Capability invocation method, apparatus and system

Abstract

The application relates to a capability calling method, device and system. The method comprises the following steps: receiving a task request sent by a request party for a target task; in response to the task request, at least one target atomic capability for executing the target task is matched from a full-amount atomic capability list pre-established by an atomic capability platform; the full-amount atomic capability list contains capability information of a plurality of atomic capabilities; a capability call is initiated to a provider of the target atomic capability, and a capability calling result is received; according to a calling protocol type associated with the task request, the capability calling result is adapted into a corresponding response format and returned to the request party. The application realizes full-process automatic processing from receiving a task, capability matching, protocol adaptation to result returning, can support coherent execution of a complex task across a plurality of atomic capabilities, and thus improves the coherence of user operation and the use experience while improving the service calling efficiency and reliability.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus and system for invoking capabilities. Background Technology

[0002] With the rapid development of artificial intelligence (AI) technology, mobile phones and other terminal devices have expanded their original functions to include AI services such as voice interaction, image recognition, and text translation by accessing large model capabilities. These services require users to actively initiate commands.

[0003] In related technologies, terminal devices typically integrate third-party services via API interfaces, pre-configuring these services as fixed service cards. The system (e.g., a service distribution framework) recognizes user intent based on input, matches preset tags, and then invokes the corresponding service, ultimately presenting it as a service card on the application or desktop interface. However, this static card display requires users to actively browse, filter, and trigger services, making the interaction cumbersome. This approach lacks proactive responsiveness, limiting further improvements in the terminal's intelligent experience. Summary of the Invention

[0004] In view of this, this application provides a capability invocation method, apparatus and system to solve at least one problem existing in the prior art.

[0005] Firstly, a capability invocation method is provided, executed by an atomic capability platform, the method comprising: The receiving party sends a task request for the target task. In response to the task request, at least one target atomic capability for performing the target task is matched from the full list of atomic capabilities pre-established by the atomic capability platform; wherein, the full list of atomic capabilities contains capability information of multiple atomic capabilities; Initiate a capability call to the provider of the target atomic capability and receive the capability call result; Based on the invocation protocol type associated with the task request, the capability invocation result is adapted to the corresponding response format and returned to the requester.

[0006] In conjunction with the first aspect, in one possible implementation, the plurality of atomic capabilities includes operating system atomic capabilities, AI atomic capabilities, and third-party service capabilities; the full list of atomic capabilities is generated by registering atomic capabilities from different providers using a standardized data structure.

[0007] In conjunction with the first aspect, in one possible implementation, the task request originates from an application running on the user terminal, and the calling protocol type associated with the task request is an application programming interface (API) protocol. Alternatively, the task request may originate from an AI agent, and the associated calling protocol type is a model context protocol.

[0008] In conjunction with the first aspect, in one possible implementation, the step of matching at least one target atomic capability for performing the target task from a pre-established full list of atomic capabilities of the atomic capability platform in response to the task request includes: In response to the task request, determine the target task corresponding to the task request; Based on the target task and the application information pre-acquired on the user terminal, atomic capabilities are matched in the full list of atomic capabilities to determine the target atomic capability; The user terminal is either the terminal where the application that initiated the task request is located, or the terminal where the application that initiated the task request is located through an AI agent.

[0009] In conjunction with the first aspect, in one possible implementation, the step of matching atomic capabilities in the full list of atomic capabilities based on the target task and pre-acquired application information installed on the user terminal to determine the target atomic capability includes: Based on the description information of the target task, a preliminary screening is performed in the full list of atomic capabilities to obtain a set of candidate atomic capabilities related to the target task. The description information of the target task, the application information, and the description information of each atomic capability in the candidate atomic capability set are input into the large language model; Based on the matching results output by the large language model, the final target atomic capability is determined.

[0010] In conjunction with the first aspect, in one possible implementation, the method further includes: The atomic capability platform combines multiple atomic capabilities from the full list of atomic capabilities into composite atomic capabilities according to preset arrangement rules. The process of initiating a capability call to the provider of the target atomic capability and receiving the capability call result includes: When the target atomic capability includes a composite atomic capability, the atomic capability platform determines each atomic capability that is combined into the composite atomic capability and the corresponding arrangement rules; According to the determined orchestration rules, a capability call is initiated to the provider corresponding to each atomic capability included in the composite atomic capability, and the capability call result is received.

[0011] In conjunction with the first aspect, in one possible implementation, the atomic capability platform includes an atomic capability platform server deployed on the cloud side and an atomic capability platform client deployed on the edge side; The process of initiating a capability invocation to the provider corresponding to the target atomic capability includes: If the target atomic capability is an edge capability, then the atomic capability platform server initiates a local call to the target atomic capability through the atomic capability platform client; If the target atomic capability includes cloud-side capabilities, then the atomic capability platform server initiates a remote call to the corresponding cloud service interface for the target atomic capability.

[0012] In conjunction with the first aspect, in one possible implementation, the task request is initiated by an AI agent for a target task; the method further includes: A list of matching atomic capabilities formed by at least one target atomic capability is returned to the AI ​​agent; The initiation of a capability invocation to the provider of the target atomic capability includes: Receive capability invocation requests initiated by the AI ​​agent based on the atomic capability matching list; In response to the capability invocation request, a capability invocation is initiated to the provider corresponding to at least one target atomic capability.

[0013] In conjunction with the first aspect, in one possible implementation, the AI ​​agent includes a master agent and multiple slave agents; The step of returning the atomic capability matching list, formed by at least one target atomic capability, to the AI ​​agent includes: In response to a task request initiated by the master agent, the atomic capability matching list is returned to the master agent so that the master agent can decompose the task based on the atomic capability matching list, generate at least one sub-task, and distribute it to the target slave agent among the plurality of slave agents. Receiving the capability invocation request initiated by the AI ​​agent based on the atomic capability matching list includes: Receive the capability invocation request initiated by the target from the agent based on the subtask, wherein the capability invocation request carries the target atomic capability information specified by the subtask.

[0014] Secondly, a capability mobilization device is provided for use on an atomic capability platform, the device comprising: The request receiving unit is used to receive task requests sent by the requester for the target task. The request processing unit is configured to, in response to the task request, match at least one target atomic capability for executing the target task from a pre-established full list of atomic capabilities in the atomic capability platform; wherein the full list of atomic capabilities contains capability information of multiple atomic capabilities; The execution unit is used to initiate a capability call to the provider of the target atomic capability and to receive the capability call result; The result return unit is used to adapt the capability call result into the corresponding response format and return it to the requester according to the call protocol type associated with the task request.

[0015] Thirdly, a capability invocation system is provided, the system comprising: an atomic capability platform, at least one requester, and at least one capability provider; The atomic capability platform is used to execute capability invocation methods provided by any possible implementation of the first aspect; The requesting party is used to send a task request for the target task to the atomic capability platform and receive the capability invocation result returned by the atomic capability platform; The capability provider is used to respond to the capability call request from the atomic capability platform, execute the corresponding atomic capability, and return the result.

