Method, apparatus and electronic device for deploying artificial intelligence system

Abstract

This disclosure relates to the field of software engineering and provides a method, apparatus, and electronic device for deploying an artificial intelligence (AI) system. The method includes: responding to a deployment instruction from the AI ​​system, loading a logical document corresponding to the AI ​​system, the logical document including task description information written in natural language; determining at least one executable task of the AI ​​system and its execution flow based on the task description information; establishing a communication connection with a server of the AI ​​system and listening to user-input interaction information, the interaction information being used to match a target task among the at least one executable task. The method provided by this disclosure deploys the AI ​​system through a logic document-driven approach, without needing to convert the task execution logic into programming language code, thereby effectively improving the deployment efficiency of the AI ​​system.

Description

Technical Field

[0001] This disclosure relates to the field of software engineering technology, and in particular to a method, apparatus and electronic device for deploying an artificial intelligence system. Background Technology

[0002] During the deployment of an artificial intelligence system, it is necessary to write professional programming language code based on the system's workflow, decision-making and planning logic, and then implement the deployment of the artificial intelligence system through standard software engineering processes.

[0003] However, this deployment method requires highly skilled personnel to write programming language code, and the writing process is time-consuming, resulting in low deployment efficiency. It may also introduce misunderstandings, affecting the actual operation of the artificial intelligence system. Summary of the Invention

[0004] To address the aforementioned technical problems, this disclosure provides a method, apparatus, and electronic device for deploying an artificial intelligence system, enabling the deployment of the system through a document-driven deployment approach, thereby improving deployment efficiency.

[0005] The first aspect of this disclosure provides a method for deploying an artificial intelligence system, comprising: In response to the deployment instructions of the artificial intelligence system, the corresponding logical document of the artificial intelligence system is loaded, which includes task description information written in natural language. Based on the task description information, determine at least one executable task of the artificial intelligence system and the execution process of at least one executable task; Establish a communication connection with the server of the artificial intelligence system and listen to the user's input interaction information, which is used to match the target task in at least one executable task.

[0006] A second aspect of this disclosure provides a deployment apparatus for an artificial intelligence system, comprising: The loading module is used to load the corresponding logical document of the artificial intelligence system in response to the deployment instructions of the artificial intelligence system. The logical document includes task description information written in natural language. The parsing module is used to parse task description information and determine at least one executable task of the artificial intelligence system and the execution flow of at least one executable task. The communication module is used to establish a communication connection with the server of the artificial intelligence system and to listen to the interactive information input by the user. The interactive information is used to match the target task in at least one executable task.

[0007] A third aspect of this disclosure provides a computer-readable storage medium storing a computer program for executing the deployment method of the artificial intelligence system proposed in the first aspect of this disclosure.

[0008] A fourth aspect of this disclosure provides an electronic device including a processor; a memory for storing processor-executable instructions; and a processor for reading executable instructions from the memory and executing the executable instructions to implement the deployment method of the artificial intelligence system proposed in the first aspect.

[0009] The technical solution provided in this disclosure, by loading the logic document corresponding to the artificial intelligence system, can parse the task description information written in natural language included in the logic document, thereby determining at least one task that the artificial intelligence system can execute after deployment, that is, at least one executable task of the artificial intelligence system, and determining the execution flow corresponding to each executable task. Then, based on the pre-deployed backend service program, a communication connection can be established with the server of the artificial intelligence system, enabling the artificial intelligence system to listen to the user's input interaction information based on the established communication connection and complete the deployment of the artificial intelligence system. In this way, during the deployment of artificial intelligence, by parsing the task description information in the logic document, natural language can be converted into an executable task flow, thereby realizing the deployment of the artificial intelligence system through a logic document-driven approach, without the need to convert the task execution logic into programming language code, thus effectively improving the deployment efficiency of the artificial intelligence system. Attached Figure Description

[0010] The above and other objects, features, and advantages of this disclosure will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this disclosure and form part of the specification. They are used together with the embodiments of this disclosure to explain the disclosure and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0011] Figure 1 This is a structural block diagram of a deployment system provided by an exemplary embodiment of the present disclosure.

[0012] Figure 2 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in an exemplary embodiment of this disclosure.

[0013] Figure 3 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in another exemplary embodiment of this disclosure.

[0014] Figure 4This is a flowchart illustrating a method for deploying an artificial intelligence system provided in yet another exemplary embodiment of this disclosure.

[0015] Figure 5 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the fourth exemplary embodiment of this disclosure.

[0016] Figure 6 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the fifth exemplary embodiment of this disclosure.

[0017] Figure 7 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the sixth exemplary embodiment of this disclosure.

[0018] Figure 8 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the seventh exemplary embodiment of this disclosure.

[0019] Figure 9 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the eighth exemplary embodiment of this disclosure.

[0020] Figure 10 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the ninth exemplary embodiment of this disclosure.

[0021] Figure 11 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the tenth exemplary embodiment of this disclosure.

[0022] Figure 12 This is a structural diagram of an artificial intelligence deployment apparatus provided in an exemplary embodiment of this disclosure.

[0023] Figure 13 This is a structural diagram of an electronic device provided by an exemplary embodiment of the present disclosure. Detailed Implementation

[0024] To explain this disclosure, exemplary embodiments according to this disclosure will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments of this disclosure, and it should be understood that this disclosure is not limited to the exemplary embodiments described herein.

[0025] It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of this disclosure.

[0026] Application Overview With the continuous development of artificial intelligence (AI) technology, AI systems have been increasingly widely used. Deploying an AI system typically requires converting its execution logic—including the task flow, task scheduling strategies, and user interaction rules—into programming language code, such as Java, Python, or C++. This process then involves compilation, building, environment configuration, and deployment. However, converting execution logic into programming language code is time-consuming, reducing deployment efficiency. Furthermore, it can introduce misunderstandings, leading to discrepancies between the deployed AI system's interactive effects and its design intent, negatively impacting user experience. This deployment method often requires complex runtime environment configurations, has a high technical threshold, and is not easily accessible to non-technical personnel for deployment and adjustments.

[0027] When the task execution logic of an artificial intelligence system needs to be adjusted, it requires the entire process of modifying code, recompiling, testing, and deploying. The response cycle is long, and each modification carries the risk of introducing new deviations, making rapid iteration difficult.

[0028] Furthermore, this deployment method requires specialized technical personnel to convert programming language code, which presents a high technical barrier, making it impossible for non-technical personnel to directly participate in deployment and adjustments. At the same time, the low readability of task execution logic in programming language code makes it difficult to quickly understand the complete execution flow, increasing the difficulty and cost of maintaining, upgrading, and handing over the AI ​​system.

[0029] To address the aforementioned technical problems, this disclosure provides a method for deploying an artificial intelligence system. By loading the logical document corresponding to the artificial intelligence system, the task description information written in natural language included in the logical document can be parsed to determine at least one task that the artificial intelligence system can execute after deployment, i.e., at least one executable task of the artificial intelligence system. At the same time, the execution flow corresponding to each executable task is determined.

[0030] Then, based on the pre-deployed backend service program, a communication connection can be established with the server of the artificial intelligence system, enabling the AI ​​system to listen to the user's input interaction information. Upon receiving the interaction information, the system can match the corresponding executable task based on the interaction information and execute the executable task according to the corresponding execution flow, thereby completing the deployment of the artificial intelligence system.

