Portable multi-platform artificial intelligence system, operating method, medium and electronic device

Abstract

The application provides a portable multi-platform artificial intelligence system, a running method, a medium and an electronic device, and the method comprises the following steps: in response to the fact that a removable storage medium is connected to a host computer, loading a corresponding installation-free running environment from the medium according to the type of the operating system of the host computer; when responding to a user request, detecting a network connection state, and adaptively selecting between a remote AI processing module and a local AI processing module pre-stored on the medium. The application also comprises the following steps: realizing remote control without a public network IP through a third-party message relay service, automatically processing user documents to build a local knowledge base, and expanding AI capability by installing a skill description file. The application realizes multi-platform plug-and-play of an AI system, cloud / local dual-mode adaptation, knowledge base self-learning and capability expansion, while guaranteeing the physical controllability of user data, and significantly reduces the use and deployment threshold of a personal AI system.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and computer software technology, specifically to a portable multi-platform artificial intelligence system, operating method, medium, and electronic device. Background Technology

[0002] In recent years, personal AI assistants have become an important tool for improving productivity. Currently, the deployment of such systems mainly falls into two categories: cloud services and local installation. The cloud service model, such as various web-based or client applications, delegates all user computational tasks and data processing to the service provider's servers. While convenient, this model offers users limited physical control over their personal data, poses a risk of privacy breaches, and is completely unusable in environments without internet access.

[0003] The local installation mode requires users to install the application, configure the runtime environment, and download the model files on their own computers. While this method keeps the data local, the deployment process is cumbersome and not user-friendly for non-technical users. More importantly, this mode ties the AI ​​system to specific hardware devices and operating systems, lacking portability. When users need to switch between their office Windows computer and their home Linux computer, they must repeat the entire installation and configuration process on each device, failing to achieve seamless integration.

[0004] To address the issue of software being tied to specific devices, portable application technology has emerged, which packages applications and their dependencies into removable storage media for "plug-and-play" functionality. However, existing portable application solutions primarily target single-function utility software and typically only support a single operating system platform. These solutions do not address the challenges of supporting complex AI systems that integrate large language model calls, local knowledge bases, and scalable skill ecosystems, nor do they solve advanced requirements such as seamless switching between different operating systems, adaptive operation in offline environments, and remote access without a public IP address.

[0005] Therefore, existing technologies generally suffer from one or more drawbacks, such as immobility, poor cross-platform compatibility, uncontrollable user data, inaccessibility when offline, fixed capabilities, and high deployment barriers. The market urgently needs a new solution that can truly portability a fully functional artificial intelligence system, enabling users to plug and play it on any computer while ensuring data security and controllability and the continuous availability of core functions. Summary of the Invention

[0006] In view of the deficiencies in the prior art, the purpose of this invention is to provide a portable multi-platform artificial intelligence system, operating method, medium and electronic device.

[0007] A portable multi-platform artificial intelligence system provided by the present invention includes: A removable storage medium on which artificial intelligence applications are deployed; A runtime environment module, pre-installed in the removable storage medium, is configured to: in response to the removable storage medium being connected to a host computer, load a corresponding installation-free runtime environment to run the artificial intelligence application based on the operating system type of the host computer, without modifying the system configuration of the host computer; and, An AI model invocation module is configured to: detect the network connection status of the host computer when responding to a user request, and adaptively select between a remote AI processing module and a local AI processing module pre-installed on the removable storage medium based on the network connection status to process the user request.

[0008] A method for operating a portable multi-platform artificial intelligence system according to the present invention includes: In response to a removable storage medium being connected to a host computer, a corresponding runtime environment is loaded from a plurality of pre-installed portable runtime environments stored in the removable storage medium, based on the operating system type of the host computer, to run an artificial intelligence system deployed on the removable storage medium; When responding to a user request, the network connection status of the host computer is detected, and based on the network connection status, an adaptive selection is made between a remote AI processing module and a local AI processing module pre-installed on the removable storage medium to process the user request.

