Qwen 4 built a PC-to-phone file-sharing app in 30 minutes
The author of the Telegram channel Neurogen fed the QoDuck model (Qwen 4) a detailed spec for DropWave, a utility that moves files and text between a Windows PC and a phone over Wi-Fi, with no internet connection. About 30 minutes passed between the first command and green tests. The result was a modular project: a FastAPI server, a mobile web page, and a PySide6 window with a QR code and tray support. The code is on GitHub, so you can check the result yourself.
AI-processed from TG: @neurogen_news (Neurogen); edited by Hamidun News
A familiar situation: you need to move a file from your phone to your computer, but you have no cable, the messenger compresses the quality, and setting up a cloud service for a single PDF feels like too much effort. Utilities like DropWave exist for exactly this. What is interesting is something else: it was generated entirely by the QoDuck model, which the author of the Neurogen channel calls Qwen 4.
This is not proof that the new model has become better than all the others. But it is a concrete example of what a "finished product in one pass" looks like, and of what is worth checking in it.
What was asked for and what came out
The task: an app for instant transfer of files and text between a Windows PC and a smartphone over a local network, without internet. The stack was set in advance: Python 3.10+, PySide6 for the window, FastAPI with uvicorn for the server, websockets, aiofiles, qrcode and pyperclip.
The user scenario is simple. The program starts on the PC and shows a QR code with a link of the form http://<local IP>:<port>. The phone scans the code and opens the page in a browser. Nothing needs to be installed on the phone. Through the page, you can upload files to the PC, send text, and download files that sit on the computer.
According to the author, the model built a modular project: separate modules for networking, the server and storage, a web front end for the phone, and a desktop GUI with a QR code and minimize-to-tray. Files are received in 1 MB chunks, so a large file does not fill up memory all at once. Existing filenames are not overwritten. A WebSocket carries progress with speed and time remaining, pings, and clipboard synchronization.
Where the engineering hides in the spec
Most of the value of this prompt lies not in the stack but in the details people usually forget.
The first detail is detecting the real IP. If the machine runs Docker, WSL or VMware, it has several virtual adapters, and "the first address that turns up" is often the wrong one. The spec prescribes a trick: open a UDP socket to an external DNS address. No data is actually sent, but the system picks the outgoing interface itself, and that interface reveals the real LAN address.
The second detail is echo protection. If auto-paste of text into the Windows clipboard is on and the phone also syncs its clipboard, messages can bounce back and forth endlessly. The author explicitly required protection against a recursive loop and reports that it is implemented.
The third is the launch requirements: a single python main.py command from the root, and validation of port and access-rights errors. This turns "code that looks plausible" into a verifiable acceptance criterion.
How it was checked and what stays off-screen
The author cites a smoke test with 28 checks and a live run of the page in a browser. One practical touch: Tailwind is loaded via CDN, but the styles are duplicated in custom CSS. If there is no internet, which is exactly the scenario the utility was built for, the interface will not fall apart.
There are plenty of limitations too:
- This is a single post by a single author. There are no independent checks.
- It is unknown how the utility behaves with files of tens of gigabytes, on unstable Wi-Fi, or under strict Windows firewall settings.
- The post does not say who wrote the tests. If the model generated them too, "green" results show that the code is consistent with itself, not necessarily that it is reliable.
- A local server without authorization essentially opens the PC to everyone on the network. On public Wi-Fi this is a risk, and the post says nothing about it.
What to take from this
For developers, the main takeaway is not about Qwen but about the format of the task. The model produces a noticeably more coherent result when it is given the stack, file structure, endpoints, edge cases, and a criterion of "runs with one command." A vague "make a file-transfer app" will not have that effect.
Product managers and founders should note something different: the time to a working prototype of an internal utility has shrunk to half an hour. For tasks like this, it is worth first asking whether it is easier to generate the tool you need than to look for a ready-made one.
Before real use, the repository needs a manual read-through: check access rights, handling of paths in filenames, and the absence of authorization. The code is on GitHub (rzaev77/DropWave), and it is a good opportunity to do such a review on a live example.
- Nov 12, 2026Eight Countries, OpenAI and Anthropic Promise an AI "Kill Switch" — but Set No Deadlines or Penalties
- Nov 12, 2026Anthropic Shuts Claude to Russian-Controlled Companies from November 12
- Nov 12, 2026n8n Releases Nod, an Open-Source Agent That Runs on Your Computer
- Nov 12, 2026Hugging Face warns: a wave of hacks could make open-source AI less open
Key Questions Answered
What was asked for and what came out?
The task: an app for instant transfer of files and text between a Windows PC and a smartphone over a local network, without internet. The stack was set in advance: Python 3.10+, PySide6 for the window, FastAPI with uvicorn for the server, websockets, aiofiles, qrcode and pyperclip. The user scenario is simple. The program starts on the PC and shows a QR code with a link of the form http://<local IP>:<port>. The phone scans the code and opens the page in a browser. Nothing needs to be installed on the phone.
Where the engineering hides in the spec?
Most of the value of this prompt lies not in the stack but in the details people usually forget. The first detail is detecting the real IP. If the machine runs Docker, WSL or VMware, it has several virtual adapters, and "the first address that turns up" is often the wrong one. The spec prescribes a trick: open a UDP socket to an external DNS address. No data is actually sent, but the system picks the outgoing interface itself, and that interface reveals the real LAN address. The second detail is echo protection.
How it was checked and what stays off-screen?
The author cites a smoke test with 28 checks and a live run of the page in a browser. One practical touch: Tailwind is loaded via CDN, but the styles are duplicated in custom CSS. If there is no internet, which is exactly the scenario the utility was built for, the interface will not fall apart. There are plenty of limitations too: This is a single post by a single author. There are no independent checks.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.