Traditional management software works like this: the vendor thinks of a scenario and builds a page. When a new question arises on-site, you have to submit a feature request and wait for the next version. But real-world questions are long-tailed: "List machines that haven't finished downloading today, allow one-click re-download", "View online rate by classroom", "Which machines are still using the old image". vDisk flips this around: instead of delivering a set of pre-built features, the platform turns every capability into a "skill" that the AI assistant can call — query data, capture screen, read logs, send commands, check switches, run CLI commands. The admin describes the goal in one sentence, the assistant picks the skills and executes them in order, pausing for manual confirmation when needed. The question "Does this feature exist?" becomes "Can we assemble it from existing skills?", and the answer is usually yes.
The following paragraph is the core conclusion of this page; all other sections expand on it
Chengcheng vDisk's AI assistant is the second way to operate the cloud desktop management console.It turns every capability of the platform into callable 「skills」——querying data, capturing client screens, reading logs, issuing commands, checking switch ports, executing command lines on the server. The administrator describes the goal in one sentence, the assistant selects the skills and executes them in order. Any action that changes system state will pause, pop up a confirmation card, and wait for the administrator's approval. Therefore, the product's capability boundary is no longer determined by 「which pages are included in this version」, but by 「whether existing skills can be combined to achieve the result」.
It's not that we haven't added enough features, but that the approach of "one question, one page" is inherently unworkable—the speed of asking questions will always outpace the speed of releasing versions.
The more features, the harder to find in the menu. Only a few are truly used, the rest become a burden; and the table the admin wants to see most at the moment is probably not in the menu.
A small table showing "online rate per classroom" takes weeks or even months to complete. By the time it's done, what the field cares about may have already changed.
Schools care about exam rooms, hospitals care about departments, enterprises care about workstations. Vendors cannot customize a version for each site, and making it a universal feature leaves no one satisfied.
"Features" are paths predetermined by the vendor; "skills" open up capabilities for you to combine on the fly based on your goals.
| Comparison Item | Traditional: Delivery of functions | vDisk: Delivery skills |
|---|---|---|
| Find a feature | Look through the menu; if you can't find it, it doesn't exist | State the goal in one sentence, and the assistant picks the skill to execute. |
| Page Not Found | Submit requirements, schedule, wait for version, upgrade | Generate instantly from a spoken sentence; ready upon refresh |
| troubleshooting | Check each machine and sift through logs one by one | The assistant collects evidence on its own, presents the evidence, and then draws a conclusion |
| capability boundaries | What decisions were made by this version? | Determined by whether skills can be combined, and expands as skills increase. |
| Cost of new capabilities | Modify code, build package, deploy | No new code is generated; deleting a record completely reverts the change |
| Differences between different sites | Making it a general-purpose feature, no one is fully satisfied | Each site creates its own required table |
| Permissions and Risks | Control permissions by role | Permission control remains unchanged, plus high-risk operations must be manually confirmed. |
For example, 'Create a page listing machines that haven't finished downloading today, with one-click re-download' — the assistant generates it on the spot, runs a test, and saves it only if it passes.
No need to know query syntax or internal IDs. Just describe what you want to see, filter by, or which button to click — clearly.
Data retrieval can only use read-only queries; buttons can only reference existing console operations. Using state-changing actions as data sources will be rejected immediately.
When actually fetching data, the process outputs step by step: Validating definition → Trial run → Trial run reaches line 12 → Added to menu. If it fails, it will not be saved.
Summary numbers, filter conditions, tables with customizable columns and formats, bar/line/pie charts, inline action buttons. When there is a lot of data, it will indicate 'Only the first N rows are displayed (M rows total).'
In "Feature Management", change sharing scope, disable, delete, or view definitions. What is shared is the page, not permissions—what a user can see is still re-evaluated based on the viewer's own permissions.
Read-only skill assistants execute automatically; write, delete, execute commands, etc., are not executed by default, only a pending confirmation card is issued.
Group online counts, network-wide download progress, endpoint status, software installation list, SSD health, system logs. Ask in one sentence, see all at once, no need to flip through pages one by one.
