AIAgent DNS-based agent discovery method and system, and AIAgent DNS-based agent calling method and system

Abstract

The invention discloses an agent discovery method and system and an agent calling method and system based on an AIAgent DNS (Domain Name Server). The discovery method comprises the following steps: receiving a request containing a query intention text by an AIAgent DNS server, vectorizing the request, performing semantic matching in a pre-stored agent vector database, and returning an address of a matched agent; the calling method comprises the steps that a model browser client sends a user task to a large language model, receives and analyzes a structured query instruction generated by the large language model to obtain a query intention, requests an AIAgentDNS server and obtains an agent address, and then calls an agent service. According to the scheme, platform locking is avoided through decentralized registration, full automation from task planning to service calling is achieved through prompt word engineering and semantic matching, manual configuration is replaced, and efficient discovery and calling of massive agents are supported.

Description

A method, system, and invocation method and system for agent discovery based on AIAgentDNS Technical Field

[0001] This invention relates to the field of interdisciplinary technology of artificial intelligence and computer networks, and in particular to an intelligent agent discovery method, system, and invocation method and system based on AIAgentDNS. Background Technology

[0002] With the rapid development of Large Language Model (LLM) capabilities, integrating external tools and services through agents to expand their capabilities has become a mainstream paradigm for AI application implementation. ModelContext Protocol (MCP), as an emerging protocol, aims to standardize the interface between agents and LLMs, reducing integration complexity.

[0003] In developing the existing technology, the inventors discovered that agent service providers are forced to publish their services to third-party commercial hosting platforms for registration and distribution. This model subjectes providers to platform rules, depriving them of the autonomy to disseminate and control their services, and posing a risk of platform lock-in. Before using LLM to perform tasks, end users must manually search for and filter suitable agents from the hosting platform's catalog, and then download, install, and configure them locally. As shown in Figure 1, this process is cumbersome and technically demanding, severely hindering the automation and widespread adoption of agent networks, and preventing LLM from dynamically and freely discovering and utilizing external agent resources.

[0004] Therefore, this application provides a decentralized network solution that enables free registration, automation, and accurate discovery of intelligent agents, so as to truly unleash the potential of the AI ​​intelligent agent ecosystem. Summary of the Invention

[0005] This application provides a decentralized network solution that enables free registration, automation, and accurate discovery of intelligent agents, thereby truly unleashing the potential of the AI ​​intelligent agent ecosystem.

[0006] Specifically, an agent discovery method based on AIAgentDNS includes the following steps: receiving an agent query request from a model browser, the request containing query intent text; converting the query intent text into a query vector; performing semantic similarity matching between the query vector and each functional description vector in a pre-stored agent vector database, the agent vector database storing functional description vectors of multiple registered agents and their access addresses; and returning the access address of at least one target agent whose matching degree meets preset conditions to the model browser.

[0007] Furthermore, the agent vector database is pre-established by: obtaining standardized description files provided by multiple agent services; parsing the functional description text of the corresponding agent from each description file; converting each functional description text into a functional description vector; and associating and storing each functional description vector with the access address of the corresponding agent service to construct the agent vector database.

[0008] Furthermore, the standardized description files are stored in a preset standard path under the domain name of each intelligent agent service provider.

[0009] This application also provides an agent discovery system based on AIAgentDNS, comprising: a request receiving module for receiving agent query requests from a model browser, the request containing query intent text; a vectorization module for converting the query intent text into query vectors; a semantic matching module for performing semantic similarity matching between the query vectors and functional description vectors in a pre-stored agent vector database, the agent vector database storing functional description vectors of multiple registered agents and their access addresses; and an address return module for returning the access address of at least one target agent whose matching degree meets preset conditions to the model browser.

[0010] The method also provides an agent invocation method based on AIAgentDNS, applied to a model browser client, comprising the following steps: receiving a natural language task input by a user and sending the task to a large language model; receiving a response returned by the large language model, the response containing a structured query instruction generated based on task decomposition for querying external agents; parsing the query intent text from the structured query instruction; sending the query intent text to an AIAgentDNS server to request access addresses of at least one target agent; and invoking the corresponding agent service to execute the task based on the access address of the target agent returned from the AIAgentDNS server.

[0011] Furthermore, the structured query instruction is generated by the large language model based on preset system prompt words. The system prompt words are configured to guide the large language model to output a predefined query tool call instruction when external information or services are needed.

[0012] Furthermore, the predefined query tool invocation command is an XML format command containing the query intent.

[0013] Furthermore, the invocation of the corresponding intelligent agent service specifically involves: initiating a request to the access address according to the Model Context Protocol (MCP), and returning the execution result to the large language model to continue the task flow.

