Competition in the AI era is not about whose model has more parameters or whose neural network is deeper — it is about who canPut AI to work faster。 信管系、信息系、经管学院……Not每个专业都需要成为算法工程师，但每一个专业的学生，都需要具备The ability to work effectively in the AI era。 vDisk AI teaching platform，让Universities各院系在No need to purchase a GPU server or build an algorithm team, truly bringing AI into the classroom.
When advancing AI-based teaching, many universities often find themselves torn between these two approaches.
Procure GPU servers, deploy a local large model, assemble an algorithm-tuning team, and first get the whole school's faculty and students to understand neural networks and the Transformer architecture...
Starting from the angle of "knowing how to use AI tools," we bring the best commercial large-model capabilities straight into the classroom, so students feel AI's real boost to learning and work efficiency from the very first lesson.
Driving a car doesn't mean you have to learn to repair the engine; likewise, making good use of AI tools doesn't mean everyone has to master neural networks.Bringing AI genuinely into daily learning and workflows is this era's most central demand on higher education.What we set out to solve is how to make this actually happen.
When universities push AI-based teaching, it's often not for lack of will but because they're stuck on these real-world problems.
The GPU servers required to self-host a large model are staggeringly expensive, and models iterate extremely fast — hardware bought today may be unable to run next year's latest version, creating an ongoing hardware burden.
Connect to leading cloud large models via API, with inference performed in the cloud; on campus, only a single ordinary server is needed to handle access and management,Near-zero hardware investment。
A self-built solution involves GPU drivers, model-weight downloads, inference-framework configuration, building a permissions system, and more — every step requires specialist involvement, and the path from project kickoff to go-live often takes several months.
Drawing on the vDisk image marketplace, select an image with AI teaching components already integrated and batch-deliver it to lab terminals in a single click,From deployment to first class in as little as a few days。
Without proper controls, a token-billed AI API can cause costs to skyrocket when large numbers of students use it simultaneously, making it hard for school administrators to approve projects with uncertain budgets.
Platform SupportPer-user Token limit settingsWith per-account statistics, cost ceilings locked in advance, fully transparent operating costs and procurement approvals backed by clear records.
Having students re-enter large blocks of background information every class, with conversation history that cannot carry over between sessions, wastes Tokens and disrupts the teaching pace; AI-generated content is also hard to save and manage consistently.
vDisk network disk mounts as a virtual drive, writing desktops, documents and AI output to the server in real time,Resume right from last time's progress in the next class, with no need to re-enter context.
A complete, truly deployable AI teaching infrastructure
From the image marketplace, choose an image template pre-installed with the AI teaching client and runtime environment, and deliver it in bulk to all terminals in the lab. No per-machine installation and no manual configuration are needed; versions are unified and environments are consistent, dramatically shortening the rollout cycle of the AI teaching environment.
Platform External Offerings统一的 OpenAI API compatible代理端点。VS Code、Cursor、各类 AI 编程插件、自研教学系统，只需在设置中填写统一的网关地址与密钥，即可接入 DeepSeek、文心一言、通义千问、智谱 GLM、豆包、讯飞星火、腾讯混元、Kimi 等国产主流大模型，无需为每家模型单独对接。
Mounted as a virtual disk (e.g., D:), directories such as Desktop, My Documents, and AppData are automatically redirected to the server and saved in real time. AI conversation logs, code drafts, and lab reports are all saved to disk, so the next class resumes from the previous breakpoint, significantly reducing repeated Token input and making it convenient for teachers to review the learning process.
It supports setting a Token usage cap per account, tracks each student's consumption in real time, and automatically blocks over-quota requests. The school's budget planning is fully documented, with no worry that a few students' overuse will drive up overall costs, and the total operating cost is completely transparent.
Large models evolve at a breakneck pace. The platform backend lets you switch the connected domestic model at any time, with no changes needed to the front-end teaching interface, so students always use the industry's current best model and avoid being locked into outdated versions by hardware bindings, keeping both teaching content and the tooling layer at the technological forefront.
By connecting AI vision to existing surveillance cameras, automatic classroom inspection, anomaly alerts and timetable linkage are achieved. During timetable periods, power, access control and air conditioning are turned on automatically and shut off after class, with no need for staff on duty, freeing lab administrators from repetitive operations.