[0031] In this way, during the deployment of artificial intelligence, by parsing the task description information in the logic document, natural language can be converted into an executable task flow, thereby enabling the deployment of artificial intelligence systems through a logic document-driven approach, without having to convert the task execution logic into programming language code, thus effectively improving the deployment efficiency of artificial intelligence systems.

[0032] Furthermore, the task description information in the logic document can be written in natural language, so no specialized technical personnel are required to write the logic document. When the AI ​​system needs to modify or add tasks, only the corresponding logic document needs to be modified or added, reducing the difficulty and cost of maintaining, upgrading, and handing over the AI ​​system.

[0033] Exemplary System Figure 1 This is a structural block diagram of a deployment system provided by an exemplary embodiment of the present disclosure.

[0034] like Figure 1 As shown, in an exemplary embodiment, the deployment system 10 of this disclosure is used for the deployment of an artificial intelligence system. The deployment system 10 may adopt a layered architecture, including a document layer 11, an execution layer 12, a reference layer 13, and a version layer 14.

[0035] In one implementation, document layer 11 carries the entire task execution logic of the artificial intelligence system in the form of logical documents and reference documents. Both logical documents and reference documents can be written in natural language and structured formats, such as SKILL format. SKILL format is a structured document format used to describe the executable tasks of an artificial intelligence system. This format writes documents through metadata definition, text data description, and reference instruction extension. As the carrier of executable logic, document layer 11 replaces traditional programming language code and is used to define the executable tasks, execution flow, triggering conditions, interaction rules, and data processing logic of the artificial intelligence system. By referencing the additional execution logic in the reference documents through the reference mechanism, modular reuse and dynamic modification of application documents can be achieved, supporting semantic matching and task scheduling based on user interaction information. At the same time, as the core object of version management, the reference documents support snapshot storage, incremental updates, and historical rollback, enabling the artificial intelligence system to complete deployment, iteration, and evolution without code conversion, reducing the deployment threshold, improving iteration efficiency, and maintainingability.

[0036] In one possible implementation, the logical document is in SKILL document format, including a metadata area, a main logical area, a reference instruction area, and a supplementary explanation area. The metadata area uses YAML frontmatter format to define document metadata, including document name, description, trigger conditions, author information, version number, etc. The metadata area is located at the beginning of the logical document and delimited by separators, such as "---". The main logical area is written in Markdown format, using hierarchical chapters to describe task flows, decision rules, and data processing logic. The chapter structure supports hierarchical structures, such as "# Heading 1" and "## Heading 2", facilitating document parser recognition and navigation. The reference area defines the reference paths to other reusable referenced documents, such as "read references / interaction.md". The reference mechanism supports relative and absolute paths and multi-level references. The supplementary area is optional and contains comments, examples, notes, and other auxiliary information to help understand the task execution logic described in the logical document.

[0037] In one implementation, Execution Layer 12 deploys the core operational unit of the AI ​​system through a logic document-driven approach. It loads and parses the natural language logic documents from Document Layer 11, extracts executable tasks and execution flows based on task description information, and constructs control structures such as sequence, branching, and looping. Execution Layer 12 can establish communication connections with the server, listen to user interaction information, determine the target task from the executable tasks through semantic matching, and complete the scheduled execution. Execution Layer 12 is also responsible for maintaining the execution context, handling execution status, managing task interruption and recovery, calling external tools and large language models, and collaborating with Reference Layer 13, Version Layer 14, and the security sandbox layer to complete the entire process of the AI ​​system's operation in a secure, isolated environment. This achieves the conversion from document logic to actual execution capabilities, ensuring the AI ​​system executes related tasks stably, efficiently, and securely.

[0038] In one implementation, the execution layer 12 may include a document parser, a process planner, and an execution controller. The document parser may implement the following functions: 1. Metadata Parsing: Extracts metadata information from the YAML frontmatter, including document name, function description, triggering conditions, etc. Parsing is handled by a YAML parsing library (such as PyYAML), supporting complex nested structures.

[0039] 2. Structure Parsing: Parses the hierarchical structure of the Markdown document and extracts the chapter tree. The parsing is based on Markdown heading syntax (#, ##, ###) to construct the parent-child relationships between chapters.

[0040] 3. Instruction Extraction: Extract execution instructions from the main body of the logical document, including task descriptions, conditional statements, loop control, etc. Instruction extraction uses pattern matching technology to identify specific instruction patterns.

[0041] 4. Reference Resolution: Parses reference instructions in the logical document, extracting the target document path and reference method. Reference resolution supports relative path conversion, circular reference detection, etc.

[0042] The process planner can perform the following functions: 1. Task Identification: Identify tasks that need to be performed from the logical document, including initialization tasks, processing tasks, and cleanup tasks. Task identification is based on the analysis of chapter titles and keywords in the main text of the logical document.

[0043] 2. Dependency Analysis: Analyzes the dependencies between tasks to determine the execution order. Dependency analysis is based on data dependencies and control dependencies between tasks and can construct a directed acyclic graph.

[0044] 3. Process Construction: Construct the execution process based on tasks and dependencies, including control structures such as sequential execution, conditional branching, and loop iteration. Process construction is implemented using state machines or workflow engines.

[0045] 4. Optimization and Adjustment: Optimize the execution process based on the execution context and historical data, such as task parallelization, cache reuse, and lazy loading. Optimization and adjustment improve execution efficiency.

[0046] The execution controller can perform the following functions: 1. Task Scheduling: Schedules task execution according to a pre-planned process, managing task start, pause, resumption, and termination states. The scheduler supports various scheduling strategies, including priority scheduling, time scheduling, and event scheduling.

[0047] 2. Context Management: Maintains the execution context, including information such as variables, state, and intermediate results. The context adopts a stack structure, supporting nested calls and scope management.

[0048] 3. Conditional Judgment: Determines the execution path based on the current state and preset conditions, supporting conditional structures such as if-then-else and switch-case. Conditional judgment is based on expression evaluation and rule matching.

[0049] 4. Loop Control: Implements loop iteration control, supporting loop structures such as while, for, and until. Loop control maintains the state of the loop counter, exit condition, etc.

[0050] In one implementation, the reference layer 13 is used to realize the modular reuse and dynamic expansion of referenced documents, and is responsible for managing, parsing, loading, and embedding referenced documents. Reference layer 13 can locate and read reusable referenced documents based on reference instructions in the main logic document, and merge their execution logic into the overall system execution flow; it also supports dynamic incremental modification of referenced documents during system operation, enabling the self-evolution of the AI ​​system's capabilities. Reference layer 13 can work in conjunction with version layer 14 to automatically generate document snapshots before modification, supporting version traceability and rollback, reducing business logic coupling, improving system maintainability and scalability, and working with execution layer 12 to complete the document-driven full-process execution.