[0009] Preferably, the method further includes: By establishing a communication connection with a server of a third-party message relay service, receiving remote instructions forwarded by the third-party message relay service, and controlling the operation of the artificial intelligence system according to the remote instructions; The third-party message relay service is at least one of the following: a publicly available instant messaging platform, an email server, or an IoT message queue service. The establishment of the communication connection includes: when the artificial intelligence system is initialized, actively establishing a long polling connection or a WebSocket persistent connection with the server of the third-party message relay service, so as to continuously listen for and receive remote instructions forwarded by the third-party message relay service as a client; The user document data, the local knowledge base index, and the user interaction data generated during the processing of the user text request are encrypted and stored in the removable storage medium. The encrypted storage includes: encrypting the entire user data partition of the removable storage medium using full-disk encryption technology, or encrypting various types of user data stored in the removable storage medium, such as user configuration files, API keys, conversation log data, source files and indexes of personal knowledge bases, and installed skill module data, at the file level or database level.

[0010] Preferably, the step of processing the user request includes: The system automatically processes user document data stored in the removable storage medium to construct a local knowledge base index associated with the user document data. When responding to a user text request that is a user request, context information related to the user text request is retrieved from the local knowledge base index; The context information is combined with the user's text request to generate a prompt word; The prompt is submitted to the adaptively selected remote AI processing module or local AI processing module to obtain a response text based on the context information.

[0011] Preferably, the step of automatically processing user document data stored in the removable storage medium to construct a local knowledge base index associated with the user document data includes: A multi-level degradation parsing strategy is adopted for content extraction to extract text information from the user document data for constructing the local knowledge base index. The degradation parsing strategy includes: for a specific file format, preferentially calling a high-performance dedicated third-party parsing library pre-installed in the installation-free runtime environment for parsing; When the dedicated third-party parsing library fails to load due to environmental limitations or an error occurs during the parsing process, it automatically downgrades and switches to use a more general standard library or alternative parsing logic for a second parsing attempt. The step of retrieving context information related to the user text request from the local knowledge base index when responding to the user text request as a user request specifically includes: First, natural language understanding is performed on the user text request to extract the core intent and keywords. Then, using the local knowledge base index, semantic similarity retrieval based on word vectors or traditional keyword weighted retrieval are performed to find several document fragments that are semantically most relevant to the user text request from the user document data as the context information, and sort them by relevance.

[0012] Preferably, the method further includes: Encapsulate domain-specific AI capabilities into independent skill description files, which contain structured workflow instructions for guiding AI models to perform specific tasks; and provide mechanisms for installing and loading the skill description files. The installation mechanism is integrated into a graphical skill management interface and supports at least one of the following installation methods: -By connecting to a remote skills distribution platform maintained by a service provider or community, you can browse, search, and download and install skills with one click. - Quick installation can be performed by directly entering the unique identifier of a known skill in the management interface; - Select and upload a skill description file package packaged in a specific format via the local file upload dialog box for offline installation.

[0013] Preferably, the mechanism for loading the skill description file includes: Before generating the prompt word, the corresponding skill description file is dynamically matched and loaded according to the user text request, and the workflow instructions contained in the skill description file are injected into the prompt word; Specifically, injecting workflow instructions into the prompts means incorporating a series of structured instructions, context examples, output format requirements, or role-playing settings defined in the skill description file for completing specific tasks into a system-level prompt. This is combined with the context information and the user text request to guide the general artificial intelligence model in the remote AI processing module or the local AI processing module, constraining and guiding its behavior so that it performs specific tasks according to a preset professional process.

[0014] Preferably, the local AI processing module includes a CPU-based inference engine and uses a quantized large language model in GGUF format; The method further includes: when loading the local AI processing module, first performing hardware capability detection to check whether the host computer has a compatible graphics processor and corresponding computing environment; if an available GPU environment is detected, then prioritizing the selection and loading of an inference backend that supports GPU acceleration to obtain higher performance; if no GPU environment is detected or the environment is incompatible, then automatically falling back to loading the CPU-based inference engine to ensure that the local AI processing module can run stably on host computers with different hardware configurations.

[0015] According to the present invention, a computer-readable storage medium is provided thereon storing a computer program, which, when executed by a processor, implements the method described herein.

[0016] An electronic device according to the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method described herein.

[0017] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention integrates multiple installation-free operating environments into a single removable storage medium, enabling seamless switching between different operating systems such as Windows and Linux. No installation or configuration is required on the host computer, making AI portable and plug-and-play.

[0018] 2. This invention can automatically detect network status, utilize the powerful computing power of the cloud when there is a network connection, and seamlessly switch to the local model when the network is disconnected, ensuring the availability of core functions in various complex network environments such as business trips and internal networks.

