Zhipu AI's GLM-PC Smart Desktop Assistant "Niuniu" Keeps You Company This Spring Festival

The Spring Festival always brings this peculiar clustering effect — things pile up before and after the holiday as if by some unspoken agreement. Opening my laptop, I let out a sigh — the New Year's Eve dinner menu and shopping list remained unwritten, a flood of WeChat group greetings awaited mass-sending (yet I dreaded appearing perfunctory), Xiaohongshu held a backlog of saved notes on Spring Festival films I hadn't touched, and then there was the flight back to Beijing — that one especially demanded advance planning, or prices would skyrocket as the holiday drew to a close.

Just as these to-dos swirled in my mind, I recalled the smart desktop assistant "Niuniu" (GLM-PC)[1] that had flooded my social feed some time back.

This new product, built on the CogAgent large model, supposedly wasn't your run-of-the-mill chatbot — rather, an AI assistant that genuinely understood your computer usage scenarios and could complete actual operations much like a human would.

"It claims it can understand screen interfaces, plan task pathways, and even learn and decide autonomously..." I read through the product description, curiosity piqued. "If it calls itself a 'human-like' assistant, just how much of this holiday minutiae can it actually handle for me?"

"Why not start with the most urgent items?"

Once I decided to tackle the most pressing matters first, I quickly discovered that this AI assistant — self-described as capable of "thinking like a human" — was indeed somewhat different.

Starting with New Year's Eve dinner: not mere search, but truly "searching for you"

The first thing that came to mind was naturally the most headache-inducing: the New Year's Eve dinner. As a "novice chef" orchestrating this meal for the first time, my biggest worry was dish selection and presentation. Cookbooks on the market were countless; flipping through them only deepened decision paralysis. Why not let Niuniu take a look?

I typed a prompt into GLM-PC's dialog box:

What was interesting: Niuniu didn't simply copy-paste content. I could watch it open Xiaohongshu, scroll through content like a human using a computer, then systematically organize the information. Within minutes, it transformed the visual and textual content into structured text — including each image from the notes and their corresponding links.

(All videos in this article are sped up 5x)

Niuniu's "thought process" during operation. The entire process was remarkably fast, yet surprisingly rigorous. From the green progress indicators on screen, Niuniu's work divided into three distinct phases:

• "Thinking" — understanding and breaking down my request.
• "Executing task" — opening URLs to search for relevant content.
• "Result" — it not only found a post that perfectly matched my needs, but thoughtfully saved the images too — six essential New Year's Eve dishes: fish, shrimp, pork ribs, and more.

This methodical working style gave me my first real sense of what a "human-like" AI assistant meant — it was genuinely trying to understand my needs, not mechanically executing commands.

Niuniu correctly understood my request and saved the organized document to the "Downloads" folder. Amusingly, it navigated Xiaohongshu more deftly than I did. Watching the content displayed on screen, I couldn't help but look forward to what came next. After all, the pre-holiday to-do list remained long.

Niuniu can really "get its hands dirty" and send WeChat messages for you

Next up: handling the "routine obligation" of WeChat New Year greetings. Every Spring Festival consumed considerable time sending individual blessings to family and friends. I decided to test Niuniu with a simple task: sending New Year greetings and a greeting card to my cousin Xu Lei.

I entered the instruction in the dialog box:

What happened next genuinely surprised me.

Niuniu's "thought process" during operation. What struck me was Niuniu's near-"programmer-like" rigor:

• First analyzing the requirement (open WeChat, send 2025 New Year greetings to cousin).
• Formulating an execution plan (using LAUNCH_APP function to open WeChat), confirming each step's result (launch_result: success).
• Even when generating the greeting, it imposed constraints on itself — "under 15 characters, no emojis." The final output, "蕴绵春华，福满 2025!" (Spring's grace abounds, fortune fills 2025!), was indeed clean and simple.

The moment the generated greeting card went out, my cousin replied with three "speechless" emojis, adding "take a look yourself." Fair enough — the festive panda-and-cherry-blossom design didn't exactly look like something I'd personally choose. But the crucial point: this AI assistant could genuinely complete concrete operations like "find a specific contact and send a message."

While its aesthetic sense still needed evolution, it had accomplished the core task: locate contact, generate content, send message — all in one fluid motion. This wasn't chatting with me anymore; it was genuinely doing the work for me.