[0051] In one implementation, version layer 14 is used to manage the version, history, and secure rollback of the logical documents and referenced documents of the artificial intelligence system. By automatically generating snapshots before modifying referenced documents and using incremental algorithms to store version differences, storage space usage is reduced. Version layer 14 can maintain a version index, supporting quick retrieval of historical versions based on version number, timestamp, and modification record. In the event of dynamic system evolution or abnormal document modification, version layer 14 can quickly roll back the document to a specified historical version, completing format, logic, and conflict checks during the rollback process. Version layer 14 works in conjunction with reference layer 13 to manage the version chains of the main logical documents and referenced documents respectively, ensuring that the artificial intelligence system is traceable, recoverable, and operates securely and stably during dynamic modification and self-evolution.

[0052] In one possible implementation, the deployment system 10 of this disclosure embodiment may further include a security layer. The security layer can be used to provide a secure and isolated operating environment for the artificial intelligence system based on a preset security isolation mechanism, such as a sandbox isolation mechanism. By implementing file access isolation, tool call whitelisting, system resource restrictions, and network access isolation, a restricted and controllable execution sandbox is constructed. The sandbox isolation mechanism restricts documents to accessing only specified directories, calling whitelisted tools, using controlled resources, and accessing authorized network addresses, preventing illegal operations, resource abuse, and data leakage. The sandbox isolation mechanism covers the entire process of document loading, logic execution, reference modification, and version update, and works in conjunction with the execution layer 12, reference layer 13, and version layer 14 to ensure that the artificial intelligence system operates securely, stably, and reliably in isolated environments such as in-vehicle, edge, and cloud environments.

[0053] The deployment system 10 provided in this embodiment loads a logic document corresponding to the artificial intelligence system. It then parses the task description information written in natural language within the logic document to determine at least one task that the artificial intelligence system can execute after deployment—that is, at least one executable task of the artificial intelligence system. Simultaneously, it determines the execution flow corresponding to each executable task. Then, based on a pre-deployed backend service program, it establishes a communication connection with the server of the artificial intelligence system, enabling the artificial intelligence system to listen to user input interaction information. Upon receiving interaction information, it matches the corresponding executable task based on the interaction information and executes the executable task according to the corresponding execution flow, thereby completing the deployment of the artificial intelligence system. During the deployment of the artificial intelligence system, parsing the task description information in the logic document converts natural language into executable task flows, thus enabling the deployment of the artificial intelligence system through a logic document-driven approach, without needing to convert the task execution logic into programming language code, thereby effectively improving the deployment efficiency of the artificial intelligence system.

[0054] In one possible implementation, deployment system 10 can be built using a lightweight runtime design. A runtime is software that provides basic support services for a specific programming language or program. Through the runtime, core capabilities such as document loading, AI capability invocation, tool invocation, and execution control can be provided, without handling specific business logic. These core capabilities are extended through plugins or modules. The runtime only starts when needed and releases resources after completing its tasks. On-demand startup reduces resource consumption, making it suitable for serverless architectures. Techniques such as lazy loading and caching preheating optimize the startup process and reduce startup latency. Resources, including memory, file handles, and network connections, are proactively released after tasks are completed to prevent resource leaks.

[0055] For example, the runtime can be deployed in edge devices, such as in-vehicle devices and smart home devices. This reduces latency, protects data privacy, and supports offline use during deployment. The runtime can also be deployed in the cloud; logic documents can be uploaded to the cloud via a web interface for the runtime to load and deploy, eliminating the need for local runtime installation and supporting remote access and collaboration. The runtime can also be packaged as a Docker container, enabling container orchestration and deployment, providing a consistent deployment environment, and facilitating scalability and management. This runtime design allows for the isolation of different users' logic document execution environments when multiple users deploy using the same runtime, improving resource utilization and supporting Software as a Service (SaaS).

[0056] In addition, the runtime provides a command-line argument interface, supporting the specification of documents, setting options, and passing parameters. The command-line interface conforms to the Portable Operating System Interface (POSIX) standard. It supports background operation, outputting the execution process to a log file, making it suitable for unattended scenarios. The runtime can also handle system signals such as termination signals (SIGTERM) and interrupt signals (SIGINT), enabling graceful exit and ensuring proper resource release and state preservation. Standardized exit codes are defined, corresponding to execution states such as success, failure, and timeout, facilitating scripted handling of task execution and error detection.

[0057] Exemplary methods Figure 2 This is a flowchart illustrating a deployment method for an artificial intelligence system provided in an exemplary embodiment of this disclosure. Embodiments of this disclosure can be applied to electronic devices, such as... Figure 2 As shown, it includes the following steps: Step S100: In response to the deployment instruction of the artificial intelligence system, load the logical document corresponding to the artificial intelligence system, which includes task description information written in natural language.

[0058] In one implementation, before the AI ​​system is deployed, deployment instructions for the AI ​​system can be listened for. These instructions can come from command line calls, user interactions with electronic devices, program triggers, or automatic system startup.

[0059] Upon receiving a deployment command, the document path, startup parameters, and runtime configuration can be extracted from it to determine the logical document to be loaded. Based on the path information carried in the deployment command, the corresponding logical document is located in the file system. For example, the logical document uses a standardized SKILL format, typically a master document (e.g., SKILL.md), stored in a system-specified working directory, and includes task description information written in natural language.

[0060] By listening to the deployment instructions of the artificial intelligence system, this step allows the system to respond to the deployment instructions by locating the corresponding logical document in the file system, loading the logical document, and reading its contents. This prepares the system for determining executable tasks and execution processes in the subsequent deployment process.

[0061] Step S200: Based on the task description information, determine at least one executable task of the artificial intelligence system and the execution process of at least one executable task.

[0062] In one implementation, after loading the logic document, task description information written in natural language can be read from the logic document. Then, by parsing the task description information in the logic document, at least one executable task of the artificial intelligence system can be determined based on the task description information. At the same time, the natural language can be converted into a task flow that can be executed by the electronic device, that is, into an execution flow corresponding to each executable task.

[0063] In this way, after the artificial intelligence system is deployed, it can directly match the corresponding executable tasks based on the user's input interaction information, and execute the executable tasks according to the execution process corresponding to the executable tasks, thus completing the interaction process of the artificial intelligence system.

[0064] By parsing the task description information, this method enables the determination of at least one executable task for the artificial intelligence system. Furthermore, it converts natural language into a task flow executable by the electronic device, allowing the device to directly execute the corresponding task without needing to convert the task execution logic into programming language code to construct the task flow. This effectively improves the deployment efficiency of the artificial intelligence system.

[0065] Step S300: Establish a communication connection with the server of the artificial intelligence system and listen to the user's input interaction information. The interaction information is used to match the target task in at least one executable task.

[0066] By employing this approach, once the executable tasks and their corresponding execution flows are determined, a communication connection can be established with the AI ​​system's server based on the pre-deployed backend service program. For example, the execution controller in the execution layer can establish a communication link with the pre-deployed backend service program (server.js) of the AI ​​system in the electronic device, thus completing the communication connection with the AI ​​system.

[0067] After establishing a communication connection with the AI ​​system's server, the system can begin listening to user interactions. Upon receiving user interactions, the AI ​​interaction system can perform semantic matching to match corresponding executable tasks. Then, based on the execution flow of the executable task, it executes the task, completing the interaction with the user.