AI in teaching is not exclusive to the computer science department—every discipline has its own AI use cases
By configuring a unified OpenAI-compatible endpoint in VS Code or Cursor, students can use AI-assisted completion, code explanation, and bug location during programming lab sessions. AI output is automatically saved to disk, so the next session resumes the previous project's progress, cutting the time spent rebuilding the engineering environment.
For students in majors such as information management, business administration, and marketing, the focus is not on training models but on learning how to leverage AI tools to efficiently complete business-scenario tasks such as market analysis, proposal writing, and data processing. The platform has a built-in conversation interface that is ready to use right after startup, offering zero-barrier access to the most powerful commercial LLMs available today.
In laboratory courses such as chemistry, physics and biology, students can use AI to quickly look up experimental principles, generate experiment report frameworks and assist with data analysis. The data generated during experiments and the AI interaction records are synchronously saved to the network drive, and teachers can track the completion progress of experiment reports in real time.
AI general-education electives are offered to all majors across the school; students need no prior AI background and can experience LLM conversations, an introduction to Prompt engineering, and hands-on AI toolchain practice simply by logging into the cloud desktop. A unified account system eliminates the need to register, top up, and configure for each person individually, greatly reducing the teaching organization cost of general-education courses.
Our goal is to get AI teaching capabilities truly up and running in the shortest possible time.
Confirm with our consultants the number of lab servers, network bandwidth, the cloud desktop version in use, and the needs of the target departments, then produce an onboarding plan and cost estimate tailored to your school's actual situation.
In the vDisk image marketplace, select or customize an image with AI teaching components already integrated; in the management console, configure an OpenAI-compatible gateway, connect to the target large model, and set Token usage policies and per-account permissions.
Use the vDisk centralized management platform to batch-deploy images to lab terminals, while applying network-disk mount policies and user-directory redirection so that AI outputs and learning records are automatically persisted to storage.
Provide brief operational training for the lead course instructor, focusing on viewing Token usage, switching models, and reviewing student data; run a first trial class to verify that the teaching workflow runs smoothly before officially launching the course.
This is the core question many schools ask when making decisions
For the vast majority of universities, the core rationale for building their own large model simply does not hold up. The issue isn't whether the technology is hard — it's that,Building your own model means falling behind from day one。
Commercial large models iterate on a monthly cadence: hardware bought for 2 million today and the model deployed on it will very likely be surpassed by a more powerful next-generation version within six months. Yet school equipment cannot automatically update as models advance, which means students end up using a set of AI capabilities "frozen" at a single point in time — disconnected from the industry, from the market, and from the tools they will actually use after graduation.
The logic of the vDisk platform is to keep schools always connected to the best commercial large models the industry currently offers. Models are switched on the backend with no impact felt on the front end; course knowledge points and learning objectives are decided by teachers, butThe tooling layer stays continuously in sync with mainstream practices. This is not a compromise on technical capability, but the most pragmatic path to achieving teaching objectives.
The real competitive edge of AI teaching lies not in how large a model you deploy, but in whether your students have learned to use AI effectively in real-world work scenarios.
| Dimension | Self-hosted AI server | vDisk API integration |
|---|---|---|
| Initial Hardware Cost | From RMB 2–3 million | Near Zero Cost |
| Deployment Cycle | Several Months | Several Days |
| Model Update Method | Repurchase Hardware | One-Click Switching in the Console |
| Model Capability Level | Fixed version, gradually falling behind | Always connected to the latest commercial models |
| O&M Requirements | Requires dedicated AI engineers | Existing data-center O&M is enough |
| Cost Predictability | A large one-time purchase plus ongoing maintenance | Billed by token, with a controllable cap |
| Student User Experience | Requires self-configuration, with a high barrier | Ready to Use on Boot, Zero Configuration |
This is neither the students' fault nor the teachers' fault — it's a disconnect that has opened up between the tools and the times.
当课堂里用的是 VC 6.0 和 VB6，而学生手机里看到的是 Cursor 一键生成完整项目、 Claude Code Pass对话重构整个代码库——这种认知落差，Not靠"Foster Proper Study Attitudes"能弥合的。 真正有效的做法，是Bring real tools that are actually in use in 2026 into the classroom， 让学生从第一堂课就感受到：这东西出去真的能用。
也不需要一上来就讲 Transformer 原理、反向传播、神经元激活函数—— 那是研究生和算法工程师的课题。对大多数学生而言， Learning to use it, getting results, and building positive feedback is where everything begins. The underlying principles can be explored naturally once curiosity has been sparked.