[0019] 3. In this invention, all personal data, including knowledge bases and conversation logs, are encrypted and stored in a removable storage medium owned by the user. Removing the medium takes away all the data without leaving any trace on the host computer, thus protecting the user's data sovereignty and privacy at the physical level.

[0020] 4. By introducing an extensible skill mechanism, this invention allows users to install new capabilities on the AI ​​system as needed, just like installing a mobile app. This transforms the AI ​​system from a tool with fixed functions into an intelligent partner that can grow together with the user's needs.

[0021] 5. This invention innovatively utilizes a third-party message relay service as a communication relay, enabling remote control without the need for a public IP address or intranet penetration, greatly facilitating users' access to and management of the AI ​​system from different locations. Attached Figure Description

[0022] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a system architecture diagram according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the multi-platform installation-free startup process according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the adaptive switching process of the AI ​​module in an embodiment of the present invention. Figure 4 This is a flowchart illustrating the local knowledge base construction and retrieval process according to an embodiment of the present invention. Figure 5 This is a diagram of the communication link for remote control without a public IP address, as described in an embodiment of the present invention. Figure 6 This is a flowchart illustrating the scalable skill loading and execution process according to an embodiment of the present invention.

[0023] Explanation of reference numerals in the attached figures: Detailed Implementation

[0024] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0025] In the context of this invention, existing AI assistants or intelligent agent systems generally suffer from poor portability, weak platform compatibility, difficulty in protecting user data privacy, and heavy reliance on network connectivity. Users often need to reconfigure their devices when changing them, personal data is stored in the cloud, posing a risk of leakage, and the systems are completely unusable in offline environments. This invention aims to address these technical pain points. Its core idea is to provide an innovative portable AI system and its operating method. This system is completely encapsulated on a removable storage medium, achieving plug-and-play functionality, cross-platform compatibility, offline operation support, and ensuring the physical privacy of user data.

[0026] To achieve the above objectives, this invention introduces and defines several core technical concepts, including: Installation-free runtime environment: refers to a standalone software environment pre-installed on a removable storage medium, containing a platform-specific interpreter and all necessary dependent libraries, which allows applications to run directly without any installation or system-level configuration on the host computer.

[0027] Remote AI processing module: Specifically refers to a module deployed on a cloud server that provides high-level artificial intelligence (such as large-scale language models) inference services through a network API interface.

[0028] Local AI processing module: This is a module pre-installed on a removable storage medium that can utilize the host computer's local computing resources (such as CPU or GPU) to perform artificial intelligence tasks.

[0029] Third-party message relay service: refers to a publicly available platform for relaying messages between clients in different network environments, such as instant messaging platforms or email servers. This invention cleverly utilizes such services to achieve remote control without the need for a public IP address.

[0030] Skill description file: This is a structured data file, usually in YAML or JSON format, which defines the workflow instructions, context examples, output format requirements and other meta-information required to guide the AI ​​model to complete a specific domain task. It is the key to realizing the scalability of AI capabilities.

[0031] like Figure 1 As shown, this invention provides a portable multi-platform artificial intelligence system, whose overall architecture is deployed on a removable storage medium 100. Logically, the system can be divided into multiple functional modules. The bottom layer is the runtime environment module 110, which provides cross-platform support for upper-layer applications. The middle layer is the core AI model invocation module 120, responsible for handling all invocation and switching logic related to the artificial intelligence model. The top layer consists of application modules that directly interact with users or provide data management capabilities, mainly including a local data storage module 130, which can be further subdivided into sub-parts such as a knowledge base, skill base, user configuration, and dialogue records. This layered architecture makes the functions of each part of the system clear, easy to maintain, and expandable.

[0032] Reference Figure 2 This invention provides a method for operating a portable multi-platform artificial intelligence system. The specific execution flow begins in response to a removable storage medium connecting to a host computer (step S201). At this point, the system does not execute immediately but waits for the user to activate it by executing a preset startup script. Once activated, the system first executes step S202, which loads a corresponding runtime environment from multiple pre-installed portable runtime environments stored on the removable storage medium, based on the host computer's operating system type. For example, the system can accurately identify the operating system by reading system environment variables (e.g., in Python, using `os.name` to determine that 'nt' represents Windows and 'posix' represents Linux / macOS) or by executing platform-specific commands (e.g., executing `uname` in a Linux environment). After identifying the operating system type, the system executes step S203 to load the corresponding portable runtime environment. Specifically, if the system is identified as a Windows system, a pre-installed Windows-based embedded Python environment is loaded; if it is identified as a domestic UOS or other Linux distribution, the corresponding Linux environment is loaded. This loading process involves only setting the interpreter path and dependency library search paths in memory, without writing any data to the host computer's system directory, registry, or any global configuration files, and without installing any drivers. This design achieves true zero-installation and plug-and-play functionality; when the removable storage medium is removed, no trace of its operation is left on the host computer, greatly improving the system's portability and security.