[0068] The technical solution of this disclosure, by loading the logic document corresponding to the artificial intelligence system, can parse the task description information written in natural language included in the logic document, thereby determining at least one task that the artificial intelligence system can execute after deployment, that is, at least one executable task of the artificial intelligence system, and determining the execution flow corresponding to each executable task. Then, based on the pre-deployed backend service program, a communication connection is established with the server of the artificial intelligence system, enabling the artificial intelligence system to listen to user input interaction information. Upon receiving interaction information, it can match the corresponding executable task based on the interaction information and execute the executable task according to the corresponding execution flow, thereby completing the deployment of the artificial intelligence system. During the deployment of artificial intelligence, by parsing the task description information in the logic document, natural language can be converted into executable task flow, thereby enabling the deployment of the artificial intelligence system through a logic document-driven approach, without the need to convert the task execution logic into programming language code, thus effectively improving the deployment efficiency of the artificial intelligence system.

[0069] Figure 3 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in another exemplary embodiment of this disclosure.

[0070] like Figure 3 As shown, in one embodiment, step S200 may include: Step S210: Parse the metadata in the task description information to determine at least one executable task and at least one functional description information of the executable task.

[0071] In one implementation, after the logic file is loaded, the execution layer's document parser can begin parsing the task description information in the logic document. First, the document parser can parse the metadata in the task description information to determine at least one executable task corresponding to the logic document. For example, the document parser can identify the YAML metadata area and read the task name field related to the task from the metadata according to preset rules, thus extracting the executable task corresponding to the logic document based on the task name field. For example, the task name field may include a question-and-answer task, a query task, a control task, etc.

[0072] The document parser can then further parse the function description fields in the metadata area to determine the function description information corresponding to each executable task. For example, it can extract keywords such as task purpose, processing object, output result, and usage scenario from the function description fields. Based on these keywords, the function description information corresponding to the executable task can be generated.

[0073] This step involves parsing the metadata in the task description information to determine the tasks that the logical document can execute, i.e., determining the executable tasks and their corresponding functional description information, thus preparing for the subsequent determination of the execution logic of the executable tasks.

[0074] Step S220: Parse the text data in the task description information that corresponds to the functional description information of at least one executable task, and determine the execution logic of at least one executable task.

[0075] In one implementation, after determining the executable task and the corresponding functional description information, the document parser can further read the text data corresponding to the functional description information and convert the task flow, judgment rules, processing steps, etc. corresponding to the functional description information written in natural language into execution logic that can be recognized by electronic devices.

[0076] For example, the main text data of the logical document is written in Markdown format. Based on the determined executable tasks and functional description information, the document parser locates the Markdown main text data of the corresponding chapter in the logical document and performs structured parsing of the main text content according to heading level, execution action, execution object and execution order. It extracts basic execution logic such as sequential execution, data reading and writing, external calls and state maintenance from the main text data, and identifies control logic such as condition judgment, loop iteration, jump execution and exception handling. It converts the business process described in natural language into standardized structured execution logic, providing a logical basis for the subsequent construction of a complete execution process.

[0077] Step S230: Based on the execution logic of at least one executable task, determine the execution flow of at least one executable task.

[0078] In one implementation, the process planner of the execution layer can plan the execution flow of at least one executable task based on at least one executable task parsed by the document parser and the corresponding execution logic.

[0079] For example, the process planner loads the determined executable tasks and corresponding execution logic, analyzes the data and control dependencies between tasks, and clarifies the execution sequence constraints. Based on the execution logic, it constructs a basic execution sequence, embeds conditional branching, loop iteration, jump and exception handling logic, and integrates the additional execution logic from the referenced documents. Finally, it forms a standardized executable process that includes startup, initialization, main process, branching, looping, and exception handling, providing the execution controller with a basis for directly scheduling and running task execution.

[0080] The technical solution of this disclosure embodiment performs layered parsing of task description information. First, it determines the executable tasks and corresponding functional description information of the artificial intelligence system from the metadata. Then, it determines the execution logic of each executable task based on the text data that matches the functional description information. Finally, it obtains the corresponding execution flow based on the execution logic. This structured parsing method can clearly separate task definition and logic implementation, improve the accuracy of document parsing and the standardization of execution flow construction, and enable the artificial intelligence system to efficiently and stably load and run various executable tasks based on task description information in a unified format.

[0081] Figure 4 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in yet another exemplary embodiment of this disclosure.

[0082] like Figure 4 As shown, in one embodiment, after step S230, which determines the execution flow of at least one executable task based on the execution logic of at least one executable task, step S200 further includes: Step S240: Based on the reference instructions in the logical document, load at least one reference document, which includes additional descriptive information of the executable task.

[0083] In one implementation, if the logical document contains a reference instruction, the document parser can also parse the reference instruction, thereby loading at least one referenced document based on the reference instruction.

[0084] For example, the document parser scans the text and configuration data in the logical document to locate reference instructions, such as the `references / interaction.md` instruction. Then, the document parser parses the reference instructions to obtain the reference path of the referenced document. Next, the document parser can process the reference path and locate the corresponding referenced document in the file system according to the reference path. During the reference path parsing process, relative paths and absolute paths can be distinguished, and relative paths can be converted into absolute paths that the system can recognize. Finally, the document parser can access the reference path, load the referenced document, and read the additional descriptive information of the executable tasks within the referenced document.

[0085] Understandably, the additional description information includes description information that differs from the original functional description information of the executable task. It is used to indicate the additional execution logic of the executable task during execution. By parsing the additional description information, the execution logic of the executable task can be further restricted during the execution of the executable task, thereby meeting the user's personalized needs.

[0086] Step S250: Parse the additional description information to determine the additional execution logic of the executable task.

[0087] In one implementation, the document parser loads and locates the additional descriptive information in the referenced document, identifies the execution actions, execution objects, execution timing, and execution constraints in the additional description through structured parsing, extracts additional execution logic from the additional description information for supplementary processing, extended judgment, enhanced output, correction of coverage, and reuse of component classes, and converts its natural language description into standardized execution logic that can be recognized at runtime; and associates and binds the additional execution logic with the corresponding executable tasks in the main logic document to provide logical support for subsequent embedding and updating of the execution process.

[0088] For example, referenced documents can use descriptive filenames, such as interaction strategy (interaction.md), response template (template.md), etc. Naming conventions ensure the recognizability and maintainability of referenced documents. Each referenced document can contain its own metadata, used to define information such as the version, dependencies, and update time of the referenced document.

[0089] Step S260: Embed the additional execution logic into the execution logic and update the execution flow of the executable task.

[0090] In one implementation, after the document parser parses out the additional execution logic corresponding to the referenced document, the process planner can embed the additional execution logic into the execution logic corresponding to the executable task according to the logic type and execution order of the additional execution logic, thereby updating the execution flow of the executable task.

[0091] For example, the process planner can insert additional execution logic steps, conditional judgments, loops, data processing, etc. into the corresponding positions of the original execution logic, keeping the original execution logic unchanged, and only expanding, supplementing or enhancing the original execution logic based on the additional execution logic.

[0092] The process planner loads the original execution logic and additional execution logic corresponding to the executable task, determines the embedding position and embedding method according to the logic type, and embeds the additional execution logic into the original execution logic by pre-positioning, post-positioning, middle insertion, or overwriting replacement; based on the merged complete execution logic, it reconstructs a unified execution flow, completes the update of the executable task execution flow, and makes the execution flow include the main logic and the additional logic provided by the referenced documents. Through reusable referenced documents, it realizes modular expansion and enhancement of executable tasks.