[0014] The system also provides an AIAgentDNS-based intelligent agent invocation system, comprising: a task receiving module for receiving a natural language task input by a user and sending the task to a large language model; an instruction parsing module for receiving a response returned by the large language model and parsing the query intent text from the structured query instructions contained in the response; a query request module for sending the query intent text to the AIAgentDNS server to request a matching intelligent agent access address; and an intelligent agent invocation module for invoking the corresponding intelligent agent service to execute the task based on the access address of the target intelligent agent returned from the AIAgentDNS server.

[0015] This application provides a method, system, and invocation method and system for intelligent agent discovery based on AIAgentDNS, aiming to solve the key problems of existing intelligent agent ecosystems relying on centralized platforms and requiring manual configuration. This solution constructs a decentralized AIAgentDNS service, enabling intelligent agent providers to complete standardized registration under their own domain names, fundamentally avoiding platform lock-in. Based on this, it guides a large language model to automatically generate structured query instructions through prompt word engineering, achieving automated triggering from "task planning" to "service scheduling." Furthermore, the system utilizes vectorization and semantic matching technologies to accurately map user natural language intent to registered intelligent agents, completely replacing the traditional cumbersome manual search and configuration process. In addition, the design based on vector retrieval and a distributed architecture enables the system to support the rapid registration, discovery, and invocation of massive amounts of intelligent agent services, possessing excellent high-concurrency processing capabilities and horizontal scalability potential, providing a complete implementation solution for open and automated intelligent agent networks. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments of this application and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 is a schematic diagram of the use of MCP intelligent agents in the prior art.

[0017] Figure 2 is a schematic diagram of the system architecture of an embodiment of the present invention.

[0018] Figure 3 is a schematic diagram of the agent discovery process according to an embodiment of the present invention.

[0019] Figure 4 is a schematic diagram of the process of matching AIAgentDNS query intent with an intelligent agent according to an embodiment of the present invention.

[0020] Figure 5 is a timing diagram of the interaction between the model browser and AIAgentDNS according to an embodiment of the present invention.

[0021] Figure 6 is a flowchart of an intelligent agent discovery and invocation method according to an embodiment of the present invention.

[0022] Figure 7 is a structural block diagram of an intelligent agent discovery system (AIAgentDNS server) according to an embodiment of the present invention.

[0023] Figure 8 is a structural block diagram of an intelligent agent invocation system (model browser client) according to an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] The following is a description of some of the English abbreviations and key terms involved in the embodiments of this application: LLM: Large Language Model, which refers to a deep learning model with a large number of parameters that can understand and generate natural language.

[0026] Agent: A software entity capable of sensing the environment, making decisions, and performing specific tasks to achieve a goal. In this application, it specifically refers to a service that conforms to the MCP protocol and can provide specific functions or data for an LLM.

[0027] MCP: Model Context Protocol, an open protocol for standardizing interactions between agents and LLMs.

[0028] AIAgentDNS: The core component for agent service registration and discovery proposed in this application, analogous to the Internet Domain Name System (DNS).

[0029] Vector database: A database specifically designed for storing, indexing, and retrieving high-dimensional vector data (embedded vectors), supporting fast similarity-based queries.

[0030] Please refer to Figure 2, which illustrates the overall system architecture of an intelligent agent network based on AIAgentDNS according to an embodiment of the present invention. The system adopts a decentralized design and mainly includes a model browser client, an AIAgentDNS server, and a distributed intelligent agent service cluster. This architecture provides the physical foundation for implementing the "discovery" and "invocation" methods and systems.

[0031] This section details the agent discovery method claimed in claim 1 and how the corresponding system is specifically implemented. The method is executed by the AIAgentDNS server shown in Figure 7.

[0032] The agent receives query requests from the model browser, which contain query intent text. The request receiving module of the AIAgentDNS server (Figure 7) receives HTTP requests from the model browser client via an open REST API interface (e.g., / api / query). The body of the request contains the query intent text expressed in natural language. For example, the request body might be: {"query": "Provide cycling route planning services from Suzhou to Beijing, including detailed route, distance, estimated time, and information on facilities along the way"}.

[0033] The query intent text is converted into a query vector.

[0034] The vectorization module (Figure 7) calls a pre-trained word embedding model (such as text-embedding-ada-002 or Sentence-BERT) after receiving the query intent text. This module inputs the text into the model, and the model outputs a high-dimensional floating-point array of fixed dimensions (such as 768 or 1536 dimensions), which is the query vector. This process transforms the unstructured semantic query into a numerical representation that can be used for mathematical calculations and similarity comparisons.

[0035] In the pre-stored agent vector database, perform semantic similarity matching between the query vector and each functional description vector.

