Shanghai Chengcheng Information Technology Co., Ltd. · The team has been established for 20 years, dedicated to R&D.
AI Assistant · Skill-based architecture

From providing features to providing skills: say one sentence, and the console gains one more feature

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.

Explain clearly in one sentence

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」.

Product name
ChengCheng vDisk Converged Cloud Management Platform · AI Assistant
Belongs to
The agent built into the cloud desktop / IDV centralized management platform is not a separately sold chat tool.
architecture change
From "vendor pre-built function pages" to "platform open skills, temporarily combined by targets"
Most Typical Capabilities
"My Features": The admin says a sentence, the console instantly generates and tests a custom page, and after approval, it is attached to the left menu.
What is generated?
A functional definition (what to query, how to display, which existing operations can be invoked), rendered by a generic renderer,Generates no new code, deleting the record fully revokes
Security Boundary
Data retrieval can only use read-only queries; buttons can only reference existing console operations; data visibility range is determined in real time based on the current user's permissions; high-risk actions must be manually confirmed by an administrator before execution.
Access
The top-level 'AI Assistant' menu in the management console, the left-side 'My Functions', and the WeChat mini-program (all three share the same set of skills and the same conversation records).
On-site deployment requirements
On-site servers do not need to configure model interfaces; access and models are uniformly distributed from the cloud, out-of-the-box for new installations.
Use Cases
Places that require centralized management of a large number of endpoints, such as school computer classrooms, college computer labs, examination halls, enterprise training rooms, training labs, call centers, hospital workstations, etc.

The ceiling of 'feature-based' software is not about the number of features

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 menu keeps getting longer, but the one you need is still missing

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.

Submit requirements → Schedule → Release version → Upgrade

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.

Each site has different ways of asking

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.

Skills, not features: the same platform, two delivery methods.

"Features" are paths predetermined by the vendor; "skills" open up capabilities for you to combine on the fly based on your goals.

Table 1: Differences between traditional 'delivery of features' and vDisk 'delivery of skills' across seven dimensions
Comparison ItemTraditional: Delivery of functionsvDisk: Delivery skills
Find a featureLook through the menu; if you can't find it, it doesn't existState the goal in one sentence, and the assistant picks the skill to execute.
Page Not FoundSubmit requirements, schedule, wait for version, upgradeGenerate instantly from a spoken sentence; ready upon refresh
troubleshootingCheck each machine and sift through logs one by oneThe assistant collects evidence on its own, presents the evidence, and then draws a conclusion
capability boundariesWhat decisions were made by this version?Determined by whether skills can be combined, and expands as skills increase.
Cost of new capabilitiesModify code, build package, deployNo new code is generated; deleting a record completely reverts the change
Differences between different sitesMaking it a general-purpose feature, no one is fully satisfiedEach site creates its own required table
Permissions and RisksControl permissions by rolePermission control remains unchanged, plus high-risk operations must be manually confirmed.

"My Functions": A single sentence adds a link on the left side of the console.

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.

STEP 01

Describe requirements in plain language

No need to know query syntax or internal IDs. Just describe what you want to see, filter by, or which button to click — clearly.

STEP 02

on-site verification

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.

STEP 03

on-site trial run

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.

STEP 04

Refresh and Use

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).'

STEP 05

Share with colleagues

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.

What skills does the assistant have?

Read-only skill assistants execute automatically; write, delete, execute commands, etc., are not executed by default, only a pending confirmation card is issued.

See the status of the entire network

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.

View client screens

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.

Fix manually

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.

Check switch port

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.

Open remote assistance

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 same brain on the phone

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."

Before delegating control to a single command, every safeguard remains in place

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.

High-risk not executed by default

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.

Execute what is confirmed

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.

Directly refuse actions that would destroy the system

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.

Self-created pages are read-only

"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.

Re-evaluate permissions each time

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.

Can be fully revoked at any time

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.

What is it really used for on-site?

It's not 'added a chat box', but compressing a person's repetitive actions of a whole day into one sentence and one click.

Table 2: Comparison of practices after switching to AI assistant for six typical daily scenarios in the computer lab.
ScenariosHow it was done beforeWhat to Do Now
Pre-semester inventoryOpen 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 incompleteCheck download progress per PC, pick out those not finished and re-download per PCList unfinished machines in one sentence, directly re-download with one click on the page.
A certain machine is very slowSuspect 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 screenTeacher describes the symptom, administrator guesses which step is wrongPaste a screenshot, or have the assistant directly capture that machine's screen to view
Exam Check-inGo to each machine to check if it's in write mode, if the USB drive is unpluggedBatch deploy exam environment detection; red bars on screen are clearly visible; after the exam starts, batch remove them.
Temporary statisticsSubmit requirements to vendor, wait for next versionSpeak a sentence, generate a form on the spot

On-site servers do not need to configure any AI parameters

Which model to use, single-response limit, and billing method are all configured in the cloud; on-site there are only two switches.

out-of-the-box

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.

Set once, Follow across the entire network

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.

Billed by the number of points, pay for what you use

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.

First, use your existing computer lab and ask it three questions.

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.

Page update date: 2026-06-30