[0093] The technical solution of this disclosure embodiment loads a reference document containing additional descriptive information through a reference instruction in the logic document, parses the additional execution logic of the executable task from it, and embeds the logic into the original execution logic to update the execution process; by referencing external documents to extend the task logic, the execution process can be modularized and flexibly extended, reducing the maintenance complexity of the logic document, and making the task capabilities of the artificial intelligence system easier to iterate and expand.

[0094] Figure 5 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the fourth exemplary embodiment of this disclosure.

[0095] like Figure 5 As shown, in one implementation, step S240, loading at least one referenced document based on reference instructions in the logical document, may include: Step S241: Based on the reference instruction, determine at least one reference path, which is used to indicate the directory where the referenced document is stored.

[0096] In one implementation, the document parser can extract features from the citation instructions to extract the citation path within them. This citation path indicates the directory where the cited document is stored. The document parser can then parse the citation path to obtain the directory indicated by the citation path and locate the corresponding cited document in the file system according to that directory.

[0097] In addition, during the parsing of reference paths, relative paths and absolute paths can be distinguished. A relative path refers to the path of the referenced document relative to the logical document. For a relative path, the standardized absolute path is calculated by taking the directory where the logical document is located as the reference and combining the relative path symbols. An absolute path refers to the path where the referenced path is located in the operating system. For an absolute path, the format validity is directly extracted and verified to obtain the file directory indicated by the referenced path.

[0098] Step S242: Access the storage directory and load the referenced document in the storage directory.

[0099] In one implementation, after the document parser parses out the directory where the referenced document is stored, the reference layer calls the system file access interface to access the target directory and verify the directory's legality and access permissions based on the determined reference path and storage directory, and locates and checks the existence, format specifications and content integrity of the target referenced document.

[0100] Then, the document parser can read the complete content of the document in read-only mode through the file reading interface, perform preliminary structured parsing on the read content and convert it into memory-processable data, complete the loading and caching of the referenced document, and provide a complete data source for subsequent parsing of additional descriptive information.

[0101] The technical solution of this disclosure determines the reference path pointing to the directory where the referenced document is stored by reference instructions, and then accesses the directory to complete the loading of the referenced document. This method realizes the accurate positioning and standardized reading of the referenced document, ensures that the system can stably obtain external additional descriptive information, and improves the reliability and execution efficiency of the logic expansion process.

[0102] Figure 6 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the fifth exemplary embodiment of this disclosure.

[0103] like Figure 6 As shown, in one implementation, after step S300—establishing a communication connection with the server of the artificial intelligence system and listening to the user's input interaction information—the method further includes: Step S400: In response to the monitored interaction information being feedback information for any additional descriptive information, determine the incremental information of the additional descriptive information based on the feedback information.

[0104] In one implementation, after the AI ​​system is deployed, if the user's input is detected as feedback on any additional descriptive information, it means that the user needs to interact with the AI ​​system to modify the additional descriptive information.

[0105] For example, the additional descriptive information can be tone style. Users can input information to provide feedback on the current tone style, thereby instructing the AI ​​system to modify the current tone style. For instance, a user can input the voice message "I hope you can be more humorous." In this case, based on the user's feedback, the incremental information for the additional descriptive information can be determined to be "more humorous tone style."

[0106] Step S500: Based on the incremental information, update the referenced document corresponding to the additional description information to obtain the target referenced document.

[0107] In one implementation, once the incremental information of the additional descriptive information is determined, the additional descriptive information can be updated based on that incremental information.

[0108] For example, the referencing layer locates the referenced document to be updated based on incremental information. According to the operation type and location identifier in the incremental information, it modifies, adds, deletes, or adjusts the corresponding supplementary description information within the referenced document to complete a partial content update. Specifically, incremental modification refers to replacing a specified chapter, sentence, or paragraph in the supplementary description information with incremental information; incremental addition refers to inserting incremental information at a specified position in the supplementary description information; incremental deletion refers to removing redundant or erroneous description information from the supplementary description information; and incremental adjustment refers to adjusting information such as the sentence order, triggering conditions, and rule descriptions in the supplementary description information.

[0109] In addition, after updating the referenced document, the updated referenced document can be validated for format validity and logical rationality. After the validation is passed, the target referenced document is saved and loaded into the runtime environment to replace the original referenced document, thus realizing dynamic updating of referenced documents based on user feedback.

[0110] The technical solution of this disclosure embodiment, after the system listens to user interaction information, if it identifies the interaction information as feedback information for additional description information, then generates incremental information for additional description information based on the feedback, and updates the corresponding referenced document based on the incremental information to obtain the target referenced document; by dynamically optimizing the content of the referenced document based on user feedback, the continuous iteration and self-evolution of additional execution logic can be realized, thereby improving the adaptability and effectiveness of the artificial intelligence system's task execution.

[0111] In some implementations, the AI ​​system can modify the additional descriptive information and then generate a corresponding confirmation response. For example, if a user enters "I hope you can be more humorous," the AI ​​system recognizes this interaction as feedback on the additional descriptive information. Based on this feedback, it modifies the corresponding referenced document to obtain the target referenced document. Then, it generates a confirmation response, "Okay, I will try to make the response more interesting." The next time the user interacts with the AI ​​system, the target referenced document can be loaded, and the interaction can be based on the additional descriptive information of the target referenced document, thus adding a humorous element to the response.

[0112] Figure 7 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the sixth exemplary embodiment of this disclosure.

[0113] like Figure 7 As shown, in one implementation, before step S500, updating the referenced document corresponding to the additional description information based on the incremental information to obtain the target referenced document, the method further includes: Step S600: Generate and save a copy of the referenced document.

[0114] In one implementation, before updating the referenced document, the version layer can first create a complete copy of the referenced document before the update, so as to achieve backup and retention before modification, which can be used for subsequent version rollback, exception rollback and operation auditing.

[0115] For example, the version layer can determine the corresponding original referenced document in the file system based on the path information of the referenced document to be updated, and obtain the name, storage directory, format, and complete content of the referenced document. Then, it reads the complete data of the referenced document, creates a copy file in the version history directory, writes the read complete data of the original document into the copy file, and completes the generation and saving of the copy file corresponding to the referenced document, providing a backup basis for subsequent version rollback and abnormal rollback.

[0116] like Figure 7 As shown, in one implementation, after step S500, updating the referenced document corresponding to the additional description information based on the incremental information to obtain the target referenced document, the method further includes: Step S700: In response to receiving a rollback instruction corresponding to the referenced document, modify the target referenced document to the referenced document based on the copy file.

[0117] In one implementation, if the AI ​​system receives a rollback instruction corresponding to a referenced document, the version layer can locate the matching copy file in the version history storage directory, read the complete original content of the copy file and overwrite it with the target referenced document, thereby modifying the target referenced document to the original referenced document and realizing version rollback.

[0118] The technical solution of this disclosure adds the operation of generating and saving the corresponding copy file, and sets a version rollback mechanism after the update is completed. When a rollback instruction is received, the target referenced document can be restored to the original version based on the copy file. Through version backup and rollback capabilities, the security and traceability of the document update process are effectively guaranteed, avoiding system logic errors caused by update anomalies and improving system operation stability.

