At Computex 2026 and the QNAP World Tour 2026, QNAP showcased a demo where an AI agent operated the QVR Client. The AI agent receives tasks via natural language instructions, switches cameras, searches recordings, exports clips, and reports results.
Current status: Release Testing. Supported AI agents are ChatGPT Codex, Claude Code, and Antigravity CLI. Beta version is available. The application method is detailed at the end of this article.
Supported Tasks
| Input | Agent Action |
|---|---|
| “Show the Lobby camera in fullscreen.” | Identifies the camera by name and displays it. |
| “When did Alex appear yesterday afternoon?” | Runs face search in the specified time period. |
| “Was license plate ABC-1234 detected this week?” | Performs license plate search and summarizes results. |
| “Export the last 5 minutes from the Door camera.” | Exports the clip and tracks progress until the file is saved. |
| “Create rule: Start recording when motion is detected on the Walkway camera.” | Creates an event rule and confirms it’s saved. |
The current build offers 92 automation methods covering: live view, playback, search, export, bookmark, event rules, schedule, PTZ, screenshot. Multi-step tasks are also supported. Example: “Find all of Alex’s face search results from this morning, export a 30-second clip before and after each, and summarize them in a table.”
How It Works
- Detection. The AI agent queries the registry service QVR Agent, installed with the client, to detect the running client.
- Getting Specifications. The AI agent gets session rules, a list of methods, and parameters for each method from the client.
- Execution and Verification. The AI agent sends JSON-RPC commands via WebSocket. After each operation, it checks the client’s status.
The client provides its own interface specifications. Any AI agent capable of WebSocket connection can be used. Methods added in future builds will be detected without updating the skill.
Setup
- In the QVR Client, open Settings > AI Agent Control, and enable Accept AI Agent commands. The client launches a local-only endpoint. Disabling this option stops the service immediately.
- Copy the prompt shown in Skill Installation and paste it in the chat field of your AI agent (e.g., Claude Code). The AI agent installs the
qvr_ctrlskill bundled with QVR Client.
That’s it for setup. Example first input: “Can you operate the QVR Client on my PC?”
Security and Privacy
- Activation. The service only operates while Accept AI Agent commands is enabled.
- Connection Scope. The endpoint listens only at 127.0.0.1. It’s not accessible from the network.
- Credentials. Passwords, PIN codes, and camera credentials are not exposed to the AI agent.
- Permissions. The AI agent operates within the permissions of the currently logged-in account.
- Method Classification. Each method is classified as read-only or writable. You can restrict the AI agent to read-only methods, or require approval for any changes.
Warning: Data read by the AI agent (camera list, server status, screenshots, recordings, etc.) is sent to the cloud of each AI provider according to their privacy policy. Usage with cameras recording highly confidential scenes is not recommended. Results may vary depending on the AI agent and model used.
Requirements and Limitations
Requirements: Beta version of QVR Client for Windows or macOS, and an AI agent running on the same PC.
Beta limitations:
- Remote control is not supported.
- Some methods are not available yet.
- Login must be performed by the user.
Getting the Beta
If you’d like the installation files, please leave a comment on this article or send me a private message. I’ll provide the installation package.