[0016] Fourthly, an electronic device is provided, including a processor configured to invoke instructions such that the terminal device, when executed, implements the capability invocation method as described in any of the first aspects.

[0017] Fifthly, a storage medium is provided having an executable program stored thereon, which, when executed by a processor, implements the capability invocation method as described in any of the first aspects.

[0018] This application provides a capability invocation method, apparatus, and system. The method is executed by an atomic capability platform. It receives a task request from a requester for a target task. In response to the task request, it matches at least one target atomic capability from a pre-established full list of atomic capabilities on the platform. This full list contains capability information for multiple atomic capabilities, enabling centralized management of distributed service capabilities and supporting on-demand combination and automatic invocation of multiple atomic capabilities. This allows the atomic capability platform to automatically and accurately match the required atomic capabilities based on task requirements, effectively alleviating the cumbersome operation of users actively searching for and triggering specific service cards in existing solutions. By initiating an invocation to the provider of the target atomic capability and receiving the result, automatic service execution is achieved. Furthermore, based on the invocation protocol type associated with the task request, the capability invocation result is adapted to the corresponding response format and returned to the requester, achieving adaptive encapsulation of the invocation result. Therefore, in this application's solution, the requester can initiate a task request in natural language or a structured task description without specifying a specific API interface. The atomic capability platform can automatically parse the user's task intent, select suitable atomic capabilities from the full list of atomic capabilities, initiate a call to the corresponding provider, and return the call result. This approach realizes the transformation from passive response to proactive service, improving the flexibility of service calls and user experience. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of a system architecture based on an atomic capability platform, as shown in one embodiment of this application; Figure 2 This is one of the flowcharts illustrating a capability invocation method according to an embodiment of this application; Figure 3 for Figure 2 The flowchart of step S104 is shown below; Figure 4 for Figure 3 The flowchart of step S1042 is shown below; Figure 5 This is a second schematic flowchart illustrating a capability invocation method according to an embodiment of this application; Figure 6 This is a third schematic flowchart illustrating a capability invocation method according to an embodiment of this application; Figure 7 This is a fourth schematic flowchart illustrating a capability invocation method according to an embodiment of this application; Figure 8 This is a schematic diagram of the capability invocation device shown in one embodiment of this application. Detailed Implementation

[0020] To make the technical solution and beneficial effects of this application more apparent and understandable, a detailed description is provided below by listing specific embodiments. The accompanying drawings are not necessarily drawn to scale, and local features may be enlarged or reduced to more clearly show the details of the local features; unless otherwise defined, the technical and scientific terms used herein have the same meanings as those in the technical field to which this application pertains.

[0021] The embodiments in this application are not exhaustive, but merely illustrative of some embodiments, and are not intended to limit the scope of protection of this disclosure. Unless otherwise specified, each step in a particular embodiment can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a particular embodiment can also be implemented as an independent embodiment, and the order of the steps in a particular embodiment can be arbitrarily interchanged. Furthermore, the optional implementation methods in a particular embodiment can be arbitrarily combined; moreover, the embodiments can be arbitrarily combined, for example, some or all steps of different embodiments can be arbitrarily combined, and a particular embodiment can be arbitrarily combined with the optional implementation methods of other embodiments.

[0022] In each embodiment of this application, unless otherwise specified or in case of logical conflict, the terminology and / or descriptions of the embodiments are consistent and can be referenced by each other. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0023] In the description of the embodiments of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0024] It should be noted that the information (including but not limited to user device information) and data (including but not limited to data used for analysis, stored data, and displayed data) collected in this public disclosure are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of this data all comply with the relevant laws, regulations, and standards of the relevant regions, and necessary confidentiality measures have been taken. This does not violate public order and good morals, and corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or organizations. Before obtaining relevant information, a request to obtain the information must be sent to the aforementioned user or organization through the interface, and the relevant information will be obtained only after the aforementioned user or organization has provided their consent.

[0025] This application provides a capability invocation method executed by an atomic capability platform. This atomic capability platform can be deployed in the cloud or on a terminal device, or it can adopt a system architecture combining the end and cloud. It is used for unified management of various access atomic capabilities, accurately matching atomic capabilities according to user needs, and establishing an interaction link between the client and the intelligent agent, enabling unified scheduling and efficient invocation of atomic capabilities.

[0026] In this application, an atomic capability refers to the smallest service unit with a specific function that can be independently invoked. It possesses the characteristic of being indivisible and atomic, and supports on-demand combination to support complex service processes. Each atomic capability encapsulates complete functional logic and can be directly invoked by upper-layer applications or intelligent agents. For example, obtaining the current geographical location, translating text from Chinese to English, and invoking a camera to complete image acquisition can all be considered as independent atomic capabilities.

[0027] Figure 1 A schematic diagram of a system architecture based on an atomic capability platform is shown. This architecture includes capability providers, atomic capability platforms, and requesters.

[0028] In some examples, the capability provider may include multiple capability sources, each providing different types of atomic capabilities, such as: OS atomic capabilities provided by the operating system (OS), AI atomic capabilities for AI model application scenarios, and third-party service capabilities provided by third-party service providers.

[0029] In some examples, atomic capabilities can be categorized into edge-side capabilities and cloud-side capabilities based on the deployment location of the atomic capability provider. Edge-side capabilities refer to atomic capabilities whose functionality depends entirely on the local hardware and software resources of the terminal. Examples include accessing the phone's camera, reading the local address book, and running lightweight AI models using the terminal's NPU. Cloud-side capabilities refer to atomic capabilities whose functionality depends on the resources of a remote cloud server. Examples include calling large language models in the cloud that require significant computing power and accessing third-party internet API services.

[0030] In some examples, the atomic capability platform is used for unified management of the recorded atomic capabilities, and its functions include one or more of the following: Data format standardization: The interface protocols and parameters of heterogeneous atomic capabilities are standardized to complete capability input using a standardized data structure. The atomic capability platform supports multiple access methods, including API and MCP (Model Context Protocol), and can automatically convert third-party API interfaces into the MCP tool format, proactively adapting to commonly used third-party services. Simultaneously, the platform provides both API and MCP dual-protocol call interfaces, catering to the calling needs of both APP applications and intelligent agents.

[0031] Capability Deployment Management: Controlling the release, version iteration, and decommissioning processes of atomic capabilities.

[0032] Full Capability Table: Centrally stores standardized metadata for all atomic capabilities, which may include one or more of the following capability information: unique capability identifier (ID), capability name, capability type (e.g., OS atomic capability, AI atomic capability, third-party service), function description, API address, deployment location (device / cloud), provider information, etc.

[0033] Capability and access control: Access control is implemented based on the full capability table, including requester authentication, user authorization verification, and compliance interception of calls involving sensitive data or device security, ensuring that the call process is secure and controllable.

[0034] Semantic recognition: Parsing user natural language input and converting it into structured task intent.