[0119] Figure 8 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the seventh exemplary embodiment of this disclosure.

[0120] like Figure 8 As shown, in one implementation, before step S200, which involves determining at least one executable task of the artificial intelligence system and the execution flow of the at least one executable task based on task description information, the method further includes: Step S800: Based on the preset security isolation mechanism, construct the operating environment of the artificial intelligence system.

[0121] In one implementation, when the deployment system starts up and begins deploying the artificial intelligence system, the security layer can build a restricted, isolated, and controllable dedicated operating environment based on a preset security isolation mechanism. This ensures that all operations of the artificial intelligence system, such as document parsing, process execution, and document updates, are performed within the isolated scope, preventing unauthorized access, resource abuse, and malicious behavior from affecting the operating system deployed by the artificial intelligence system.

[0122] The runtime environment is used to instruct the file access configuration, tool invocation configuration, resource usage configuration, and network access configuration during the deployment of the artificial intelligence system. In other words, within a runtime environment built on a security isolation mechanism, the artificial intelligence system can only run based on the file access configuration, tool invocation configuration, resource usage configuration, and network access configuration indicated by the runtime environment, to prevent unauthorized access, resource abuse, and malicious behavior by the artificial intelligence system.

[0123] For example, the security layer can provide a secure and isolated operating environment for artificial intelligence systems based on a sandbox isolation mechanism. By isolating file access, whitelisting tools, restricting system resources, and isolating network access, a restricted and controllable execution sandbox is constructed. The sandbox isolation mechanism restricts documents to accessing only specified directories, calling whitelisted tools, using controlled resources, and accessing authorized network addresses, preventing illegal operations, resource abuse, and data leakage. The sandbox isolation mechanism covers the entire process of document loading, logic execution, reference modification, and version update.

[0124] The technical solution of this disclosure embodiment constructs the operating environment of the artificial intelligence system based on a preset security isolation mechanism, and uniformly configures file access, tool invocation, resource consumption and network access; by pre-establishing a secure and controllable isolated operating environment, it can effectively limit the system access scope, prevent resource abuse and unauthorized operations, ensure the security and stability of task parsing and execution process, and improve the overall system security and reliability.

[0125] Figure 9 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the eighth exemplary embodiment of this disclosure.

[0126] like Figure 9 As shown, in one implementation, after step S300—establishing a communication connection with the server of the artificial intelligence system and listening to the user's input interaction information—the method further includes: Step S900: In response to the detected interaction information, the large language model deployed on the server side of the artificial intelligence system is used to analyze the interaction information to obtain the task intent corresponding to the interaction information.

[0127] In one implementation, after the AI ​​system is deployed, the execution controller can begin listening to user interactions. When the execution controller receives user interactions, it can input the information into a large language model deployed on the AI ​​system's server. The large language model then performs semantic understanding and intent recognition on the interactions, outputting analysis results including intent type, core parameters, and confidence levels, thereby determining the task intent corresponding to the user's input.

[0128] It is understandable that by determining the task intent, the executable task corresponding to the user's interaction information can be matched based on the task intent, and then the user's instructions can be completed by executing the executable task.

[0129] Step S1000: Match the task intent with the functional description information corresponding to at least one executable task to determine the target task corresponding to the task intent.

[0130] In one implementation, after the large language model analyzes the user's interaction information to obtain the user's task intent, the execution controller can perform semantic matching between the task intent and the functional description information of each executable task, and find the task that best matches the user's intent as the final target task to be executed.

[0131] For example, the execution controller can first use techniques such as word embedding and sentence embedding to convert task intent and functional description information into vectorized representations. The similarity between the vector corresponding to the task intent and the vector corresponding to the functional description information is calculated, for example using cosine similarity, dot product, or Euclidean distance. The similarity score reflects the degree of matching between the task intent and each functional description. Then, the top K functional descriptions by similarity are selected, and the logical documents corresponding to these functional descriptions are used as candidate documents. Based on multi-dimensional factors, including similarity, priority, and historical usage, the candidate documents are further filtered to obtain the final logical document. The executable task indicated by this document is the target task to be executed based on the user's input interaction information.

[0132] Step S1100: Execute the target task according to the target execution process corresponding to the target task.

[0133] In one implementation, once the target task is determined, the execution controller can call its corresponding target execution flow to execute the target task completely in sequence, branching, and looping logic, thereby realizing the business logic corresponding to user interaction.

[0134] For example, the execution controller loads and parses the corresponding target execution flow based on the task identifier of the target task, and initializes the task execution context and execution state; according to the execution order, conditional branches, loop rules and jump logic defined in the target execution flow, it schedules and executes each execution node in sequence, and completes operations such as file access, interface calls and logical operations in the secure isolation environment of the sandbox layer; during the execution process, it maintains the task execution state and intermediate results, and handles execution exceptions according to the preset exception handling mechanism; after the execution is completed, it outputs the task execution result and completes resource release, realizing the complete execution of the target task.

[0135] The technical solution of this disclosure embodiment, after listening to user interaction information, analyzes the interaction information through a server-side large language model to obtain the task intent, matches the task intent with the functional description information of executable tasks to obtain the target task, and completes the task execution according to the corresponding target execution process; it realizes a complete closed loop from user interaction to intent recognition, task matching and automatic execution, enabling the artificial intelligence system to intelligently schedule and execute corresponding tasks according to user input, thereby improving the intelligence level and execution efficiency of interaction response.

[0136] In one implementation, the AI ​​system can provide a transparent user experience through a user interface while performing its target task. This means the user interface is pre-generated based on ui / index.html, and during the AI ​​system's operation, it displays the user interface by calling ui / index.html.

[0137] For example, during the interaction, the AI ​​system can report the execution progress to the user through the user interface, including completed steps, the current step, and remaining steps. Progress reports are displayed through text output, progress bars, status indicators, etc. The user interface displays the execution results, including output data, status information, and error messages. The results are displayed in a structured format for easy user understanding and processing. During the interaction, user confirmation can also be provided through the user interface, requesting user confirmation at critical steps, such as modifying strategy documents or deleting important data, to prevent accidental operations. The system can also monitor user input in the user interface in real time, responding to interruption requests by pausing the execution of the target task while preserving the current state for later resumption of the target task.

[0138] Figure 10 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the ninth exemplary embodiment of this disclosure.

[0139] like Figure 10 As shown, in one implementation, step S300 involves listening to user input interaction information, including: Step S310: Read the target interaction file according to the preset time interval. The target interaction file is the latest generated interaction file in the preset folder. The interaction file is a file generated based on the user's interaction information.

[0140] In one implementation, the execution controller initiates a timed polling mechanism at preset time intervals to locate and access preset folders within the authorized scope of the security layer; it traverses the files in the preset folders and filters out interactive files according to file rules, sorts them according to the creation or modification time of each interactive file, and determines the latest generated interactive file as the target interactive file; it reads the content generated based on user interaction information in the target interactive file in read-only mode, performs format parsing and legality verification on the content, and caches it in the execution context to provide a data foundation for subsequent interactive information analysis.

[0141] For example, the execution controller reads the input.json file in a preset folder at preset time intervals. It can be understood that the input.json file is a formatted file generated based on user interaction information, used to indicate the content of the user's interaction information.