[0036] The semantic matching module (Figure 7) calls the searchTools function. This module takes the query vector generated by S102 as input and compares it with all the functional description vectors pre-stored in the agent vector database. The comparison algorithm usually uses cosine similarity calculation, that is, calculating the cosine value of the angle between the query vector and each functional description vector. This value ranges between [-1, 1], and the larger the value, the more semantically similar the two vectors are. The system calculates all similarity scores and sorts them.

[0037] The agent vector database (as shown in Figure 4) is pre-established by the method described in claim 2, and it stores the functional description vectors of multiple registered agents and their access addresses (URLs). The structure of each record can be abstracted as: {index ID, functional description vector, agent access address, functional description text}.

[0038] The access address of at least one target agent whose matching degree meets the preset conditions is returned to the model browser.

[0039] The address return module (Figure 7) selects one or more records with the highest scores (e.g., greater than the threshold of 0.7) as matching results based on the similarity ranking results obtained in S103. Subsequently, this module extracts the agent access address (URL) from these records and encapsulates it into structured response data, returning it to the model browser client. The return format is, for example: {"agents": [{"url": "https: / / mcp.example.ai / sse","description": "Cycling route planning for planning cycling commuting schemes...","request_id": "a1b2c3d4"}]} This completes the full discovery process defined in claim 1.

[0040] The agent vector database is pre-built in the following way: standardized description files provided by multiple agent services are obtained; the functional description text of the corresponding agent is parsed from each description file; each functional description text is converted into a functional description vector; and each functional description vector is associated with and stored with the access address of the corresponding agent service to construct the agent vector database. The specific steps are as follows: The registration management module of AIAgentDNS (Figure 7) receives registration information submitted by the service provider through an API (such as POST / api / registerMcp), which includes the URL of the manifest file, for example, https: / / mcp.example.ai / .well-known / mcp.json. This file is stored in the default standard path ( / .well-known / mcp.json) of the service provider's own domain; the system obtains this JSON file and parses its content. This file contains a tools array, where each tool has a description field. The system extracts these descriptive texts, such as "Cycling route planning is used to plan cycling commuting schemes..."; the system uses the same word embedding model to transform each functional description text into a functional description vector, and then stores the vector, the corresponding tool name, the agent service root address (baseUrl), and the original descriptive text in the vector database.

[0041] As shown in Figure 7, the AIAgentDNS-based agent discovery system specifically includes a request receiving module, a vectorization module, a semantic matching module, and an address return module. Furthermore, the figure also includes a vector database module and a registration management module for database creation, constituting a complete system implementation.

[0042] This section details the claimed agent invocation method and how the corresponding system is specifically implemented. The method is executed by the model browser client shown in Figure 8.

[0043] Receive natural language tasks input by the user and send the tasks to the large language model.

[0044] The task receiving module of the model browser client (Figure 8) receives user input, such as "cycling route from Suzhou to Beijing", through a graphical user interface (GUI) or API. Subsequently, the large language model interface module (Figure 8) sends the task text, along with the system configuration context and parameters, to the backend large language model (such as GPT-4, Claude, etc.) service via an HTTP request.

[0045] The system receives a response from the large language model, which contains a structured query instruction generated based on task decomposition for querying external intelligent agents; and parses the query intent text from the structured query instruction.

[0046] The large language model decomposes tasks based on its system prompts (which are key to achieving the "predefined system prompts" in claim 6). These prompts explicitly define the specific query_agent tool the model must use when external information is needed, and specify its XML output format (corresponding to the "predefined query tool invocation instructions" in claim 7). For example, the LLM might return: <thinking>Users need cycling routes and should use map services.< / thinking> <query_agent> <information> We offer cycling route planning services from Suzhou to Beijing, including detailed route information, distance, estimated travel time, and information on facilities along the way.< / information> < / query_agent> The model browser's instruction parsing module (Figure 8) will recognize the instructions in the response.<query_agent> Tags (XML format instructions), and extract them. <information>The text content within the tag is the pure text of the query intent.

[0047] The query intent text is sent to the AIAgentDNS server to request access addresses for at least one target agent.

[0048] The query request module (Figure 8) constructs the query intent text parsed in S202 into an HTTP request (e.g., a POST request to https: / / aiagentdns.example.com / api / query) and sends it to the pre-configured AIAgentDNS server address, thereby triggering the discovery process described in claim 1.

[0049] Based on the access address of the target agent returned from the AIAgentDNS server, the corresponding agent service is invoked to execute the task.