These three tools are the AI coding assistants most frequently used by engineers and students in 2026 — deploy them into the computer lab so students start using real-world tools right from the classroom.
Deeply reworked from VS Code, it has built-in AI chat, code completion, whole-file editing and Composer multi-file coordination. Students don't need to switch tools — they can interact with the AI in real time while writing code, with an extremely low barrier to entry.
A command-line AI programming tool from Anthropic that can directly operate on project directories, read and write files, execute commands, and commit code. Ideal for senior computer-science students to experience a real AI-collaborative development workflow.
Delivered as a VS Code plugin with inline code completion, function generation, and comment-to-code conversion. Highly compatible and supporting nearly all programming languages, it offers the smoothest entry point to AI for beginners.
In the image marketplace of the vDisk management console, select a teaching image with Cursor preinstalled, or install Cursor on a base image and create a snapshot, then batch-deploy it to all terminals in the room to ensure version consistency.
Open Cursor → Settings → Models → OpenAI API Key, enter the Key issued centrally by the school; in Override OpenAI Base URL enter the vDisk OpenAI-compatible gateway address here. Once done, all of Cursor's AI requests are routed through the school gateway to a domestic large model.
Guide students to set Cursor's Workspace under a vDisk network drive letter (such as D:\projects\). The vDisk network disk isA personal disk mounted separately for each student after login, the D: drive itself belongs solely to the currently logged-in student, so there is no need to create subdirectories by student ID to distinguish them. All code files, conversation history (.cursor directory) is written to the server in real time, so the next class simply opens it and the previous lesson's progress and AI conversation context carry over intact.
OpenAI API Key sk-school-xxxxxxxxxxxxxxxx Override Base URL https://ai.school.edu.cn/v1 Model deepseek-chat # 或 gpt-4o / qwen-max # 学生无需注册任何大模型账号 # 统一由School网关鉴权与计量 # Token 超额后自动提示，不产生额外Cost
# D: 是每位学生登录后独立挂载的Personal Exclusive盘 # 无需学号子目录，D: 本身就只属于当前学生 D:\projects\ # Create the project directly under the personal disk ├─ .cursor\ # AI Chat History ← Auto-saved │ └─ chat-history\ ├─ src\ # Code files ← Written to the server in real time └─ README.md
After pre-installing Node.js (LTS version recommended) in the lab image, run npm install -g @anthropic-ai/claude-code. We recommend including this step in the image-creation process to avoid taking up class time on student operations.
Since the vDisk gateway provides an OpenAI-compatible protocol, a proxy pointer must be set in the system environment variables (or login script) so that Claude Code requests are forwarded through the school's unified gateway. This can be pushed in bulk via GPO or login script at user logon, taking effect automatically once the student logs in, with no manual configuration required.
Guide Students In D:\projects\my-project directory, then launch Claude Code (claude command). The vDisk network disk gives each studentA personal disk mounted independently after login, D: drives are naturally isolated and mutually invisible, with no need for student-ID subdirectories. Claude Code reads all files in the current directory as context; conversation records and file changes are written to disk in real time, so the project state remains exactly consistent after logging in from another machine.
# Deploy in bulk via GPO or vDisk policy ANTHROPIC_API_KEY = sk-school-xxxxxxxxxxxxxxxx ANTHROPIC_BASE_URL = https://ai.school.edu.cn/v1 CLAUDE_CODE_MAX_OUTPUT = 8192 # 学生Open终端直接运行： # cd D:\projects\my-project # claude # 无需任何额外Configuration即可开始 AI 编程对话
Claude Code can directly read an entire project directory, understand the code structure and then make changes through conversation. It is ideal for senior students to experience the complete workflow of "letting AI take over a real project" — currently one of the most authentic ways enterprise engineers collaborate with AI.
Pre-install VS Code and the GitHub Copilot extension in the image (GitHub.copilot). If you wish to use a domestic model instead, you can install a third-party Copilot-alternative plugin that supports the OpenAI-compatible protocol (such as Continue, Codeium, etc.) and configure it to point to the school gateway.
After installing the Continue extension in VS Code, open its configuration file ~/.continue/config.json, and enter the school's gateway address and Key. Students simply open VS Code to chat with the AI directly in the editor's sidebar or trigger inline code completion.
