponding to the workflow in the metadata service.

As an optional implementation scheme of the present application, optionally, the workflow includes a plurality of work nodes for connecting to corresponding third-party business systems.

It should be noted that workflow configuration is used to enable the agent to interact with external business systems. In the process of configuring workflow metadata information, the platform administrator calls the workflow service through the web front end to configure the workflow that can be connected to the third-party business system. Among them, after the workflow service receives the workflow node, it configures the workflow node to the Activiti workflow engine. After the workflow service saves the workflow, it calls the metadata service and sends the metadata information stored in the workflow to the metadata service, which stores the corresponding workflow metadata information.

As an optional implementation scheme of the present application, optionally, when calling the large model agent service to create an agent framework, guide words are configured based on the web front end, and the guide words are matched to the corresponding large language model in a preset manner.

It should be noted that the platform administrator calls the large model agent service through the web front end to create an agent framework, including configuring prompt words, which large language model to use, what the opening remarks are, etc., and submits the prompt words, opening remarks and large language model metadata to the metadata service for storing the agent metadata. Among them, the large language module in this embodiment can be any one of GPT4 or Wenxin Yiyan. The administrator inputs the prompt words through the web front end. After the backend service receives the prompt word information, it will give the prompt words to the large language model, so that the agent can have the ability to answer questions in vertical fields. And based on the opening remarks entered by the administrator, the first message sent when the agent starts a conversation with the user is determined. In other words, the configuration of the prompt words is used to guide the role, capabilities, requirements and constraints of the agent, and the large model agent service enables the agent to flexibly use various existing large language models to reason about the answer. Furthermore, after the large model agent service creates the agent framework, it calls the metadata service to obtain other agent metadata information of the above steps and binds it to the agent framework, finally realizing the development of a complete agent, so that the agent has the capabilities of knowledge base, plug-in, workflow and database query.

The following will take the development process of a hospital assistant intelligent agent as an example to explain the method of this application in detail.

1. Build a hospital intelligent body framework.

The agent development platform provides agent configuration capabilities. Guide words can be configured through the web front end. For example, the hospital agent guide words:

##Role: Hospital Assistant

##introduce:

Your name is xx. You are a hospital management expert. You can provide a series of services to patients, hospital leaders, and hospital properties.

##Serve

You have the following service capabilities:

###VR Navigation: xxxxxx.

###Smart registration: xxxxx.

###Test report query: xxxxx.

###Patient Services: xxxxx.

###Medical consultation: xxxxx.

##constraint:

You need to think about the user's questions step by step and then give the answer;

It is not allowed to add fabricated elements to the answers;

In addition to the existing services, you do not have other capabilities. When you encounter user demands that exceed your capabilities, you need to reject the user's request and tell the user that you are still learning these service capabilities.

After receiving the guide word information, the backend server will submit the guide word to the large language model through a prompt (command). For example, the web frontend inputs: What can you do for me? After the backend server submits the question and guide word to the large language model through a command, it returns to the web frontend: Hello! My name is xx. As a hospital management expert, I can provide you with the following services:

###VR Navigation: xxxxxx.

###Smart registration: xxxxx.

###Test report query: xxxxx.

###Patient Services: xxxxx.

###Medical consultation: xxxxx.

This enables the intelligent agent to have the ability to answer questions in the vertical field of the hospital.

2. Bind the knowledge base to the hospital agent and build a private knowledge base system for the hospital agent. The specific configuration process is as follows:

The intelligent agent development platform provides the ability to upload knowledge files, and the file formats include pdf, doc, docx, ppt, pptx, xlsx, markdown, png, jpg, and txt. Upload hospital-related information to the knowledge base through the web front end, including basic hospital information such as buildings, departments, canteens, as well as location information and medical consultation information. After the backend server receives the uploaded knowledge base, it is stored in two ways. One is to vectorize the knowledge through the text2vec-bge-large-chinese model and then save it to the Milvus vector knowledge base. The text2vec-bge-large-chinese model can effectively improve the recall rate of Chinese knowledge; the second is to use text frequency, classification and clustering algorithms to extract knowledge keywords and save them to the es search engine. Finally, the semantic + keyword knowledge hybrid recall method is realized, which greatly improves the recall rate.