[0035] Capability matching: Based on historical call data or semantic understanding of the large model, suitable atomic capabilities are retrieved from the full capability table. In specific implementation, user requests can be encoded by a semantic model and matched with a vector library. A hybrid strategy combining semantic retrieval and keyword retrieval is adopted, and the matching score is dynamically optimized through a weighted / subtractive mechanism. In addition, application information of applications installed on the user's device can be combined to filter the list of related capability candidates. Then, the large model performs intent alignment and precise sorting, and finally outputs the matching capabilities that meet the user's needs.

[0036] Capability orchestration: Supports combining multiple atomic capabilities into a composite capability to customize business flows for complex scenarios, such as creating travel guides and batch image enhancement capabilities.

[0037] Capability routing: Selects the execution path based on factors such as capability deployment location (endpoint or cloud), initiates capability calls to the provider on the endpoint or cloud, and returns the result after execution.

[0038] Capability Invocation: A collaborative interaction link is established between the edge, cloud, and intelligent agents. The atomic capability platform server serves as the unified exit point for capability invocation, responsible for initiating capability invocations to third-party servers or OS application clients and receiving the invocation results back to the requesting party. The atomic capability platform client performs unified scheduling of edge-side atomic capabilities and collaborates with the atomic capability platform server to transmit invocation information. By implementing unified exit control for edge and cloud-side capabilities, the performance and power consumption overhead of the terminal operating system is reduced while ensuring the legality and security of capability invocations.

[0039] In some examples, the atomic capability call requester, such as an agent or an application, can initiate a capability call (e.g., submit a task request) to the atomic capability platform via the AI ​​business gateway. After the atomic capability platform executes the capability call, it returns the call result to the requester through the gateway, forming a complete service loop.

[0040] Understandable Figure 1 The architecture shown is merely an example. The specific module division and deployment location (edge-side, cloud-side, or edge-cloud integrated deployment) of its atomic capability platform can be flexibly adjusted according to actual applications. This application does not impose any specific limitations on this.

[0041] Figure 2 This is a flowchart illustrating a capability invocation method according to an embodiment of this application. See also... Figure 2 The method includes the following steps: S102: Receive the task request sent by the requester for the target task; S104: In response to the task request, match at least one target atomic capability for executing the target task from the full list of atomic capabilities pre-established by the atomic capability platform; wherein, the full list of atomic capabilities contains capability information of multiple atomic capabilities; S106: Initiate a capability call to the provider of the target atomic capability and receive the capability call result; S108: Based on the invocation protocol type associated with the task request, adapt the capability invocation result to the corresponding response format and return it to the requester.

[0042] In step S102, the requester can be an application (APP) installed on the user's terminal device, such as a map application, social application, or shopping application, or it can be an AI agent built based on artificial intelligence (such as a large language model). The task request is used to request the atomic capability platform to execute capability invocation.

[0043] For apps, the invocation methods can be divided into two scenarios: one is that the app can directly submit task requests to the atomic capability platform. For example, when a photo app calls image enhancement capabilities, it can submit a task request through a predefined API interface. The other is that the app itself does not directly construct task requests, but forwards the user's natural language instructions (such as "This photo is too dark, please brighten it") to the AI ​​agent, which then performs intent parsing and task generation.

[0044] For AI agents, after receiving user natural language input and completing intent understanding, they submit structured task requests to the atomic capability platform. These task requests can be encapsulated using the Model Context Protocol (MCP).

[0045] In an exemplary interaction scenario, a user initiates a task request through a voice assistant app on a terminal device. This request is forwarded to an AI agent via an AI business gateway. After the AI ​​agent parses the user's intent, it submits a structured task request to the atomic capability platform through the same gateway. The atomic capability platform executes the capability call and returns the call result layer by layer, which is finally presented to the user by the voice assistant app, forming an end-to-end service loop.

[0046] In step S104, after receiving a task request, the atomic capability platform can determine the corresponding target task by parsing and standardizing the task request. For task requests in natural language form (e.g., from an AI agent), the atomic capability platform extracts the task intent and key parameters through semantic recognition to generate a structured target task description; for structured API requests (e.g., JSON instructions from an app), the atomic capability platform can directly parse predefined fields, extract the operation type and parameters, and form a target task representation; the atomic capability platform can obtain a target task with a unified format after parsing and processing both types of requests.

[0047] The aforementioned semantic recognition can be performed by the semantic recognition module within the atomic capability platform. For example, for the task request "Help me find a nearby Sichuan restaurant," the semantic recognition module parses the "target task" as "search for restaurants." The semantic recognition module can be a rule-based or template-based parser; in a preferred implementation, this module can integrate or invoke a Natural Language Processing (NLP) model or a Large Language Model (LLM) to parse the user's input natural language, identify the task intent, extract key parameters, and understand the user's true needs in conjunction with the context, ultimately outputting a structured task description to provide accurate input for subsequent capability matching.

[0048] Based on the parsed target task, the atomic capability platform can use the entire list of atomic capabilities as the search scope to match one or more functionally compatible target atomic capabilities for the target task. The matching process can employ various strategies. For example, the atomic capability platform can compare keywords in the target task (e.g., "search for restaurants") with the capability information (e.g., capability metadata) of the atomic capabilities. Preferably, semantic vector matching can be used, encoding the task description and capability metadata into vectors respectively, and identifying the most semantically relevant atomic capabilities by calculating their cosine similarity (or Euclidean distance). For example, the atomic capability with the highest similarity (or smallest Euclidean distance) can be selected as the target atomic capability.

[0049] In step S106, after matching the target atomic capability, the atomic capability platform can initiate a capability call request to the corresponding provider based on the call information registered in the list of the target atomic capability (such as API endpoints of third-party services, OS local function interfaces, or cloud-side service addresses). The capability provider can be deployed in the cloud (e.g., third-party services) or on the local terminal device (e.g., OS system services or installed applications). The atomic capability platform can be responsible for parameter assembly, network communication, and security authentication during the capability call process, and receive the capability call result generated by executing the target atomic capability returned by the provider.

[0050] In step S108, the atomic capability platform can perform necessary processing on the received call result (such as format conversion and data integration) and return it to the requester through the same channel used when receiving the request. This entire process allows the requester to obtain the required services without needing to concern themselves with the specific implementation, deployment location, or invocation method of the atomic capability; they can simply use the unified interface provided by the atomic capability platform to access the services in natural language or a structured task description.

[0051] The result of a capability call may include valid data returned when the call is successful, such as translated text or captured images; or it may include error information returned when the call fails, such as error codes or error descriptions, to provide feedback on the execution status of the call process.