[0033] Next, in response to a user request, the system executes a crucial adaptive selection process, which... Figure 3 The process is detailed below. Upon receiving a user request S301, the system first executes step S302, which checks the network connectivity status of the host computer. A specific detection method involves sending a lightweight HTTP PING or ICMP ECHO request to one or more preset, highly available network endpoints (e.g., the health check API address of the remote AI processing module or a public DNS server). Based on whether the request successfully returns within a preset timeout period, the system can determine the current network connectivity. Based on the network connectivity detection result, the system adaptively selects between the remote AI processing module and the local AI processing module pre-installed on the removable storage medium. If a smooth network connection is detected, step S303 is executed, selecting to invoke the remote AI processing module. At this point, the user's request is formatted and sent to the large language model in the cloud via the API interface for processing, leveraging its powerful computing capabilities and model performance. Conversely, if a network interruption or extremely poor connection quality is detected, the system automatically executes step S304, selecting to load and use the local AI processing module. This design enables the system to provide the best experience when there is a network connection, and to ensure the availability of core functions when there is no network connection, thus achieving strong adaptability to different network environments.

[0034] Furthermore, to achieve remote access without complex network configuration, the method of this invention also includes establishing a communication connection with a server of a third-party message relay service to receive and respond to remote commands. For example... Figure 5 As shown, the communication link involves a remote user 501, a third-party message relay service 502, a host computer 503 running the present invention, and an artificial intelligence system 504 on it. Specifically, during the initialization of the artificial intelligence system 504, it acts as a client, actively establishing a long polling connection or a WebSocket persistent connection with the server of the third-party message relay service. For example, the system can log in to a dedicated instant messaging account and continuously listen for messages sent to that account. When the remote user 501 wishes to interact with the system, he / she only needs to send a message to the dedicated account through the corresponding client on his / her own device. This message is routed and forwarded by the server of the third-party message relay service 502 and is ultimately received by the artificial intelligence system 504 running on the host computer 503. After receiving the instruction, the system processes it and replies to the remote user with the result through the same relay service. A significant advantage of this approach is that the initiator of the communication is always the artificial intelligence system located on the internal network, which actively establishes a connection to the public network server, thus cleverly bypassing common network limitations such as the host computer lacking a public IP address, port forwarding, or internal network penetration, greatly reducing the technical threshold for remote use.

[0035] As a preferred embodiment, the operating method of the portable multi-platform artificial intelligence system provided by the present invention is based on the aforementioned portable multi-platform artificial intelligence system. It should be understood that the operating method provided by the present invention is not limited to the aforementioned portable multi-platform artificial intelligence system, but can also be applied to portable multi-platform artificial intelligence systems with other hardware architectures.

[0036] As a preferred implementation, to make the AI's answers more personalized and professional, the process of handling user requests also includes deep integration with a local knowledge base. For example... Figure 4 As shown, the process begins in step S401, which automatically processes user document data stored on the removable storage medium. Users can place documents containing their personal or professional knowledge (such as .docx, .pdf, .txt, etc.) in a preset directory. The system monitors this directory, and automatically triggers the processing flow once a new file is detected. In processing user document data, to accurately extract text information from files of various complex formats, the system employs a multi-level degradation parsing strategy in step S402. The core idea of ​​this strategy is "prioritize dedicated libraries, with backup in case of failure." For example, when parsing a PDF file, the system will first attempt to call a high-performance dedicated third-party parsing library, such as `pymupdf`. If the library fails to load or parses abnormally due to limitations of the host computer environment, the system will not directly terminate with an error message, but will automatically degrade and switch to a more general, less dependent backup library (such as `pypdf`) for a second parsing attempt. This degradation mechanism ensures that even in various non-standardized, installation-free operating environments, the system can maximize the success rate of parsing user documents, improving the robustness of knowledge base construction.