3. Furthermore, it also includes the construction of a synonym database for the hospital's private knowledge base.

Since there are many synonyms in the hospital knowledge system, such as treatment, medical treatment, diagnosis and treatment, and diagnosis and treatment, which may express the same meaning at the same time, these words will appear at the same time in the hospital knowledge base. The platform provides the ability to maintain synonyms. Knowledge entity words and synonyms are input through the web front end to perform one-to-many mapping between knowledge entity words and synonyms. After the backend server receives the input from the web front end, it also vectorizes the knowledge entity words and synonyms through the text2vec-bge-large-chinese model and saves them to the Milvus vector knowledge base. Subsequent synonyms entered by users will uniformly hit the same knowledge entity word. For example, if a user enters "diagnosis and treatment" or "medical treatment" on the web front end, after the backend server receives the information, it will first vectorize "diagnosis and treatment" or "medical treatment", and then submit it to the Milvus vector knowledge base for query through the cosine distance technology. The standardized knowledge entity word "treatment" will be uniformly queried, thereby achieving normalization processing, reducing knowledge ambiguity, and effectively improving the knowledge hit rate.

4. Conduct a hit test on the hospital’s proprietary knowledge.

The intelligent agent development platform provides autonomous hit testing capabilities. Users enter a piece of text on the web front end and perform a recall test on the knowledge base to see if the recalled data is correct.

For example, the user inputs: Where is the Department of Neurology? After receiving the message, the backend server first vectorizes the question and then performs knowledge query in two ways. The first is to query knowledge by matching keywords to the es search engine, and the second is to query knowledge by matching vectorized semantics to the Milvus knowledge base. After the two query results are superimposed, the results are rearranged using the ReRank rearrangement technology to further ensure the validity and accuracy of the query results, and the final output result is given to the web frontend: The Department of Neurology is in Room 205, 3rd Floor, Hospital Building.

5. The intelligent development platform is connected with the third-party system.

The intelligent agent development platform provides workflow configuration capabilities, which is used to quickly connect the intelligent agent with third-party business systems. It can be used in business scenarios such as smart registration and test order query.

Take smart registration as an example: the web front end configures a workflow, including the following nodes, namely the start node, semantic parsing node, registration system API call node, and end node. Then enter the information "I want to make a dental appointment at 15:00 this afternoon" on the web front end. After receiving the information, the backend server calls the semantic parsing node and returns a json data structure {"name":"李某","time":"2024-05-15 15:00:00","depart":"dentistry"} by calling the large model capability; call the registration system API node and submit the data structure to the registration system according to the API requirements. Finally, the registration is successfully returned to the web front end to tell the user: Hello, your registration has been successful, please see the doctor in time.

It can be seen from this that the intelligent agent development platform of the present application can provide capabilities such as intelligent agent framework configuration, knowledge base configuration, workflow configuration, etc., and efficiently and accurately build a hospital intelligent agent with service capabilities in the vertical field of the hospital. In addition, through the web visualization configuration mechanism, the difficulty of developing the intelligent agent can be greatly reduced during the configuration process, and the underlying large language model, knowledge base and other technologies can be shielded, so that business personnel who do not understand technology can also easily develop the intelligent agent. Moreover, it is highly maintainable and can quickly iterate the intelligent agent, greatly increasing its flexibility. In other words, through the configuration method of the present application, users do not need to care about how to use technical means to call the large model capabilities, such as what API to call the large model to submit prompt words and what development language to use for the interaction of the large model. The reason is that there are many types of large models, and the calling methods of each large model are different, but through the configuration of the intelligent agent, the calling method of the large model can be shielded, so that users can use the large model without feeling. At the same time. Through the configuration method of this application, users do not need to worry about how to use technical means to use the knowledge base, for example, they do not need to worry about how to segment, vectorize, and recall knowledge documents, and what knowledge base technology to use to save knowledge documents, so that users can use the knowledge base without feeling.