For hard-to-describe issues like blue screens, error pop-ups, or freezing on a screen, the assistant can directly capture the current screen of that machine and analyze the image; you can also paste a screenshot and ask a question.
Deploy desktops, set default desktop, change access mode, restart/shutdown/wake up, create computers/groups/scheduled tasks, and even execute commands on the server—high-risk operations always prompt a confirmation card first.
View per-port connection status, negotiated speed, duplex, spanning tree, and error packets/collisions/packet loss. Identify slowness, loops, collisions, and error packets at once, and provide commands in Huawei, H3C, and Cisco formats. Read-only, and can only query registered devices.
When human intervention is needed for repair, the assistant can directly generate a time-limited remote link, without having to search through the console or go to the computer lab.
The WeChat mini-program shares the same skills, same session records, and the same depth of troubleshooting with the management console—depth is determined by the problem, not by "which device the query comes from."
Skill-based expansion expands 'what can be expressed', not 'what can be bypassed' — the original permission system remains unchanged, with an additional manual confirmation layer on top.
For operations like writing files, deleting files, executing commands, batch deployment, the assistant only shows a card specifying which machine, which desktop, and what parameters to operate, and the administrator clicks to confirm.
Directly executed after confirmation, no longer going through the large model. What is applied is the snapshot at the moment the card was popped, eliminating the situation where 'what was confirmed is A but what is executed is B'. Approval is valid for 15 minutes, can only be used once, and cannot be delegated to others.
The server command line is not a universal gateway: for commands that would damage the operating system or the desktop cloud software itself, the assistant refuses directly without asking for approval.
"My Functions" can only retrieve data through read-only queries; the page does not execute any scripts; users can only fill in "value", which operation to call and which parameters to pass are locked by definition, and changing the address bar cannot change to other operations.
What data can be viewed is recalculated each time based on the current user's permissions. Sharing a feature with a colleague does not lend your own permissions to them.
What is created is a "function definition" rather than a piece of code, rendered on the fly by a generic renderer. Deleting that record is equivalent to a complete rollback, leaving no residue in the system.
It's not 'added a chat box', but compressing a person's repetitive actions of a whole day into one sentence and one click.
| Scenarios | How it was done before | What to Do Now |
|---|---|---|
| Pre-semester inventory | Open classrooms one by one, manually count online devices and check image versions. | Create a page for 'Online Rate and Image Version by Classroom'; thereafter, simply open it daily. |
| Image download incomplete | Check download progress per PC, pick out those not finished and re-download per PC | List unfinished machines in one sentence, directly re-download with one click on the page. |
| A certain machine is very slow | Suspect the network cable, suspect the switch, suspect the hard drive, test one by one. | The assistant simultaneously checks NIC negotiation speed, switch port errors, and download link, presenting the evidence. |
| Student computer blue screen | Teacher describes the symptom, administrator guesses which step is wrong | Paste a screenshot, or have the assistant directly capture that machine's screen to view |
| Exam Check-in | Go to each machine to check if it's in write mode, if the USB drive is unplugged | Batch deploy exam environment detection; red bars on screen are clearly visible; after the exam starts, batch remove them. |
| Temporary statistics | Submit requirements to vendor, wait for next version | Speak a sentence, generate a form on the spot |
Which model to use, single-response limit, and billing method are all configured in the cloud; on-site there are only two switches.
No need to enter the API endpoint, model name, and API key on every server. Newly installed servers are enabled out of the box; for scenarios where the entire machine needs to be disabled, simply turn it off in the settings page.
Models and execution profiles are maintained in the cloud; each site automatically syncs within 5 minutes at most, without needing to log into any on-site server.
Balance is always displayed on the chat page, can be self-recharged (WeChat Pay / Alipay), and self-invoicing is available when accumulated amount reaches a threshold. Creating a "My Feature" consumes one Q&A point, but after creation, opening it daily costs no points.
Apply for a free trial: Let the assistant check your site's download and online status, capture a machine's screen, and then create a table you've always wanted on the spot—whether it works, you'll know in one go.