[0052] This application provides a capability invocation method executed by an atomic capability platform. In response to a task request sent by a requester for a target task, at least one target atomic capability is matched from a pre-established full list of atomic capabilities on the platform. This full list contains capability information for multiple atomic capabilities, enabling centralized management of distributed service capabilities and supporting on-demand combination and automatic invocation of multiple atomic capabilities. This allows the atomic capability platform to automatically and accurately match the required atomic capabilities based on task requirements, effectively alleviating the cumbersome operation of users actively searching for and triggering specific service cards in existing solutions. By initiating a call to the provider of the target atomic capability and receiving the result, automatic service execution is achieved. Furthermore, based on the invocation protocol type associated with the task request, the capability invocation result is adapted to the corresponding response format and returned to the requester, achieving adaptive encapsulation of the invocation result. Therefore, in this application's solution, the requester can initiate a task request in natural language or a structured task description without specifying a specific API interface. The atomic capability platform can automatically parse the user's task intent, select suitable atomic capabilities from the full list of atomic capabilities, initiate a call to the corresponding provider, and return the call result. This approach realizes the transformation from passive response to proactive service, improving the flexibility of service calls and user experience.

[0053] In some embodiments, the plurality of atomic capabilities includes operating system atomic capabilities, AI atomic capabilities, and third-party service capabilities; the full list of atomic capabilities is generated by registering atomic capabilities from different providers using a standardized data structure.

[0054] In this embodiment, the atomic capability platform defines a set of mandatory standardized encapsulation specifications. All entered atomic capabilities must be described in accordance with these specifications. Capability information may include metadata such as a unique capability identifier, functional description, input / output parameter format, provider type, and API call. The atomic capability platform can register the encapsulated capability descriptions through a management interface or automated interface and continuously update the full capability list.

[0055] Operating system atomic capabilities refer to the native functions exposed by the terminal device's operating system through system APIs, such as basic system services like accessing contacts, calling the camera, obtaining geolocation, and managing calendar events.

[0056] AI atomic capabilities refer to the AI ​​atomic capabilities of AI model application scenarios, which can be deployed on the edge or cloud side, such as image enhancement, style transfer, text generation, speech recognition and other atomic capabilities.

[0057] Third-party service capabilities refer to commercial or public services provided and maintained in the cloud by third-party service providers, such as online map navigation, real-time weather queries, payment gateways, ticket booking, etc., which are made available to the outside world through standard API interfaces.

[0058] For example, taking the "send SMS" OS atomic capability as an example, after standardized encapsulation, its record in the full capability list can include information such as capability ID, function description, input parameter specifications (such as recipient number and message content), output format, provider type ("OS"), and local call interface.

[0059] In this embodiment, a full list of atomic capabilities is generated by registering atomic capabilities from different providers using a standardized data structure. This process transforms original services from different sources into atomic resources with a unified format that can be managed uniformly by the platform, providing a data foundation for subsequent capability matching and capability invocation.

[0060] In some embodiments, the task request originates from an application running on a user terminal, and the calling protocol type associated with the task request is an application programming interface (API) protocol; or, the task request originates from an AI agent, and the calling protocol type associated with the task request is a model context protocol.

[0061] In this embodiment, before the atomic capability platform returns the capability call result to the requester, it needs to perform format adaptation according to the protocol used by the requester to initiate the task request.

[0062] For example, when an application running on a user terminal has a known target capability, it can directly initiate a structured call request (i.e., a task request) to the atomic capability platform via a standard application programming interface (API) protocol (such as RESTful API or gRPC). For instance, when a photo album application calls an image enhancement capability, it can directly submit a JSON instruction containing the operation type and parameters to the atomic capability platform. This calling method is suitable for scenarios where the capability is known. The capability call result is returned to the application by the atomic capability platform according to the response format (such as JSON or Protobuf) agreed upon by the API protocol, thus completing the capability call.

[0063] For example, when the requester of a task is an AI agent, it submits the task request to the atomic capability platform through the Model Context Protocol (MCP). The MCP is a standardized protocol specifically designed for AI agents to interact with external tools. For instance, when an AI agent initiates a task request "check tomorrow's weather in Beijing" through the MCP protocol, the atomic capability platform encapsulates the capability call result according to the format specified in the MCP specification so that the agent can directly parse and use it.

[0064] In this embodiment, the atomic capability platform adapts the format according to the calling protocol type (application interface protocol or model context protocol) associated with the task request and returns the capability call result that conforms to the corresponding protocol specification to the requester. In this way, it is compatible with AI agent calls on the basis of API interface service, without the need for third-party service providers to develop additional MCP tools for encapsulation, which greatly reduces the cost of capability access.

[0065] In some embodiments, such as Figure 3 As shown, in step S104 above, in response to a task request, matching at least one target atomic capability for performing the target task from a pre-established list of all atomic capabilities on the atomic capability platform may include: S1041: In response to a task request, determine the target task corresponding to the task request; S1042: Based on the target task and the application information pre-acquired on the user terminal, match the atomic capabilities in the full list of atomic capabilities to determine the target atomic capability.

[0066] The user terminal is either the terminal where the application that initiated the task request resides, or the terminal where the application that initiated the task request resides through an AI agent. In other words, the application on the user terminal (such as a voice assistant app) can directly initiate a task request to the atomic capability platform, or it can initiate a task request to the atomic capability platform through an AI agent.

[0067] To improve matching accuracy, the atomic capability platform uses a list of applications installed on the user's device as context when matching capabilities. Application information can be obtained through privacy-compliant interfaces pre-installed in the device's operating system (such as Package Manager for Android and Limited App Tracking for iOS) after user authorization.

[0068] In some examples, when performing capability matching, priority can be given to matching based on the terminal-side atomic capabilities provided by applications already installed on the user's terminal. These terminal-side atomic capabilities are then ranked according to their matching degree with the target task, and the terminal-side atomic capability with the highest ranking is determined as the target atomic capability. If no terminal-side atomic capability matches the target task, then cloud-side atomic capabilities are matched. The terminal-side atomic capabilities include OS atomic capabilities and terminal-side AI atomic capabilities. The cloud-side atomic capabilities include third-party atomic capabilities and cloud-side AI atomic capabilities.

[0069] For example, when a user requests "navigation to the company," semantic parsing determines the target task as "execute route planning and navigation." The atomic capability platform matches the "map navigation" atomic capability with matching functionality from the full list of atomic capabilities. It then filters and sorts the edge capabilities based on the application information installed on the user's terminal (e.g., at least one map application is installed). Prioritizes calling the installed map application with the highest matching degree to execute the navigation task. If the terminal does not have any map application installed that supports this capability, it automatically calls the cloud-side navigation service (e.g., an online map API).

[0070] In this embodiment, priority is given to matching the target atomic capabilities on the terminal side, which can ensure that the matched target atomic capabilities are executable in the terminal environment and realize low latency and high energy efficiency localized services. When there are no available capabilities on the terminal side, the cloud-side atomic capabilities are automatically matched to ensure service availability, thereby improving the task execution success rate and user experience.

[0071] In some embodiments, such as Figure 4 As shown, step S1042 above, which involves matching atomic capabilities in the full list of atomic capabilities based on the target task and pre-acquired application information installed on the user terminal to determine the target atomic capability, may include: S10421: Based on the description information of the target task, perform preliminary screening in the full list of atomic capabilities to obtain a set of candidate atomic capabilities related to the target task.

[0072] The atomic capability platform can perform functional semantic matching in the full list of atomic capabilities based on the structured description information of the target task to obtain a set of candidate atomic capabilities, which includes candidate edge-side atomic capabilities and candidate cloud-side atomic capabilities.