[0037] After the text information is successfully extracted, the system will execute step S403 to build a local knowledge base index using this text information. This can be achieved using various techniques, such as using a full-text search engine index, or using a vector database, where the text is sliced ​​and converted into vectors using a pre-trained embedding model and stored to support subsequent semantic retrieval. After building the index, when responding to a user's text request, the system will execute step S404 to retrieve contextual information related to the user's question from the local knowledge base index. Specifically, this retrieval step can first perform natural language understanding on the user's text request to extract its core intent and keywords. Subsequently, the retrieval operation is performed using the built local knowledge base index. If a vector database is used, the user request is also converted into a vector, and a semantic similarity-based retrieval is performed in the database; if a full-text search engine is used, a keyword-based retrieval can be performed. These retrieved document fragments will be used as contextual information and sorted according to relevance scores, selecting the top-ranked parts.

[0038] After obtaining the relevant contextual information, the system combines this information with the user's original text request to generate a richer and more targeted prompt. For example, a template like "Based on the following information: [Contextual Information], please answer the question: [User Text Request]" can be used. Finally, this combined prompt is submitted to either the remote or local AI processing module adaptively selected in the preceding steps, resulting in a more accurate and personalized response text that closely incorporates the user's private knowledge. This process transforms AI from a vague, general model into a personal expert familiar with the user's specific knowledge domain.

[0039] To ensure the confidentiality and integrity of user data, the method of this invention also emphasizes encrypted data storage. Specifically, all sensitive user data, including but not limited to user document data, constructed local knowledge base indexes, user interaction data generated during user request processing, user configuration files, API keys, and installed skill description file data, will be encrypted and stored on the removable storage medium. The encryption scheme can be flexibly selected. For example, full-disk encryption technology can be used to create an encrypted partition on the removable storage medium, requiring a password to mount the partition during system operation. This method provides the highest level of physical security. As another or supplementary encryption method, finer-grained encryption can also be performed at the file system level, such as using standard encryption libraries to encrypt specific files or database entries. Regardless of the method used, the core objective is to ensure that in the event of loss of the removable storage medium or access by an unauthorized party, the statically stored data is inaccessible due to encryption, thereby effectively protecting user privacy and data security.

[0040] In addition to utilizing local knowledge bases, this invention introduces a scalable skill mechanism to further enhance the capabilities of artificial intelligence systems. The core of this mechanism is the encapsulation of domain-specific AI capabilities into independent skill description files. These files are typically structured text files (such as Markdown, YAML, or JSON) containing detailed instructions to guide AI models in performing specific tasks. The system provides mechanisms for installing and loading these skill description files. To facilitate user skill management, the skill installation mechanism is integrated into a graphical skill management interface that supports various flexible installation methods. For example, users can install skills online by connecting to a remote skill distribution platform, similar to the experience of a mobile app store; they can also quickly install skills by entering their unique identifier; and offline installation is also supported via a local file upload dialog box, where users can select and upload a skill description file package packaged in a specific format. These diverse installation methods greatly enrich the sources of skills and promote the formation of a skill ecosystem.

[0041] Once a skill is installed, its loading mechanism is activated when processing user requests. For example... Figure 6 As shown, before generating the final prompt word submitted to the AI ​​model, the system dynamically matches and loads the corresponding skill description file S601 from the installed skill library based on the user's text request, using methods such as intent recognition or keyword matching. If the match is successful, the system executes step S602, which injects the workflow instructions contained in the skill description file into the prompt word. Specifically, "injecting workflow instructions into the prompt word" means merging a series of structured instructions, context examples, output format requirements, or role-playing settings defined in the skill description file for completing a specific task, as part of the system-level prompt, with the aforementioned context information retrieved from the knowledge base and the user's original request. For example, a complete prompt word structure might be: "[Skill Workflow Instruction] Based on the following background knowledge: [Context Information] Please complete the user's request: [User Text Request]". In this way, the behavior of the general AI model (whether remote or local) will be constrained and guided by these instructions, causing it to exhibit professional and expected behavioral patterns. Finally, in step S603, the system submits the combined prompt word to the AI ​​model, enabling it to perform domain-specific tasks with high quality.