Set the VS Code Workspace folder directly to D:\projects\ is enough. The vDisk network disk gives each studentPersonal ExclusiveThe mounted drive is attached on login and detached on logout; D: itself belongs only to the current student, so there is no need to create subdirectories by student ID. The Continue plugin's conversation index and history are stored under the project directory and written to disk in real time, seamlessly continuing across devices and across class sessions.
{
"models": [
{
"title": "School AI Gateway",
"provider": "openai",
"model": "deepseek-chat",
"apiBase": "https://ai.school.edu.cn/v1",
"apiKey": "sk-school-xxxxxxxxxxxxxxxx"
}
],
"tabAutocompleteModel": {
"title": "Auto-complete",
"provider": "openai",
"model": "deepseek-coder",
"apiBase": "https://ai.school.edu.cn/v1",
"apiKey": "sk-school-xxxxxxxxxxxxxxxx"
}
}
AI 编程工具最大的价值之一，是上下文的延续——AI 记得你上次在做什么，不用每次重新解释。 而这一切，Dependency于Conversation history files are persisted to unified storage, rather than existing only in a local temporary directory.
# D: 是每位学生登录后独立挂载的Personal Exclusive盘 # 登录即挂载，注销即卸载，天然隔离互不可见 D:\ ← Student personal vDisk network disk ├─ .cursor\ ← Cursor conversation history │ ├─ chat\ │ └─ composer\ ├─ .continue\ ← VS Code Continue history ├─ claude-sessions\ ← Claude Code session summary └─ projects\ ├─ week01\ ← Week 1 course project ├─ week02\ └─ final\ ← Final Project
Core Shift:The goal of an AI programming course is not to "teach students every feature of some tool" but to cultivate in studentsThe ability to complete real tasks in collaboration with AI— including how to describe a problem, how to evaluate AI's output, and how to iterate continuously with AI's help. This is exactly the skill every job in 2026 demands.
A student-side tool integrated into the vDisk teaching space — from doing assignments to submitting them, AI is involved throughout and data is persisted to disk the whole way.
After students finish a stage of work in Cursor, VS Code or any AI tool, they can complete it directly within the teaching space without leaving their workspace.Screenshot Review与File SubmissionTwo actions - AI gives instant feedback and the teacher simultaneously receives the submission record.
Students take a screenshot of their current workspace (code, runtime results, or design drafts), and the screenshot is sent to the large model together with the assignment requirements. Based on preset evaluation dimensions (correctness, coding standards, logical clarity, completeness, etc.), the model generatesStructured scoring and written feedback, students see it instantly while the teacher side records it in sync.
Beyond screenshot-based assessment, students can also submit actual project files (source code, report documents, data files, archives, etc.). Files are selected directly from the assignment directory on the vDisk network drive with no extra upload required, and on the teacher side each student's submission status and file content can be viewed in a unified assignment panel.
All workspace data — screenshots, AI evaluation results, and submitted files — is stored in standardized directories on the vDisk network disk, written to disk in real time, so teachers and students can review historical records at any time.
# D: is each student's personal dedicated drive, naturally isolated, with no need for student-ID subdirectories D:\assignments\ ← Create the assignment folder directly under the personal disk ├─ week03_作业1\ │ ├─ screenshot.png ← Original Screenshot │ ├─ ai_eval.json ← AI scoring results (structured) │ ├─ ai_feedback.md ← AI text feedback │ └─ main.py ← Submitted Source Code ├─ week05_作业2\ │ ├─ screenshot.png │ ├─ ai_eval.json │ └─ project.zip ← Project Archive └─ final_project\ ├─ screenshot.png ├─ ai_eval.json ├─ report.pdf └─ src\ ← Full Project Source Code
The moment a screenshot is taken, an AI evaluation is generated or a file is selected, the data is written to the network disk server—no reliance on local cache and no loss on shutdown.
Screenshots and feedback records for every assignment throughout the semester are fully retained, and a full-process report on the student's AI-assisted learning can be generated at term's end as a basis for formative assessment.
Each student can view only their own assignment directory; the teacher side has read-only access to all students' directories in the class — non-interfering and data-secure.
Teachers can export all students' assignment files and AI evaluation results for the whole class with one click, organized into a standard format, with support for batch grading and score aggregation.