[0073] S10422: Input the description information of the target task, application information, and description information of each atomic capability in the candidate atomic capability set into the large language model.

[0074] The atomic capability platform can input the structured description information of the target task, the list of applications installed on the user terminal, and the metadata description information of each atomic capability in the candidate atomic capability set into the large language model.

[0075] The large language model can be a general-purpose large language model or a specialized model fine-tuned for capability matching tasks, such as an autoregressive language model based on the Transformer architecture. The large language model can be trained using a combination of pre-training and instruction fine-tuning: first, pre-training is performed on a large-scale general corpus to obtain basic language understanding capabilities; then, instruction fine-tuning is performed based on atomic capability metadata, task description samples, and task-capability matching annotation data to enhance its reasoning capabilities in semantic matching, edge-cloud capability assessment, and context-aware ranking.

[0076] In some examples, the large language model can use the list of applications installed on the user terminal as context to evaluate the availability of terminal-side atomic capabilities, and combine task semantic features with the deployment location information of each atomic capability for comprehensive ranking, so that terminal-side atomic capabilities have higher ranking priority than cloud-side atomic capabilities in the ranking results, and terminal-side atomic capabilities corresponding to applications not installed have the lowest ranking or are excluded from the valid matching results.

[0077] S10423: Determine the final target atomic capability based on the matching results output by the large language model.

[0078] For example, the matching results output by the large language model may include capability matching ranking results. The atomic capability platform selects the atomic capability ranked first as the target atomic capability based on the capability matching ranking results.

[0079] In this embodiment, by performing functional semantic screening on the full list of atomic capabilities, a set of candidate atomic capabilities is obtained, which can effectively narrow the search scope and reduce the computational load of the large model. Furthermore, by combining the large language model to perform deep semantic understanding of the target task description and candidate atomic capability metadata, and by matching capabilities based on the list of applications installed on the terminal, the accuracy of capability matching and the feasibility of execution can be improved.

[0080] In some embodiments, based on Figure 2 ,like Figure 5 As shown, prior to step S104, the method may further include: S101: The atomic capability platform combines multiple atomic capabilities from the full list of atomic capabilities into composite atomic capabilities according to preset arrangement rules.

[0081] Here, the orchestration rule corresponds to a multi-step task and can be registered in the full atomic capability table as part of the metadata of the composite atomic capability. The orchestration rule can be used to control the cooperative execution mode between atomic capabilities, including at least one of the following: call order, data transfer rules, conditional branch judgment conditions, and parallel execution relationship.

[0082] Among them, multi-step tasks are tasks that need to be completed through the collaboration of multiple atomic capabilities. That is, a multi-step task is a complex task that cannot be completed independently by a single atomic capability and requires the sequential or parallel invocation of multiple atomic capabilities and the implementation through data flow.

[0083] In some examples, the atomic capability platform supports orchestrating and encapsulating multiple independent atomic capabilities from the full atomic capability list according to specific business logic through visual orchestration or configuration, forming a composite atomic capability, and registering it to the full atomic capability list.

[0084] For example, the four atomic capabilities of querying destination weather, searching for attractions, planning routes, and booking hotels can be encapsulated into a composite atomic capability called "creating travel guides"; and the four atomic capabilities of batch selecting images, image denoising, color enhancement, and style transfer can be encapsulated into a composite atomic capability called "batch image beautification".

[0085] The process of initiating a capability call to the provider of the target atom capability in step S106 above, and receiving the capability call result, may include: S1061: When the target atomic capability includes a composite atomic capability, the atomic capability platform determines each atomic capability that makes up the composite atomic capability and the corresponding arrangement rules. When the target atomic capability is a composite atomic capability, the atomic capability platform can determine the individual atomic capabilities it contains and the orchestration rules corresponding to the composite atomic capability during registration by parsing the metadata of the composite atomic capability.

[0086] S1062: Based on the determined orchestration rules, initiate capability calls to the providers corresponding to each atomic capability contained in the composite atomic capability, and receive the capability call results.

[0087] According to the orchestration rules, the atomic capability platform sequentially initiates capability calls to the providers corresponding to each atomic capability, and passes the call results of the preceding atomic capabilities to the subsequent atomic capabilities, until the execution of all atomic capabilities contained in the composite atomic capability is completed, and finally returns the overall execution result of the composite atomic capability to the requester.

[0088] In this embodiment, the atomic capability platform supports the orchestration and encapsulation of multiple atomic capabilities according to predefined business logic to form composite atomic capabilities for handling complex tasks. The atomic capability platform automates the processing of complex tasks by sequentially invoking, conditionally branching, or executing the atomic capabilities contained in the composite atomic capability in parallel. Thus, for the requesting party, only one task request needs to be initiated. If the atomic capability platform matches a composite atomic capability to the task request, it sequentially initiates calls to each atomic capability contained in the composite atomic capability according to the orchestration rules that formed the composite atomic capability. This automates the multi-step call process of the entire task, thereby improving the efficiency and continuity of task execution and reducing the operational complexity for users.

[0089] In some embodiments, the atomic capability platform includes an atomic capability platform server deployed on the cloud side and an atomic capability platform client deployed on the edge side; the step S106 above, initiating a capability call to the provider corresponding to the target atomic capability, includes: If the target atomic capability is an edge capability, the atomic capability platform server initiates a local call to the target atomic capability through the atomic capability platform client; if the target atomic capability includes a cloud-side capability, the atomic capability platform server initiates a remote call to the corresponding cloud service interface to the target atomic capability.

[0090] In this embodiment, the atomic capability platform can adopt an edge-cloud integrated distributed architecture, including an atomic capability platform server centrally deployed on a cloud server and an atomic capability platform client deployed on terminal devices such as mobile phones.

[0091] When a call to a target atomic capability is required, the atomic capability platform server determines whether the target atomic capability is an edge-side capability or a cloud-side capability based on its capability attributes. Specifically, for any atomic capability, its capability attributes are defined in the full atomic capability table during registration.

[0092] For edge-side capabilities, since the execution entity is located locally on the terminal, the atomic capability platform server in the cloud sends the invocation command to the atomic capability platform client deployed on the terminal via a network channel. After the client executes the target atomic capability by calling the corresponding system API or application SDK locally, it sends the capability invocation result back to the server. For cloud-side capabilities, since the execution entity is located in the cloud, the atomic capability platform server directly initiates a remote network call (such as an HTTP request) to the interface of the target cloud service provider and obtains the feedback capability invocation result.

[0093] In this embodiment, the atomic capability platform initiates calls to the target atomic capability using different call routing methods based on whether the target atomic capability is an edge-side capability or a cloud-side capability. This balances response speed, privacy and security, and resource consumption, and can provide users with a better capability service experience.