[0042] In the specific implementation of the local AI processing module, to maximize performance while ensuring universality, this invention adopts a hardware-adaptive strategy. The local AI processing module typically includes a CPU-based inference engine and uses a large language model format such as GGUF, which is optimized for CPUs and supports different precision quantizations. However, relying solely on the CPU can be slow when processing large models, impacting user experience. Therefore, the method of this invention first performs hardware capability detection when loading the local AI processing module. This step detects whether the host computer has a compatible graphics processing unit (GPU) and the corresponding computing environment, such as NVIDIA's CUDA toolkit or AMD's ROCm environment. If an available GPU environment is detected, the system prioritizes and loads an inference backend that supports GPU acceleration, offloading most of the computational tasks of model inference to the GPU, thereby achieving a performance improvement of several times or even tens of times. If the hardware capability detection does not detect an available GPU, or the GPU model is incompatible or the driver is incorrect, the system automatically reverts to loading a general-purpose, purely CPU-based inference engine. This design ensures that the system can run at full power on high-performance desktops with powerful GPUs, while also running stably on thin and light laptops with only integrated graphics cards, thus achieving a clever balance between extreme performance and broad compatibility.

[0043] From a system implementation perspective, the above method is accomplished by a series of collaborative modules. A portable artificial intelligence system includes: a removable storage medium on which an artificial intelligence application is deployed; a runtime environment module pre-installed in the removable storage medium and configured to: in response to the removable storage medium being connected to a host computer, load a corresponding installation-free runtime environment to run the artificial intelligence application based on the operating system type of the host computer, without modifying the system configuration of the host computer; and an AI model invocation module configured to: upon responding to a user request, detect the network connection status of the host computer, and adaptively select between a remote AI processing module and a local AI processing module pre-installed on the removable storage medium based on the network connection status to process the user request. These modules together constitute the core system architecture of this invention.

[0044] Those skilled in the art will understand that all or part of the steps in the above method can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, such as a read-only memory, random access memory, magnetic disk, or optical disk. When the computer program is executed by a processor, the above method can be implemented.

[0045] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement all or part of the steps in the above method.

[0046] As a specific application scenario, suppose a user inserts a removable storage medium pre-installed with this invention into a computer running Windows 11. The user double-clicks to start the script, and the system begins running. The system first recognizes the current Windows environment and loads the pre-installed, portable runtime environment. Next, the system performs a network scan, finds the computer's network connection is normal, and enters online mode, preparing to call the configured remote AI processing module. The user accesses the local web interface through a browser, enters "What's the weather like today?" in the dialog box, and the request is sent to the remote AI processing module, which quickly returns the local weather information. Subsequently, the user removes the removable storage medium and boards a train without Wi-Fi. He inserts the same removable storage medium into his laptop. After the system starts, it finds that it cannot connect to any external network during the network scan phase and automatically switches to offline mode. The system loads a local AI processing module pre-installed on the removable storage medium. The user then asks, "Write a poem about traveling alone," and the request is received by the local AI processing module, which uses the laptop's CPU for inference calculations. Although the speed may not be as fast as in online mode, the system still successfully generates and returns a poem that meets the requirements a few seconds later. This scenario fully demonstrates the seamless switching and core capability retention of this invention under different network environments, reflecting its high adaptability and practicality.

[0047] Building upon the above embodiments, this invention also supports convenient remote control functionality. Continuing with the previous scenario, suppose a user leaves a removable storage medium on their office computer and has logged into an account dedicated to a third-party message relay service via the system's built-in remote control function. On their way home from get off work, the user suddenly remembers needing a summary of a document processed that afternoon. They take out their phone, open the client of the third-party message relay service, find the contact representing the AI ​​system, and send a message: "Summarize the 'Phoenix Project' document I asked you to process this afternoon, just the key points." The system running on the office computer receives this instruction through a long-lived connection with the third-party message relay service server, then retrieves relevant documents from its local knowledge base, calls the AI ​​model to generate a summary, and sends the summary back to the user's phone via the third-party message relay service. The entire process requires no public IP address on the office computer and no complex network settings from the user.