教师在布置作业时Configuration评价Dimension（如"代码是否能运行"、"是否满足作业要求"、"逻辑层次是否清晰"等）及权重，形成结构化评分 Prompt 模板，后续所有提交自动套用此模板。
学生在教学空间点击"Screenshot Review"，系统自动捕获当前屏幕画面，将截图与教师Configuration的评价模板一同发送至大模型（Support视觉理解能力的模型，如 DeepSeek-VL、GPT-4o 等）。
The large model analyzes the screenshot content, scores it item by item across dimensions, and generates written improvement suggestions, returned in structured JSON format (including per-dimension scores, total score, and a feedback summary). Students immediately see the visualized scoring results on the interface.
AI assessment results are automatically written to the corresponding assignment directory on the student's network drive (ai_eval.json + ai_feedback.md), while the teacher side is notified that the student has completed the submission. Teachers can manually review and adjust the AI scores to serve as the basis for the final grade.
Quota allocation, usage monitoring, and model scheduling under a unified account system — keeping the use of AI resources fully under control
学生PassSchool统一网关调用大模型，每一次请求都会消耗 Token。 如果缺乏管控，少数高频Usage的学生可能挤占大量资源，而大多数学生的Usage体验反而受影响； 管理员也无法了解 AI 资源的实际Usage效率，更难以在下一学期制定合理的配额计划。
Token Usage Management的目标，Not限制学生Usage AI， 而是确保Every student gets a stable, fair AI usage quota， 同时让教师和管理员对整体资源消耗保持清晰的感知与掌控。
Configure an independent Token quota for each student account, set by session, week, or semester. When the quota is used up, requests are automatically paused, the student receives a notice, and the teacher can top it up with one click; this prevents a few high-frequency users from crowding out overall resources.
Teachers and administrators can view, in the back end, overall class consumption trends, per-student usage rankings, and a heatmap of Token distribution for each lesson. Data updates in real time, and usage anomalies (extremely high or low) are automatically flagged for timely intervention.
Set an overall balance threshold for the class; when the available quota falls below the warning line, notifications are pushed automatically to the responsible instructor and administrator. Sense resource trends in advance and top up calmly, so a depleted quota never disrupts class continuity.
Different course scenarios can be configured with different models: everyday code completion uses a lightweight, fast model, while in-depth analysis and architecture design use the flagship model. The same student account is automatically routed across scenarios, balancing response speed and usage efficiency.
It supports exporting detailed Token usage by class, by course and by time period, in a format compatible with common spreadsheet tools. The reports can be used for end-of-term resource reviews, next-term quota planning and usage summaries submitted to the school's IT department.
Different classes and courses within the same school can have independent Token pools that do not affect one another. If Class A exceeds its usage, it will not affect Class B's normal use; teachers can also allocate a more generous quota to key courses, flexibly accommodating different teaching intensities.
The greatest bonus of AI-powered teaching is that every student interaction with AI leaves behind quantifiable learning data
When students interact with AI, the quality of their prompts directly reflects theirClarity of thinking and degree of knowledge mastery。 一个描述模糊、只会说"Write a program for me"的学生，与一个能精确描述需求、指定边界条件的学生， 处于完全不同的学习Stage——而这个差异，在传统Exam中几乎无法被量化捕捉。
Token usage is alsoProxy metrics for learning engagement： 主动探索的学生会持续追问、修改、延伸； 被动应付的学生往往只取第一个答案、不做进一步交互。 两种学习行为，在Data上清晰可见。
"帮我写代码"
"这个怎么做"
"报错了怎么办"
"用 Python 实现一个读取 CSV 并按第3列排序的脚本，需要处理空值"
"这段代码在并发访问时会有线程安全问题吗？"
Only 2–3 messages sent in a single class
Closes the window after a single conversation
Multi-turn dialogue with continuous follow-up on details
Explore the same question repeatedly from different angles
"ImportError"、"undefined"、"null pointer"
"这个语法怎么写"、"为什么不能运行"
Many students start proactively asking about architecture design
Independent exploration beyond the course scope
Generated once and submitted directly, never revised
Each question is entirely different, with no continuity
Code files with multiple revision records
Progressively refine the same project, continually adding features
Learning analytics is based on anonymized usage statistics and keyword clustering; the core principle is:To assist teaching, not to surveil students. It is recommended to explain the scope of data collection to students before the course begins, and to use the analysis results to improve instructional design rather than as a direct basis for assessment grading.
Whether you are a department head, a course instructor or a school IT administrator, contact us to obtain an implementation plan and cost estimate tailored to your institution's actual situation.