[0094] In some embodiments, a capability invocation method is provided, executed by the atomic capability platform, such as... Figure 6 As shown, the method includes the following steps: S201: Receive a task request sent by the AI ​​agent for the target task; S202: In response to a task request, match at least one target atomic capability for performing the target task from the full list of atomic capabilities pre-established by the atomic capability platform. S203: Return the list of matching atomic capabilities formed by at least one target atomic capability to the AI ​​agent; S204: Receive a capability call request initiated by an AI agent based on an atomic capability matching list; S205: In response to a capability invocation request, initiate a capability invocation to the provider corresponding to at least one target atomic capability; S206: Receive capability invocation results from the provider corresponding to at least one target atomic capability; S207: Adapt the capability invocation result to the corresponding response format according to the invocation protocol associated with the task request; S208: Return the capability call result to the AI ​​agent in an adapted response format.

[0095] In this embodiment, the AI ​​agent acts as the requester. The AI ​​agent can initiate a task request through the Model Context Protocol (MCP), which is routed to the atomic capability platform via the AI ​​business gateway. The atomic capability platform then provides the AI ​​agent with richer atomic capability support and tool execution links.

[0096] For example, after receiving natural language input from a user and understanding their intent, the AI ​​agent submits a structured task request to the atomic capability platform. The platform extracts the task intent and key parameters through semantic recognition, generating a structured target task description. Based on the parsed target task, the platform can search the entire list of atomic capabilities to match one or more functionally compatible target atomic capabilities. For instance, semantic vector matching can be used, encoding the task description and capability metadata into vectors respectively. The platform then identifies the most semantically relevant atomic capabilities by calculating their cosine similarity, forming an atomic capability matching list that is returned to the AI ​​agent.

[0097] The atomic capability matching list includes at least one target atomic capability that matches the target task function. If there are multiple providers of the same target atomic capability (for example, if the user terminal has installed two or more map applications that can be used for navigation), the atomic capability matching list includes atomic capability instances corresponding to each provider.

[0098] In this embodiment, the atomic capability platform returns the atomic capability matching list to the AI ​​agent, which enables the AI ​​agent to autonomously select a target atomic capability based on the user's historical preferences, dialogue context and / or the parameter requirements of each atomic capability (such as parameter name, type, etc.) and initiate a capability call request for the selected target atomic capability to the atomic capability platform, thereby enabling the atomic capability platform to initiate a capability call to the provider corresponding to the target atomic capability.

[0099] This approach improves the success rate and flexibility of capability invocation. On the one hand, when there is missing information in the capability parameters, the AI ​​agent can interact with the user to complete the missing information according to the capability parameter requirements, avoiding capability invocation failure due to incomplete parameters. On the other hand, it enables the AI ​​agent to make autonomous decisions on capability selection and invocation order based on the context, and to perform dynamic capability orchestration for complex tasks. While reducing the coupling between the platform and business logic, it can provide the AI ​​agent with scalable tool invocation support.

[0100] In some embodiments, optional implementations of step S202 described above can be found in [reference needed]. Figure 2 Step S104 Figure 3 Steps S1041 to S1042, and Figure 4 Optional implementations of steps S10421 to S10423, and Figure 2 , Figure 3 , Figure 4 , Figure 5 Other related parts in the embodiments involved will not be described in detail here.

[0101] In some embodiments, the atomic capability platform encapsulates the atomic capability matching list into a tool description set conforming to the Model Context Protocol (MCP) specification and returns it to the AI ​​agent. The tool description set may contain metadata such as identifiers, functional descriptions, parameters, and invocation methods related to each target atomic capability, enabling the AI ​​agent to understand available atomic capabilities and initiate capability invocation requests. The capability invocation request is submitted to the atomic capability platform in the tool invocation message format defined by the MCP protocol.

[0102] In some embodiments, the optional implementations of initiating the capability call in step S205 and receiving the capability call result in step S206 can be found in [reference needed]. Figure 2 Step S106 Figure 5 Optional implementations of steps S1061 to S1062, and Figure 2 , Figure 5 Other related parts in the embodiments involved will not be described in detail here.

[0103] In some embodiments, optional implementations of steps S207 to S208 described above can be found in [reference needed]. Figure 2 Step S108 Figure 5 Optional implementation of step S108, and Figure 2 , Figure 5 Other related parts in the embodiments involved will not be described in detail here.

[0104] In some embodiments, the AI ​​agent includes a master agent and multiple slave agents; the step S203 above, returning the atomic capability matching list formed by at least one target atomic capability to the AI ​​agent, may include: in response to a task request initiated by the master agent, returning the atomic capability matching list to the master agent, so that the master agent can decompose the task based on the atomic capability matching list, generate at least one subtask and distribute it to the target slave agents among the multiple slave agents. In step S204 above, receiving a capability call request initiated by the AI ​​agent based on the atomic capability matching list may include: receiving a capability call request initiated by the target from the agent based on a subtask, wherein the capability call request carries the target atomic capability information specified by the subtask.

[0105] In this embodiment, to better handle complex tasks, the AI ​​agent adopts a master-slave architecture, including a master agent (which may be called the system agent) and multiple slave agents (which may be called vertical agents). The master agent is used to receive user requests, perform task parsing and decomposition, and initiate capability matching requests to the atomic capability platform. After obtaining a list of candidate atomic capabilities, it coordinates and schedules the execution processes of each slave agent. Each slave agent, targeting a specific vertical field (such as healthcare, travel, e-commerce, etc.) or specialized function (such as information retrieval, numerical calculation, content generation, etc.), initiates targeted capability call requests to the atomic capability platform based on the sub-tasks assigned by the master agent and executes them.

[0106] In this embodiment, a master-slave architecture consisting of a master agent and slave agents, combined with the interaction link with the atomic capability platform, is used to achieve hierarchical decoupling and collaborative parallelism of task processing, thereby improving task execution efficiency.

[0107] Next, with specific examples, the capability invocation method provided in the embodiments of this application will be further explained.

[0108] See Figure 7 As shown in the figure, this application provides a capability invocation method, which may include: 1. Applications (such as voice assistant apps) transmit the user's original request to the system's intelligent agent through the AI ​​business gateway; 2. After parsing the user's intent, the system intelligent agent (i.e., the main intelligent agent) submits a structured task request to the atomic capability platform through the AI ​​business gateway; 3. The atomic capability platform matches the associated capability list based on the applications installed on the user's device, and uses a large language model to accurately filter and determine the most suitable target atomic capability; 4. The system agent decomposes the task based on the atomic capability matching list, generates at least one sub-task, and distributes it to the corresponding vertical agent (i.e., the agent from the agent). 5. The vertical intelligent agent initiates a capability call request to the atomic capability platform server; 6. When the atomic capability platform server responds to a capability call request, it determines whether the atomic capability to be called is an edge capability. If it is an edge capability, the atomic capability service client initiates the edge capability call and returns the call result to the atomic capability platform server. If it is a cloud capability, the atomic capability platform server directly initiates a cloud capability call to the third-party service and obtains the call result. 7. The system's intelligent agent encapsulates the call results returned by the atomic capability platform server and finally returns a response to the user through the voice assistant application.