It should be noted that the overall topology of the intelligent development platform of the present application is shown in Figure 5, which includes an agent, a knowledge base, a plug-in, a workflow, a large language model and metadata. Specifically, it is necessary to first obtain a configured agent through manual configuration. Among them, in the process of configuring the agent, the meta information of the agent needs to be stored in the metadata service, including knowledge base metadata, plug-in metadata, database metadata, workflow metadata, and the opening remarks, prompt words, and large language model metadata corresponding to the agent framework. And the above metadata are finally stored in the database by calling the metadata service. Subsequently, in the process of using the configured agent, when the user submits a query instruction to the agent development platform, the agent development platform retrieves relevant information from the knowledge base according to the instruction input by the user, and obtains the relevant data returned by the knowledge base, and then submits the information returned by the knowledge base and the user's query instruction to the large language model, and outputs the final answer through the semantic understanding ability of the large language model, that is, using the agent to answer the question.

Example 2

As shown in FIG. 3 and FIG. 4 , the present application also provides a method for using a large model application agent, and the large model application agent configured based on the configuration method of the large model application agent described in any of the above items is implemented, including the following steps:

S10, the terminal user initiates a query to the configured agent through the web front end, calling the large model agent service;

S20, the large model agent service calls the metadata service according to the query question, and queries the metadata information configured in the metadata service; and the metadata information includes knowledge base metadata information, plug-in metadata information, workflow metadata information, and database metadata information;

S30, the large model agent service calls the knowledge base service according to the knowledge base metadata information to obtain the corresponding knowledge base information;

S40, the large model agent service calls the plug-in service according to the plug-in metadata information to obtain the corresponding plug-in information;

S50, the large model agent service calls the workflow service according to the workflow metadata information to obtain the corresponding workflow information;

S60, the large model agent service calls the database service according to the database metadata information to query the required data;

S70. The large model agent service submits the knowledge base information, plug-in information, workflow information and required data, as well as the prompt words and large language model metadata obtained from the metadata service to the large language model for reasoning, and finally returns the answer to the end user.

It should be noted that in step S10, the interaction between the terminal user and the agent development platform can be text or voice. That is, the terminal user can automatically understand and answer questions based on the configured large language model, prompt (instructions), tool plug-ins, processes and knowledge base during the text or voice interaction with the agent.

To summarize, the present application configures the metadata information of the intelligent agent, including the role definition of the intelligent agent, the knowledge base used, and the called large language model, so as to intercept the process of each user submitting an instruction, dynamically query the corresponding information based on the configured metadata information, and submit the queried information and questions to the configured large language model to infer the final answer.

It should be noted that, although the above is introduced as an example, those skilled in the art will understand that the present disclosure should not be limited thereto. In fact, users can flexibly set according to actual application scenarios, as long as the technical functions of the present application can be realized according to the above technical methods.

Example 3

Based on the implementation principle of Example 1, the present application provides, on the other hand, a device for implementing the configuration method of the large model application agent described in any one of the above items, including:

The instruction receiving module is configured to receive the agent configuration instruction; and the agent configuration instruction is obtained based on the web front-end configuration;

The metadata information configuration module is configured to configure the corresponding agent metadata information according to the agent configuration instruction and store it in the metadata service; and the agent metadata information includes knowledge base metadata information, plug-in metadata information, workflow metadata information and database metadata information;

The intelligent agent creation module is configured to call the large model intelligent agent service to create an intelligent agent framework, and bind the intelligent agent metadata information with the intelligent agent framework to create an intelligent agent.

Obviously, those skilled in the art should understand that all or part of the processes in the above-mentioned embodiment method can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned control methods. The modules or steps of the present invention can be implemented by a general-purpose computing device. They can be concentrated on a single computing device or distributed on a network composed of multiple computing devices. Optionally, they can be implemented by program codes executable by computing devices, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules respectively, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. In this way, the present invention is not limited to any specific combination of hardware and software.

Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment method can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned control methods. Among them, the storage medium can be a disk, an optical disk, a read-only memory (ROM), a random access memory (RAM), a flash memory (Flash Memory), a hard disk (Hard Disk Drive, abbreviated as: HDD) or a solid-state drive (SSD), etc.; the storage medium can also include a combination of the above-mentioned types of memory.

Example 4

Furthermore, in another aspect, the present application provides a control system, comprising:

processor;

a memory for storing processor-executable instructions;

Wherein, the processor is configured to implement any of the above-mentioned methods for configuring the large model application agent when executing the executable instructions.