[0048] The local knowledge base function of this invention has also demonstrated significant value in practical work. Users can directly drag and drop a large number of documents generated in their daily work into a designated folder on a removable storage medium. The knowledge base building service in the system's background will automatically detect the new files and initiate the processing flow. For example, for a complex PDF document, the system will first attempt to use a high-performance parsing library; if that fails, it will automatically downgrade to a backup library to ensure maximum text extraction. The extracted text is segmented, vectorized, and stored in a local vector database. After a period of time, when a user needs to query information, they can ask a question like in a chat: "What is the budget for the 'Phoenix Project'?" After receiving the question, the system will convert it into a vector, perform a semantic search in the knowledge base, find the most relevant original text fragments, and combine these fragments as context to generate a precise answer: "According to the document you provided, 'Phoenix Project Budget Planning.xlsx,' the total project budget is 5 million yuan." To handle more complex tasks, users can leverage the scalable skill mechanism of this invention. For example, a project manager needs to create a presentation PPT based on a large amount of project documentation. He can use the system's built-in skill management interface to search for and install a "PPT Generation" skill description file from a remote skill store. After installation, he instructs the system: "Generate a 5-page project presentation PPT outline based on all the documents I gave you last time regarding the 'Phoenix Project'." The system, through intent recognition, matches the "PPT Generation" skill. When processing the request, the system injects the workflow instructions defined in the skill description file at the beginning of the prompt, then appends relevant information retrieved from the knowledge base and the user's original request, finally sending them together to the AI ​​model. Ultimately, the AI ​​model generates a well-structured and detailed PPT outline, rather than just generic text.

[0049] The hardware adaptability of this invention is also demonstrated in switching between different devices. Assume a user has a desktop computer equipped with a high-performance GPU and a thin and light laptop with only a CPU and integrated graphics. When using this system on the desktop in offline mode, the system detects an available GPU environment during the initialization of the local AI processing module and loads a GPU-accelerated inference backend, resulting in very fast local inference speed. However, when the user inserts the same removable storage medium into the thin and light laptop, the system cannot detect a compatible GPU environment and automatically reverts to pure CPU mode. Although the response speed is slower, the system still functions normally, ensuring the availability of core functions. This adaptability allows users to obtain the optimal performance experience under the current hardware conditions, regardless of the device they use.

[0050] To more comprehensively demonstrate the integrated application value of this invention, a solution in a secure, offline environment can be envisioned. An employee needs to process a confidential report on a computer using the domestically developed UOS operating system, physically isolated from the external network. First, in the preparation phase, the employee, in an environment with internet access, stores necessary reference materials on a fully encrypted removable storage medium that has the system deployed, and installs professional skill description files such as "code review" and "report generation." After entering the intranet office area, the employee inserts the removable storage medium into the UOS computer. The system automatically recognizes the Linux kernel-based UOS system and loads the corresponding portable runtime environment. Upon detecting no external network connection, the system automatically activates the local AI processing module and selects the optimal inference backend based on hardware configuration. During the work process, the employee can interact with the AI, query the local knowledge base, use the "report generation" skill to generate a draft report, or use the "code review" skill to check the code. The entire process is performed locally, and the data does not leave the removable storage medium. After the work was completed, the employee removed the removable storage medium, leaving no information on the host machine. All data was physically taken away with the medium, fully meeting the stringent requirements of this confidential environment for data security and physical isolation.

[0051] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0052] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A portable multi-platform artificial intelligence system, characterized in that, include: A removable storage medium on which artificial intelligence applications are deployed; A runtime environment module, pre-installed in the removable storage medium, is configured to: in response to the removable storage medium being connected to a host computer, load a corresponding installation-free runtime environment to run the artificial intelligence application based on the operating system type of the host computer, without modifying the system configuration of the host computer; and, An AI model invocation module is configured to: detect the network connection status of the host computer when responding to a user request, and adaptively select between a remote AI processing module and a local AI processing module pre-installed on the removable storage medium based on the network connection status to process the user request.

2. A method for operating a portable multi-platform artificial intelligence system, characterized in that, include: In response to a removable storage medium being connected to a host computer, a corresponding runtime environment is loaded from a plurality of pre-installed portable runtime environments stored in the removable storage medium, based on the operating system type of the host computer, to run an artificial intelligence system deployed on the removable storage medium; When responding to a user request, the network connection status of the host computer is detected, and based on the network connection status, an adaptive selection is made between a remote AI processing module and a local AI processing module pre-installed on the removable storage medium to process the user request.