[0109] In summary, the atomic capability platform provided in this application provides AI intelligent hardware devices with rich atomic capabilities, matches precise capabilities to user scenario requirements, and connects the interaction link between the client and the intelligent agent to achieve unified capability distribution and invocation. This solution has at least the following beneficial effects: 1. A unified standardized access mechanism is established for the atomic capability platform, supporting access to operating system-level capabilities, AI-native capabilities, and third-party capabilities. It also supports API interfaces and the MCP protocol standard for calling atomic capabilities, better compatibility with application-side and agent-side calling scenarios. Furthermore, the platform automatically encapsulates the API interfaces into MCP tools, eliminating the need for third-party MCP development and reducing capability access costs. 2. Applications can choose the calling path according to the scenario. One approach is for the APP to directly call known atomic capabilities through standard API interfaces, suitable for scenarios with clearly defined functions and complete parameters. The other approach is for the AI ​​agent to access the atomic capability platform via the AI ​​business gateway using the MCP protocol, where the platform completes semantic parsing, capability matching, and scheduling, suitable for natural language interaction and complex task scenarios. The AI ​​agent link is designed at the OS underlying architecture level. Through the AI ​​business gateway and atomic capability platform, the business division between modules is clear, providing richer atomic capability support and tool execution links for the AI ​​agent. 3. Establish an interaction link between the edge / cloud / intelligent agent. The atomic capability platform server acts as the unified exit for capability invocation, initiating capability invocations to third-party servers or OS application clients and receiving the invocation results back to the intelligent agent. The atomic capability client acts as the unified scheduler of edge-side atomic capabilities, transmitting invocation information with the atomic capability platform server. The edge and cloud have a unified interface, reducing the impact on OS system performance and power consumption, and ensuring the legality, security, efficiency, and stability of capability invocations. 4. By integrating vector libraries and semantic models, capability matching is achieved on the full capability set on the atomic capability platform. Combined with capability matching training of large models, the matching accuracy is improved in two ways, and an AI proactive service experience is realized. 5. Services can be proactively recommended based on user needs through service distribution entry points, including system-level entry points (global search, system navigation bar, etc.), intelligent agent entry points, and in-application entry points, and matching atomic capabilities can be invoked, resulting in diverse application scenarios; 6. It can dynamically expand to multiple devices in people, vehicles, and homes, and call atomic capabilities in a unified manner through the system's intelligent agent. It can also be linked to user accounts and memories to continuously optimize the matching degree of capabilities and personalized capability recommendations.

[0110] The various embodiments or implementation methods described in this specification are presented in a progressive manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other.

[0111] In the various embodiments of the specification, some or all of the steps and their optional implementations can be arbitrarily combined with some or all of the steps in other embodiments, or arbitrarily combined with the optional implementations in other embodiments.

[0112] Based on the capability invocation method provided in the preceding embodiments, this application also provides a capability invocation device applied to an atomic capability platform. The capability invocation device 100 includes: The request receiving unit 101 is used to receive a task request sent by the requester for the target task. The request processing unit 102 is configured to, in response to the task request, match at least one target atomic capability for executing the target task from a pre-established full list of atomic capabilities in the atomic capability platform; wherein the full list of atomic capabilities contains capability information of multiple atomic capabilities; The execution unit 103 is used to initiate a capability call to the provider of the target atomic capability and receive the capability call result; The result return unit 104 is used to adapt the capability call result into the corresponding response format and return it to the requester according to the call protocol type associated with the task request.

[0113] In some embodiments, the plurality of atomic capabilities includes operating system atomic capabilities, AI atomic capabilities, and third-party service capabilities; the full list of atomic capabilities is generated by registering atomic capabilities from different providers using a standardized data structure.

[0114] In some embodiments, the task request originates from an application running on a user terminal, and the calling protocol type associated with the task request is an application programming interface (API) protocol. Alternatively, the task request may originate from an AI agent, and the associated calling protocol type is a model context protocol.

[0115] In some embodiments, the request processing unit is configured to: In response to the task request, determine the target task corresponding to the task request; Based on the target task and the application information pre-acquired on the user terminal, atomic capabilities are matched in the full list of atomic capabilities to determine the target atomic capability; The user terminal is either the terminal where the application that initiated the task request is located, or the terminal where the application that initiated the task request is located through an AI agent.

[0116] In some embodiments, the request processing unit is specifically used for: Based on the description information of the target task, a preliminary screening is performed in the full list of atomic capabilities to obtain a set of candidate atomic capabilities related to the target task. The description information of the target task, the application information, and the description information of each atomic capability in the candidate atomic capability set are input into the large language model; Based on the matching results output by the large language model, the final target atomic capability is determined.

[0117] In some embodiments, the apparatus further includes: The capability management unit is used by the atomic capability platform to combine multiple atomic capabilities in the full list of atomic capabilities into composite atomic capabilities according to preset arrangement rules. The call execution unit is specifically used for: When the target atomic capability includes a composite atomic capability, the atomic capability platform determines each atomic capability that is combined into the composite atomic capability and the corresponding arrangement rules; According to the determined orchestration rules, a capability call is initiated to the provider corresponding to each atomic capability included in the composite atomic capability, and the capability call result is received.

[0118] In some embodiments, the atomic capability platform includes an atomic capability platform server deployed on the cloud side and an atomic capability platform client deployed on the edge side; The call execution unit is specifically used for: If the target atomic capability is an edge capability, then the atomic capability platform server initiates a local call to the target atomic capability through the atomic capability platform client; If the target atomic capability includes cloud-side capabilities, then the atomic capability platform server initiates a remote call to the corresponding cloud service interface for the target atomic capability.

[0119] In some embodiments, the task request is initiated by an AI agent for a target task; the request processing unit is further configured to: A list of matching atomic capabilities formed by at least one target atomic capability is returned to the AI ​​agent; The call execution unit is specifically used for: Receive capability invocation requests initiated by the AI ​​agent based on the atomic capability matching list; In response to the capability invocation request, a capability invocation is initiated to the provider corresponding to at least one target atomic capability.

[0120] In some embodiments, the AI ​​agent includes a master agent and multiple slave agents; The request processing unit is specifically used for: In response to a task request initiated by the master agent, the atomic capability matching list is returned to the master agent so that the master agent can decompose the task based on the atomic capability matching list, generate at least one sub-task, and distribute it to the target slave agent among the plurality of slave agents. The call execution unit is specifically used for: Receive the capability invocation request initiated by the target from the agent based on the subtask, wherein the capability invocation request carries the target atomic capability information specified by the subtask.

[0121] This application also provides a capability invocation system, the system comprising: an atomic capability platform, at least one requester, and at least one capability provider; The atomic capability platform is used to execute the capability invocation method provided in any of the foregoing embodiments; The requesting party is used to send a task request for the target task to the atomic capability platform and receive the capability invocation result returned by the atomic capability platform; The capability provider is used to respond to the capability call request from the atomic capability platform, execute the corresponding atomic capability, and return the result.

[0122] This application also provides an electronic device, including a processor and steps for the processor to invoke instructions to cause the electronic device to perform the capability invocation method provided in any of the foregoing embodiments.