[0142] Step S320: Compare the target file identifier of the target interactive file with the historical file identifier, wherein the historical file identifier is the file identifier of the interactive file generated before the target interactive file.

[0143] In one implementation, the execution controller can read the file identifier of the target file and read the historical file identifier of the last monitoring record from the system cache. By comparing the target file identifier with the historical file identifier, it can determine whether the target file is interactive information that needs to be processed or interactive information that was processed last time and can be ignored this time.

[0144] For example, the execution controller reads the input.json file in a preset folder at preset time intervals. The input.json file is then compared with last_processed_id, which is the identifier of the historical file from the last monitoring record.

[0145] Step S330: If the target file identifier is different from the historical file identifier, then it is determined that interactive information has been detected.

[0146] In one implementation, the execution controller makes a judgment based on the comparison between the target file identifier and the historical file identifier. If they are different, it indicates that the target interactive file is a newly generated, unprocessed file, confirming that the system has detected the user's interaction information and proceeding to the subsequent interaction information parsing and processing flow. If they are the same, it is determined that no new interaction information has been generated, and the current monitoring process ends. After a preset time interval, the target file is retrieved again, and the file identifier is compared.

[0147] The technical solution of this disclosure reads the latest target interactive file generated in a preset folder at preset time intervals, compares its file identifier with the historical file identifier, and determines that new interactive information has been detected when the identifiers are inconsistent. This method of timed polling plus file identifier verification can efficiently and stably monitor user interactions, avoid repeatedly processing the same information, and reduce system resource consumption while ensuring real-time performance.

[0148] Figure 11 This is a flowchart illustrating a method for deploying an artificial intelligence system provided in the tenth exemplary embodiment of this disclosure.

[0149] like Figure 11 As shown, in one implementation, before step S300—establishing a communication connection with the server of the artificial intelligence system and listening to the user's input interaction information—the method further includes: Step S1200: In response to receiving a trigger instruction, match the trigger instruction with the trigger conditions in the logical document to obtain a matching result.

[0150] It should be noted that the matching result includes whether the trigger command meets the trigger condition or not. The trigger command meeting the trigger condition is used to indicate that the user's input interaction information should be started.

[0151] In one implementation, before the AI ​​system begins listening to user interactions, it is in a non-triggered state. At this point, the AI ​​system can listen for trigger commands and determine if the command meets the triggering conditions in the logical document. If the command meets the conditions, it indicates that the user needs to interact with the AI ​​system, and the AI ​​system can then begin listening to user interactions. Conversely, if the command does not meet the conditions, it means the AI ​​system has not detected a trigger command and there is no need for interaction.

[0152] For example, the execution controller responds to the received trigger command by parsing and formatting the trigger command; extracting preset trigger conditions from the logic document; matching the command type, parameters, and system running status of the trigger command with the trigger conditions item by item to generate a matching result containing whether the trigger conditions are met or not; if the matching result is that the trigger conditions are met, then the process of listening to user input interaction information is started; if the trigger conditions are not met, then the trigger command is ignored and listening is not started.

[0153] The technical solution of this disclosure, before starting to listen to user interaction information, first receives a trigger command and matches it with preset trigger conditions in a logical document. The listening process is only started when the trigger command meets the trigger conditions. Through the trigger condition verification mechanism, controllable initiation and security verification of the listening behavior are achieved, avoiding invalid or illegal initiation of listening and improving the standardization and controllability of system operation. The solutions involved in this disclosure are not limited to the embodiments mentioned above.

[0154] Exemplary device The deployment method of the artificial intelligence system provided in the embodiments of this disclosure has been described above. It is understood that the deployment apparatus for the artificial intelligence system may include corresponding hardware and software for implementing the hardware functions in order to realize the various functions of the deployment method.

[0155] Those skilled in the art will readily recognize that the steps of the deployment method for an artificial intelligence system described in conjunction with the embodiments of this disclosure can be implemented in hardware or in a software-driven hardware manner. Whether a function is executed in hardware or software-driven hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0156] The beneficial technical effects corresponding to the exemplary embodiments of this device can be found in the corresponding beneficial technical effects in the exemplary method section above, and will not be repeated here.

[0157] Figure 12 This is a structural diagram of an artificial intelligence deployment apparatus provided in an exemplary embodiment of this disclosure.

[0158] See Figure 12 The artificial intelligence deployment device 1200 includes a loading module 1210, a parsing module 1220, and a communication module 1230.

[0159] The artificial intelligence deployment device 1200 can be used to implement the corresponding method embodiments of this disclosure.

[0160] In one possible implementation, the loading module 1210 is used to load the logical document corresponding to the artificial intelligence system in response to the deployment instructions of the artificial intelligence system. The logical document includes task description information written in natural language. The parsing module 1220 is used to parse the task description information to determine at least one executable task of the artificial intelligence system and the execution flow of at least one executable task. The communication module 1230 is used to establish a communication connection with the server of the artificial intelligence system and listen to the interactive information input by the user. The interactive information is used to match the target task in at least one executable task.

[0161] In one possible implementation, the parsing module 1220 is used to parse the metadata in the task description information to determine at least one executable task and at least one functional description information of the executable task; parse the text data in the task description information corresponding to the functional description information of at least one executable task to determine the execution logic of at least one executable task; and determine the execution flow of at least one executable task based on the execution logic of at least one executable task.

[0162] In one possible implementation, the loading module 1210 is further configured to load at least one reference document based on reference instructions in the logical document, the reference document including additional descriptive information of the executable task; the parsing module 1220 is further configured to parse the additional descriptive information, determine the additional execution logic of the executable task; embed the additional execution logic into the execution logic, and update the execution flow of the executable task.

[0163] In one possible implementation, the loading module 1210 is used to determine at least one reference path based on a reference instruction, the reference path indicating the directory where the referenced document is stored; access the storage directory, and load the referenced document in the storage directory.

[0164] In one possible implementation, the AI ​​deployment device 1200 further includes an update module, which, in response to the monitored interaction information being feedback information for any additional descriptive information, determines incremental information of the additional descriptive information based on the feedback information; and updates the referenced document corresponding to the additional descriptive information based on the incremental information to obtain the target referenced document.

[0165] In one possible implementation, before updating the referenced document corresponding to the additional description information to obtain the target referenced document, the update module is also used to generate and save a copy file corresponding to the referenced document; after updating the referenced document corresponding to the additional description information based on the incremental information to obtain the target referenced document, the update module is also used to respond to receiving a rollback instruction corresponding to the referenced document and modify the target referenced document to the referenced document based on the copy file.

[0166] In one possible implementation, the AI ​​deployment device 1200 further includes a building module for constructing an operating environment for the AI ​​system based on a preset security isolation mechanism; wherein the operating environment is used to instruct file access configuration, tool invocation configuration, resource usage configuration, and network access configuration during the deployment of the AI ​​system.

[0167] In one possible implementation, the AI ​​deployment device 1200 further includes an execution module, which, in response to listening for interactive information, uses a large language model deployed on the server side of the AI ​​system to analyze the interactive information to obtain the task intent corresponding to the interactive information; matches the task intent with functional description information corresponding to at least one executable task to determine the target task corresponding to the task intent; and executes the target task according to the target execution flow corresponding to the target task.