3. The method for operating a portable multi-platform artificial intelligence system according to claim 2, characterized in that, The method further includes: By establishing a communication connection with a server of a third-party message relay service, receiving remote instructions forwarded by the third-party message relay service, and controlling the operation of the artificial intelligence system according to the remote instructions; The third-party message relay service is at least one of the following: a publicly available instant messaging platform, an email server, or an IoT message queue service. The establishment of the communication connection includes: when the artificial intelligence system is initialized, actively establishing a long polling connection or a WebSocket persistent connection with the server of the third-party message relay service, so as to continuously listen for and receive remote instructions forwarded by the third-party message relay service as a client; The user document data, the local knowledge base index, and the user interaction data generated during the processing of the user text request are encrypted and stored in the removable storage medium. The encrypted storage includes: encrypting the entire user data partition of the removable storage medium using full-disk encryption technology, or encrypting various types of user data stored in the removable storage medium, such as user configuration files, API keys, conversation log data, source files and indexes of personal knowledge bases, and installed skill module data, at the file level or database level.

4. The method for operating a portable multi-platform artificial intelligence system according to claim 2, characterized in that, The steps for processing the user request include: The system automatically processes user document data stored in the removable storage medium to construct a local knowledge base index associated with the user document data. When responding to a user text request that is a user request, context information related to the user text request is retrieved from the local knowledge base index; The context information is combined with the user's text request to generate a prompt word; The prompt is submitted to the adaptively selected remote AI processing module or local AI processing module to obtain a response text based on the context information.

5. The method for operating a portable multi-platform artificial intelligence system according to claim 4, characterized in that, The step of automatically processing user document data stored in the removable storage medium to construct a local knowledge base index associated with the user document data includes: A multi-level degradation parsing strategy is adopted for content extraction to extract text information from the user document data for constructing the local knowledge base index. The degradation parsing strategy includes: for a specific file format, preferentially calling a high-performance dedicated third-party parsing library pre-installed in the installation-free runtime environment for parsing; When the dedicated third-party parsing library fails to load due to environmental limitations or an error occurs during the parsing process, it automatically downgrades and switches to use a more general standard library or alternative parsing logic for a second parsing attempt. The step of retrieving context information related to the user text request from the local knowledge base index when responding to the user text request as a user request specifically includes: First, natural language understanding is performed on the user text request to extract the core intent and keywords. Then, using the local knowledge base index, semantic similarity retrieval based on word vectors or traditional keyword weighted retrieval are performed to find several document fragments that are semantically most relevant to the user text request from the user document data as the context information, and sort them by relevance.

6. The method for operating a portable multi-platform artificial intelligence system according to claim 2, characterized in that, The method further includes: Encapsulate domain-specific AI capabilities into independent skill description files, which contain structured workflow instructions for guiding AI models to perform specific tasks; and provide mechanisms for installing and loading the skill description files. The installation mechanism is integrated into a graphical skill management interface and supports at least one of the following installation methods: -By connecting to a remote skills distribution platform maintained by a service provider or community, you can browse, search, and download and install skills with one click. - Quick installation can be performed by directly entering the unique identifier of a known skill in the management interface; - Select and upload a skill description file package packaged in a specific format via the local file upload dialog box for offline installation.

7. The method for operating a portable multi-platform artificial intelligence system according to claim 6, characterized in that, The mechanism for loading the skill description file includes: Before generating the prompt word, the corresponding skill description file is dynamically matched and loaded according to the user text request, and the workflow instructions contained in the skill description file are injected into the prompt word; Specifically, injecting workflow instructions into the prompts means incorporating a series of structured instructions, context examples, output format requirements, or role-playing settings defined in the skill description file for completing specific tasks into a system-level prompt. This is combined with the context information and the user text request to guide the general artificial intelligence model in the remote AI processing module or the local AI processing module, constraining and guiding its behavior so that it performs specific tasks according to a preset professional process.

8. The method for operating a portable multi-platform artificial intelligence system according to claim 2, characterized in that, The local AI processing module includes a CPU-based inference engine and uses a quantized large language model in GGUF format. The method further includes: when loading the local AI processing module, first performing hardware capability detection to check whether the host computer has a compatible graphics processor and corresponding computing environment; if an available GPU environment is detected, then prioritizing the selection and loading of an inference backend that supports GPU acceleration to obtain higher performance; if no GPU environment is detected or the environment is incompatible, then automatically falling back to loading the CPU-based inference engine to ensure that the local AI processing module can run stably on host computers with different hardware configurations.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 2 to 8.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 2 to 8.

Images