[0123] This application also provides a storage medium, including an executable program stored thereon, which, when executed by a processor, implements the steps of the capability invocation method provided in any of the foregoing embodiments.

[0124] For ease of understanding, the following focuses on explaining the terminology used in this embodiment: In this application embodiment, the processor is a circuit with signal processing capabilities. In one implementation, the processor can be a circuit with instruction read and execute capabilities, such as a Central Processing Unit (CPU), a microprocessor, a Graphics Processing Unit (GPU) (which can be understood as a type of microprocessor), or a Digital Signal Processor (DSP). In another implementation, the processor can implement certain functions through the logical relationships of hardware circuits. The logical relationships of the aforementioned hardware circuits are fixed or reconstructable. For example, the processor is a hardware circuit implemented using an Application-Specific Integrated Circuit (ASIC) or a Programmable Logic Device (PLD), such as an FPGA. In a reconstructable hardware circuit, the processor loads a configuration document, implementing a cyclical process of hardware circuit configuration. This can be understood as the processor loading instructions to implement the functions of some or all of the above units or modules in a cyclical process. In addition, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as a Neural Network Processing Unit (NPU), a Tensor Processing Unit (TPU), a Deep Learning Processing Unit (DPU), etc.

[0125] The computer-readable storage medium provided in this embodiment can execute the capability invocation method of the above embodiment. Its implementation principle and technical effect are similar to those of the above embodiment, and will not be repeated here.

[0126] The aforementioned computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0127] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in an electronic device or a host device.

[0128] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0129] The various embodiments or implementation methods described in this specification are presented in a progressive manner. Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other.

[0130] In the description of this specification, references to "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0131] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for invoking capabilities, characterized in that, Executed by an atomic capability platform, the method includes: The receiving party sends a task request for the target task. In response to the task request, at least one target atomic capability for performing the target task is matched from the full list of atomic capabilities pre-established by the atomic capability platform; wherein, the full list of atomic capabilities contains capability information of multiple atomic capabilities; Initiate a capability call to the provider of the target atomic capability and receive the capability call result; Based on the invocation protocol type associated with the task request, the capability invocation result is adapted to the corresponding response format and returned to the requester.

2. The capability invocation method according to claim 1, characterized in that, The multiple atomic capabilities include operating system atomic capabilities, AI atomic capabilities, and third-party service capabilities; the full list of atomic capabilities is generated by registering atomic capabilities from different providers using a standardized data structure.

3. The capability invocation method according to claim 1, characterized in that, The task request originates from an application running on the user terminal, and the calling protocol type associated with the task request is an application programming interface (API) protocol. Alternatively, the task request may originate from an AI agent, and the associated calling protocol type is a model context protocol.

4. The capability invocation method according to claim 1, characterized in that, In response to the task request, at least one target atomic capability for performing the target task is matched from a pre-established full list of atomic capabilities on the atomic capability platform, including: In response to the task request, determine the target task corresponding to the task request; Based on the target task and the application information pre-acquired on the user terminal, atomic capabilities are matched in the full list of atomic capabilities to determine the target atomic capability; The user terminal is either the terminal where the application that initiated the task request is located, or the terminal where the application that initiated the task request is located through an AI agent.

5. The capability invocation method according to claim 4, characterized in that, The step of matching atomic capabilities in the full list of atomic capabilities based on the target task and pre-acquired application information installed on the user terminal to determine the target atomic capability includes: Based on the description information of the target task, a preliminary screening is performed in the full list of atomic capabilities to obtain a set of candidate atomic capabilities related to the target task. The description information of the target task, the application information, and the description information of each atomic capability in the candidate atomic capability set are input into the large language model; Based on the matching results output by the large language model, the final target atomic capability is determined.

6. The capability invocation method according to claim 1, characterized in that, The method further includes: The atomic capability platform combines multiple atomic capabilities from the full list of atomic capabilities into composite atomic capabilities according to preset arrangement rules. The process of initiating a capability call to the provider of the target atomic capability and receiving the capability call result includes: When the target atomic capability includes a composite atomic capability, the atomic capability platform determines each atomic capability that is combined into the composite atomic capability and the corresponding arrangement rules; According to the determined orchestration rules, a capability call is initiated to the provider corresponding to each atomic capability included in the composite atomic capability, and the capability call result is received.

7. The capability invocation method according to claim 1, characterized in that, The atomic capability platform includes an atomic capability platform server deployed on the cloud side and an atomic capability platform client deployed on the edge side; The process of initiating a capability invocation to the provider corresponding to the target atomic capability includes: If the target atomic capability is an edge capability, then the atomic capability platform server initiates a local call to the target atomic capability through the atomic capability platform client; If the target atomic capability includes cloud-side capabilities, then the atomic capability platform server initiates a remote call to the corresponding cloud service interface for the target atomic capability.

8. The capability invocation method according to any one of claims 1 to 7, characterized in that, The task request is initiated by an AI agent for a target task; the method further includes: A list of matching atomic capabilities formed by at least one target atomic capability is returned to the AI ​​agent; The initiation of a capability invocation to the provider of the target atomic capability includes: Receive capability invocation requests initiated by the AI ​​agent based on the atomic capability matching list; In response to the capability invocation request, a capability invocation is initiated to the provider corresponding to at least one target atomic capability.

9. The capability invocation method according to claim 8, characterized in that, The AI ​​agent includes a master agent and multiple slave agents; The step of returning the atomic capability matching list, formed by at least one target atomic capability, to the AI ​​agent includes: In response to a task request initiated by the master agent, the atomic capability matching list is returned to the master agent so that the master agent can decompose the task based on the atomic capability matching list, generate at least one sub-task, and distribute it to the target slave agent among the plurality of slave agents. Receiving the capability invocation request initiated by the AI ​​agent based on the atomic capability matching list includes: Receive the capability invocation request initiated by the target from the agent based on the subtask, wherein the capability invocation request carries the target atomic capability information specified by the subtask.

10. A capability mobilization device, characterized in that, Applied to an atomic capability platform, the device includes: The request receiving unit is used to receive task requests sent by the requester for the target task. The request processing unit is configured to, in response to the task request, match at least one target atomic capability for executing the target task from a pre-established full list of atomic capabilities in the atomic capability platform; wherein the full list of atomic capabilities contains capability information of multiple atomic capabilities; The execution unit is used to initiate a capability call to the provider of the target atomic capability and to receive the capability call result; The result return unit is used to adapt the capability call result into the corresponding response format and return it to the requester according to the call protocol type associated with the task request.

11. A capability invocation system, characterized in that, The system includes: an atomic capability platform, at least one requester, and at least one capability provider; The atomic capability platform is used to execute the capability invocation method as described in any one of claims 1 to 9; The requesting party is used to send a task request for the target task to the atomic capability platform and receive the capability invocation result returned by the atomic capability platform; The capability provider is used to respond to the capability call request from the atomic capability platform, execute the corresponding atomic capability, and return the result.

Images