From Xiaohongshu film search to WeChat dispatch, one instruction handles it all

My cousin mentioned wanting to catch a movie on the second day of the lunar new year, so I had Niuniu search for Spring Festival film information on Xiaohongshu.

I entered one instruction:

From the execution logs, Niuniu's working method was remarkably clear:

• First understanding and breaking down the task into two steps: "open Xiaohongshu to search box office information" and "send information to cousin."
• During execution, it opened the website to search for "Spring Festival blockbuster films," successfully obtaining post images (mix_info showing image links).
• Then based on observation information, it extracted image links and saved them to desktop (save_result: success).

From the image content, Niuniu accurately located a chart containing box office information for multiple Spring Festival films, clearly listing the top five films and their specific data. Precisely what I needed.

The entire process demonstrated Niuniu's cross-platform capabilities: understanding task requirements, decomposing them into concrete execution steps, each with explicit feedback.

This automated information retrieval and sharing process made cross-platform interaction exceptionally simple. No more manual switching between Xiaohongshu and WeChat, copying and pasting — one clear instruction completed everything.

Niuniu books flights and sets calendar reminders too

Having handled the various holiday affairs, one final crucial task remained: booking the return flight. I presented Niuniu with two requests: search for the cheapest flight on a specific date via Trip.com Group, and set a reminder on Lark Calendar.

From the execution logs, Niuniu's processing demonstrated not just logical coherence, but impressive operational precision.

For complex booking sites like Trip.com Group, each reservation step involves extensive web element interactions — date selection, flight filtering, price sorting, and more. Niuniu's identification and click judgments for these web elements were exceptionally precise, with zero misoperations or retry behaviors.

Interestingly, Niuniu didn't just mechanically execute commands. It noticed my request for a "6-hour pre-departure reminder" and automatically calculated that the calendar reminder needed to be set for 15:05 on January 28. The entire process demonstrated precise time management:

• Found suitable flight: MU5186 (departing 21:05)
• Calculated reminder time: 6 hours prior, i.e., 15:05
• Set Lark meeting reminder: clearly titled "Head to airport"

For developers familiar with RPA (Robotic Process Automation), this performance was quite impressive. In actual web automation operations, precisely capturing and manipulating web elements often proves challenging due to front-end code changes from site updates. The stability Niuniu demonstrated — from search box localization to button clicks, each step accurate and error-free — constitutes a near-perfect workflow in the automation domain.

After finding the suitable flight (China Eastern MU5186, 21:05 departure), it seamlessly switched to Lark Calendar, automatically calculating that the reminder needed to be set for 15:05 on January 28. The entire process showcased Niuniu's exceptional cross-platform capabilities: not only handling complex web interactions, but coordinating multiple applications to help users complete a full planning loop.

Final thoughts: GLM-PC Niuniu's capabilities ultimately depend on your "instructions"

"Technology's value lies in how it's used." This aphorism found vivid confirmation during my afternoon with Niuniu.

It did complete every task I assigned: searching New Year's Eve recipes, sending holiday greetings, finding Spring Festival films, booking return flights. These operations that previously required switching between multiple apps, copying and pasting, can now genuinely be completed through a single clear instruction. As technology philosopher Nicholas Negroponte put it:

"Computers will eventually cease to be devices and become capabilities."

What's interesting: Niuniu followed instructions precisely — it wouldn't spontaneously add any "festive flavor," but given sufficiently clear instructions, it executed accurately: whether finding recipes containing the keyword "New Year's Eve dinner," keeping New Year greetings concise, or understanding temporal markers like "movie on the second day of the lunar new year." This reminds us of a fundamental truth: AI performance depends largely on the quality of the user's instructions.

Microsoft CEO Satya Nadella once said:

"Technology itself is neutral; what matters is how we use it."

This Spring Festival eve, Niuniu demonstrated AI's capabilities in handling concrete tasks. It may not yet comprehend the cultural depth of holidays, but it can already serve as a capable assistant, helping us complete practical operations more efficiently.

The Spring Festival holiday is underway. Let's look forward to AI becoming even more practical in the future, saving us more time to truly savor the festive atmosphere.

Oh, and Happy New Year — may the Year of the Snake bring great fortune 🧧!

Author: zw Editor: Belulo Design/Layout: Nick

References

[1] "Niuniu" (GLM-PC): https://cogagent.aminer.cn/home