[0168] In one possible implementation, the execution module is further configured to read a target interaction file according to a preset time interval. The target interaction file is the latest interaction file generated in a preset folder, and the interaction file is a file generated based on the user's interaction information. The target file identifier of the target interaction file is compared with the historical file identifier, where the historical file identifier is the file identifier of the interaction file generated before the target interaction file. If the target file identifier and the historical file identifier are different, it is determined that interaction information has been detected.

[0169] In one possible implementation, the execution module is further configured to, in response to receiving a trigger instruction, match the trigger instruction with trigger conditions in a logical document to obtain a matching result, the matching result including whether the trigger instruction meets the trigger conditions or not, wherein the trigger instruction meeting the trigger conditions is used to indicate the start of listening for user input interaction information.

[0170] Exemplary electronic devices Figure 13 This is a structural diagram of an electronic device provided in an exemplary embodiment of this disclosure. Figure 13 As shown, the electronic device 1300 includes at least one processor 1310 and a memory 1320.

[0171] The processor 1310 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 1300 to perform desired functions.

[0172] The memory 1320 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 1310 may execute one or more computer program instructions to implement the deployment methods of the artificial intelligence systems of the various embodiments of this disclosure and / or other desired functions.

[0173] In some examples, the electronic device 1300 may also include an input device 1330 and an output device 1340, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0174] The input device 1330 may also include, for example, a keyboard, a mouse, etc.

[0175] The output device 1340 can output various information to the outside, including, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0176] Of course, for the sake of simplicity, Figure 13 Only some of the components of the electronic device 1300 relevant to this disclosure are shown, omitting components such as buses, input / output interfaces, etc. In addition, the electronic device 1300 may include any other suitable components depending on the specific application.

[0177] Exemplary computer program products and computer-readable storage media In addition to the methods and apparatus described above, embodiments of this disclosure may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps in the methods for deploying an artificial intelligence system according to various embodiments of this disclosure as described in the "Exemplary Methods" section of this specification.

[0178] Computer program products can be written in any combination of one or more programming languages ​​to perform the operations of embodiments of this disclosure. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0179] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps in the deployment methods of an artificial intelligence system according to various embodiments of this disclosure as described in the "Exemplary Methods" section above.

[0180] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0181] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0182] The block diagrams of devices, apparatuses, devices, and systems disclosed herein are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0183] It should also be noted that in the apparatus, devices, and methods of this disclosure, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions to this disclosure.

[0184] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0185] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations therein.

Claims

1. A method for deploying an artificial intelligence system, the method comprising: In response to the deployment command of the artificial intelligence system, the logical document corresponding to the artificial intelligence system is loaded, and the logical document includes task description information written in natural language; Based on the task description information, at least one executable task of the artificial intelligence system and the execution flow of the at least one executable task are determined; A communication connection is established with the server of the artificial intelligence system, and the user's input interaction information is listened to. The interaction information is used to match the target task in the at least one executable task.

2. The method according to claim 1, wherein, The step of parsing the task description information to determine at least one executable task of the artificial intelligence system and the execution flow of the at least one executable task includes: Parse the metadata in the task description information to determine the at least one executable task and the functional description information of the at least one executable task; Parse the text data in the task description information that corresponds to the functional description information of the at least one executable task, and determine the execution logic of the at least one executable task; Based on the execution logic of the at least one executable task, the execution flow of the at least one executable task is determined.

3. The method according to claim 2, wherein, The step of determining at least one executable task of the artificial intelligence system and the execution flow of the at least one executable task based on the task description information further includes: Based on the reference instructions in the logical document, at least one reference document is loaded, the reference document including additional descriptive information of the executable task; Parse the additional description information to determine the additional execution logic of the executable task; The additional execution logic is embedded into the execution logic to update the execution flow of the executable task.

4. The method according to claim 3, wherein, The step of loading at least one referenced document based on the reference instructions in the logical document includes: Based on the reference instruction, at least one reference path is determined, the reference path being used to indicate the storage directory of the referenced document; Access the storage directory and load the referenced document from the storage directory.

5. The method according to claim 3, wherein, After establishing a communication connection with the server of the artificial intelligence system and listening to the user's input interaction information, the method further includes: In response to the monitored interaction information being feedback information for any of the additional description information, incremental information of the additional description information is determined based on the feedback information; Based on the incremental information, the referenced document corresponding to the additional description information is updated to obtain the target referenced document.

6. The method according to claim 5, wherein, Before updating the referenced document corresponding to the additional description information based on the incremental information to obtain the target referenced document, the method further includes: Generate and save a copy of the referenced document; After updating the referenced document corresponding to the additional description information based on the incremental information to obtain the target referenced document, the process further includes: In response to receiving a rollback instruction corresponding to the referenced document, the target referenced document is modified to the referenced document based on the copy file.

7. The method according to any one of claims 1 to 6, wherein, Before determining at least one executable task of the artificial intelligence system and the execution flow of the at least one executable task based on the task description information, the method further includes: The operating environment of the artificial intelligence system is constructed based on a preset security isolation mechanism; The operating environment is used to instruct the file access configuration, tool invocation configuration, resource usage configuration, and network access configuration during the deployment of the artificial intelligence system.

8. The method according to any one of claims 1 to 6, wherein, After establishing a communication connection with the server of the artificial intelligence system and listening to the user's input interaction information, the method further includes: In response to the detected interaction information, the system analyzes the interaction information using a large language model deployed on the server side of the artificial intelligence system to obtain the task intent corresponding to the interaction information. The task intent is matched with the functional description information corresponding to the at least one executable task to determine the target task corresponding to the task intent; The target task is executed according to the target execution process corresponding to the target task.

9. The method according to any one of claims 1 to 6, wherein, The monitoring of user input interaction information includes: According to a preset time interval, a target interaction file is read. The target interaction file is the latest interaction file generated in the preset folder. The interaction file is a file generated based on the user's interaction information. The target file identifier of the target interactive file is compared with the historical file identifier, wherein the historical file identifier is the file identifier of the interactive file generated before the target interactive file; If the target file identifier is different from the historical file identifier, then it is determined that the interaction information has been detected.

10. The method according to any one of claims 1 to 6, wherein, Before listening to the user's input interaction information, the method further includes: In response to receiving a trigger instruction, the trigger instruction is matched with the trigger conditions in the logical document to obtain a matching result. The matching result includes whether the trigger instruction satisfies the trigger conditions or whether the trigger instruction does not satisfy the trigger conditions. The trigger instruction satisfying the trigger conditions is used to indicate that the user's input interaction information should be started.

11. A deployment apparatus for an artificial intelligence system, the apparatus comprising: A loading module is used to load a logical document corresponding to the artificial intelligence system in response to the deployment command of the artificial intelligence system. The logical document includes task description information written in natural language. The parsing module is used to parse the task description information and determine at least one executable task of the artificial intelligence system and the execution flow of the at least one executable task; A communication module is used to establish a communication connection with the server of the artificial intelligence system and to listen to the interactive information input by the user, the interactive information being used to match a target task in the at least one executable task.

12. A computer-readable storage medium storing a computer program for executing a deployment method of the artificial intelligence system according to any one of claims 1-10.

13. An electronic device, the electronic device comprising: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the deployment method of the artificial intelligence system according to any one of claims 1-10.

Images