The control system of the disclosed embodiment includes a processor and a memory for storing processor executable instructions, wherein the processor is configured to implement any of the above-mentioned configuration methods of the large model application agent when executing the executable instructions.

Here, it should be noted that the number of processors can be one or more. At the same time, in the control system of the embodiment of the present disclosure, an input device and an output device may also be included. Among them, the processor, memory, input device and output device may be connected through a bus or in other ways, which are not specifically limited here.

The memory, as a computer-readable storage medium, can be used to store software programs, computer executable programs and various modules, such as the program or module corresponding to the configuration method of the large model application agent in the embodiment of the present disclosure. The processor executes various functional applications and data processing of the control system by running the software programs or modules stored in the memory.

The input device can be used to receive input numbers or signals. The signal can be a key signal related to user settings and function control of the device/terminal/server. The output device can include a display device such as a display screen.

Example 5

On the other hand, the present application provides a non-volatile computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implements any of the above-mentioned methods for configuring a large model application agent.

The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims (10)

1. A configuration method for a large model application agent, characterized in that it includes the following steps: Receiving agent configuration instructions; wherein the agent configuration instructions are obtained based on the web front-end configuration; According to the agent configuration instruction, corresponding agent metadata information is configured and stored in the metadata service; and the agent metadata information includes knowledge base metadata information, plug-in metadata information, workflow metadata information and database metadata information; The large model agent service is called to create an agent framework, and the agent metadata information is bound to the agent framework to create an agent. 2. The method for configuring a large model application agent according to claim 1, characterized in that the corresponding agent metadata information is configured according to the agent configuration instruction and stored in the metadata service, comprising: Calling the knowledge base service according to the agent configuration instruction, and storing the knowledge document in the knowledge base service in a preset manner; The metadata service is called, and the knowledge base metadata information corresponding to the knowledge document is stored in the metadata service. 3. The configuration method of a large model application agent according to claim 2, characterized in that the preset storage method comprises: After the knowledge document is vectorized through a preset model, it is stored in a corresponding vector knowledge base. 4. The configuration method of a large model application agent according to claim 2, characterized in that the preset storage method further comprises: Keywords of the knowledge in the knowledge document are extracted using text frequency, classification and clustering algorithms, and the knowledge and the keywords are stored in a corresponding search engine. 5. The method for configuring a large model application agent according to claim 1, wherein the configuring corresponding agent metadata information according to the agent configuration instruction and storing it in a metadata service further comprises: Call the plug-in service according to the agent configuration instruction, configure the corresponding plug-in, and persist it in the plug-in service; The metadata service is called, and plug-in metadata information corresponding to the plug-in persistence is stored in the metadata service. 6. The method for configuring a large model application agent according to claim 1, characterized in that the step of configuring corresponding agent metadata information according to the agent configuration instruction and storing it in a metadata service further comprises: Calling database services according to the agent configuration instructions to create databases and data tables; The metadata service is called, and database metadata information corresponding to the database and the data table is stored in the metadata service. 7. The method for configuring a large model application agent according to claim 1, wherein the configuring corresponding agent metadata information according to the agent configuration instruction and storing it in a metadata service further comprises: Calling the workflow service according to the agent configuration instruction to configure the workflow; The metadata service is called, and the workflow metadata information corresponding to the workflow is stored in the metadata service. 8. The configuration method of the large model application agent according to claim 7 is characterized in that the workflow includes several working nodes for connecting to corresponding third-party business systems. 9. The configuration method of the large model application agent according to any one of claims 1-8 is characterized in that when calling the large model agent service to create an agent framework, guide words are configured based on the web front end, and the guide words are matched to the corresponding large language model in a preset manner. 10. A device for implementing the configuration method of a large model application agent according to any one of claims 1 to 9, characterized in that it comprises: An instruction receiving module is configured to receive an agent configuration instruction; and the agent configuration instruction is obtained based on the web front-end configuration; A metadata information configuration module is configured to configure corresponding agent metadata information according to the agent configuration instruction and store it in the metadata service; and the agent metadata information includes knowledge base metadata information, plug-in metadata information, workflow metadata information and database metadata information; The agent creation module is configured to call the large model agent service to create an agent framework and claim The agent metadata information is bound to the agent framework to create an agent.