[0050] Upon receiving the address returned by AIAgentDNS (e.g., https: / / mcp.example.ai / sse), the agent invocation module (Figure 8) establishes a connection with that address (e.g., an SSE connection or an HTTP call) according to the Model Context Protocol (MCP) (a feature of claim 8). Then, a tool invocation request based on the specific requirements of the MCP is constructed. For example, the `maps_direction_bicycling` tool is invoked, passing the `origin` and `destination` parameters. After execution, the agent returns a result, which can be returned to the LLM by the result integration module (Figure 8) for further processing or presented directly to the user.

[0051] The AIAgentDNS-based intelligent agent invocation system comprises various modules that execute the aforementioned invocation methods. As shown in Figure 8, these modules specifically include: a task receiving module, an instruction parsing module, a query request module, and an intelligent agent invocation module. Furthermore, it includes a large language model interface module, a result integration module, and a configuration management module, collectively forming a complete client system.

[0052] Combining the sequence diagram in Figure 5 and the overall flowchart in Figure 6, a complete application example of this application's solution is as follows: The user inputs "cycling route from Suzhou to Beijing" into the model browser. The model browser sends the task to the LLM, and the LLM generates a system containing prompts based on the prompts.<query_agent> The command response: The model browser parses the command, extracts the intent text, and then sends a query to the AIAgentDNS server. AIAgentDNS uses vectorization and semantic matching to find the map agent from a pre-built database and returns its address. Finally, the model browser uses this address to call the agent via the MCP protocol to obtain the route result.

[0053] In summary, this specific implementation clearly demonstrates, through the core component AIAgentDNS, the decentralized registration of intelligent agent services, semantic-based accurate discovery, and automated collaborative invocation with LLM, effectively solving the platform lock-in and manual configuration problems mentioned in the background technology.

[0054] It should be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0055] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.< / information>

Claims

1. An agent discovery method based on AIAgentDNS, applied to an AIAgentDNS server, characterized in that, The process includes the following steps: receiving a query request from a model browser, the request containing query intent text; converting the query intent text into a query vector; performing semantic similarity matching between the query vector and each functional description vector in a pre-stored agent vector database, the agent vector database storing functional description vectors of multiple registered agents and their access addresses; and returning the access address of at least one target agent whose matching degree meets preset conditions to the model browser.

2. The method according to claim 1, characterized in that, The agent vector database is pre-established by: obtaining standardized description files provided by multiple agent services; parsing the functional description text of the corresponding agent from each description file; converting each functional description text into a functional description vector; and associating and storing each functional description vector with the access address of the corresponding agent service to construct the agent vector database.

3. The method according to claim 2, characterized in that, The standardized description files are stored in the preset standard paths of each agent service provider's own domain name.

4. An agent discovery system based on AIAgentDNS, characterized in that, include: The request receiving module is used to receive intelligent agent query requests from the model browser, the requests containing query intent text; The vectorization module is used to convert the query intent text into a query vector; the semantic matching module is used to perform semantic similarity matching between the query vector and each functional description vector in a pre-stored agent vector database, wherein the agent vector database stores the functional description vectors of multiple registered agents and their access addresses; the address return module is used to return the access address of at least one target agent whose matching degree meets the preset conditions to the model browser.

5. An agent invocation method based on AIAgentDNS, applied to a model browser client, characterized in that: The method includes: receiving a natural language task input by a user and sending the task to a large language model; receiving a response returned by the large language model, the response containing a structured query instruction generated based on task decomposition for querying an external intelligent agent; parsing the query intent text from the structured query instruction; sending the query intent text to an AIAgentDNS server to request access addresses of at least one target intelligent agent; and, based on the access addresses of the target intelligent agents returned from the AIAgentDNS server, invoking the corresponding intelligent agent service to execute the task.

6. The method according to claim 5, characterized in that, The structured query command is generated by the large language model based on preset system prompt words. The system prompt words are configured to guide the large language model to output a predefined format query tool call command when external information or services are needed.

7. The method according to claim 6, characterized in that, The predefined query tool invocation command is an XML format command containing the query intent.

8. The method according to claim 5, characterized in that, The specific steps of calling the corresponding intelligent agent service are as follows: according to the Model Context Protocol (MCP), a request is sent to the access address, and the execution result is returned to the large language model to continue the task flow.

9. An intelligent agent invocation system based on AIAgentDNS, characterized in that, include: The task receiving module is used to receive natural language tasks input by the user and send the tasks to the large language model; The instruction parsing module is used to receive the response returned by the large language model and parse the query intent text from the structured query instructions contained in the response; The query request module is used to send the query intent text to the AIAgentDNS server to request the matching agent access address; the agent invocation module is used to invoke the corresponding agent service to execute the task based on the access address of the target agent returned from the AIAgentDNS server.

Images