The author's hands-on record of connecting 15 mainstream apps to Ant Lingguang 2.0

Since I got the beta for Ant Lingguang 2.0, this Personal Agent,

it has become my personal butler. So far, I've connected 15 major consumer apps through Skills, MCP, GitHub projects, or built-in methods.

Basically, in the "Help Me Manage" section I open, the scheduled tasks regularly refresh information worth paying attention to.

Lingguang isn't the first Personal Agent I've used—I've tested 8 in total, and the more I tested, the more I felt that a Personal Agent on a phone and one on a computer are simply two different species.

For a computer Agent, I want it to be like the ones we tested before—able to use computer use to browse web pages and click buttons to retrieve information for me.

For the phone one, my requirements are actually quite plain: as long as it's easy to configure. Whether you send me a link or a QR code, as long as it can connect, and stays effective once connected, that's enough.

So where exactly is the advantage of Personal Agents made in China?

I think the answer is actually pretty straightforward: they can connect to domestic services.

Connecting to GitHub and email is standard practice. Those overseas fitness-management and meeting-notes apps—I rarely use them, and even if I force myself to, I can rarely find anyone else using them with me. What I actually open every day are Mijia, Didi, Meituan, Taobao, and WeChat Reading. No matter how smart a Personal Agent is, if it can't connect to these, it just doesn't feel smooth enough for me to use.

Take Luckin and McDonald's, for example—I previously connected them on a computer Agent, so I could directly order coffee and food. So I, the idea guy, was thinking: in theory, could I make a super-capability expansion pack for Lingguang and see whether all 15 apps can be connected?

Let's start with the first card—the whole day began with it.

At 11:31, I casually tossed out a line: before I go to Disney's Halloween event on the 24th, keep an eye on what to wear based on Shanghai's weather for me.

Lingguang's reply was that on the 24th Shanghai would have light rain during the day, 22 to 25 degrees, and I should guard against rain and damp. A raincoat is more cost-effective than an umbrella; holding an umbrella while queuing for the parade is too troublesome, and you have to dig through your bag at security. Don't wear canvas shoes or white sneakers—stepping in puddles all day is very uncomfortable. Indoor attractions like Pirates of the Caribbean and TRON Lightcycle Power Run can serve as rain shelters.

Up to here it was still fairly ordinary. What was interesting was the line of gray text at the very bottom of the card.

The forecast is from the morning of 10/20; there are still 4 days until the 24th, so it may still be adjusted. On the 23rd it will give the final clothing list, and on the morning of the 24th it will remind me once more before I head out.

This kind of proactiveness is a fairly signature specialty of a Personal Agent.

I only asked once, and it broke it down into three time points for me.

I didn't tell it to schedule anything—it set that itself.

In other words, before, once you finished asking, the conversation ended. Now with Lingguang 2.0, even if I close the app, it keeps watching weather changes in the background.

Public information like weather, stocks, and flights can be checked without connecting an account, so I won't expand on that. What I was curious about was the other half, so I asked it another question: which of my accounts or CLIs can you connect to, to help me look at information?

It listed a whole bunch. Mijia, Roborock robot vacuums, Xiaomi cars, DingTalk, Didi, six email providers—QQ, 163, 126, Sina, Sohu, and Gmail—plus Baidu Netdisk, and so on and so on. The number didn't look alarming.

OK, I followed up with another question: is what you're connected to behind the scenes a computer that's always on? Are all these services I've connected to you saved long-term? Also, list them in a table for me—after all, this is also a classic feature of a Personal Agent, a long-running cloud computer. This is also the underlying infrastructure that lets Lingguang 2.0 "manage things to the end"—relying on an independent cloud host (with its own computer and browser) to carry long-range task scheduling, combined with unified cross-session memory. Even if I exit the conversation or even close the app, my task goals, preferences, and historical progress stay connected, and it still runs tasks and monitors status in the background.

At that point there were already 7 in the table: Meituan Errands, Didi Ride-Hailing, Amap, Fliggy Travel, and Mijia Smart Home. More would come later—it said 15, not one fewer.

Before breaking them down one by one, let me first compress the experience from a whole day of tinkering into one master prompt.

If you've also got Lingguang, just send it the whole thing, and it will walk you through connecting them step by step in one conversation.

## Personal Agent Configuration Prompt

I want to connect all the commonly used national-level apps to you and turn them into my own Personal Agent. Please walk me through connecting them one at a time in this conversation, handling only one at a time, and after each one is connected, ask me whether to continue with the next. For each service, follow this order: 1. First check whether it is currently connected; if it is, skip it 2. If it is not connected, tell me the easiest way to connect it, prioritizing authorization links or QR codes, followed by official MCP, official Skill, or command-line tools 3. If I need to apply for a Key or authorization code myself, tell me which page to apply on and which items I need to get back 4. After connecting, verify with one real query, then save the connection method as a reusable skill 5. Keys are stored only in your environment; do not repeat them in replies 6. For any action that costs money, such as placing an order, hailing a ride, or making a payment, first list the cost and details for me, and execute only after I say confirm 7. If it cannot be connected, find out the reason and tell me, then skip to the next one Checklist: Mijia (QR code authorization) Baidu Netdisk (QR code authorization) DingTalk (authorization link) QQ Mail (IMAP/SMTP authorization code) Didi Ride-Hailing (apply for a Key on the official Didi MCP website) Luckin Coffee open. lkcoffee. com McDonald's open .mcd .cn/mcp/doc Amap lbs .amap .com Fliggy flyai .open .fliggy .com/docs/quickstart Meituan Errands github .com/meituan/MT-Paotui-For-Client Alipay open .alipay .com WeRead weread .qq .com/r/weread-skills NetEase Cloud Music github .com/NetEase/skills Meitu clawhub .ai/meituskills/designkit-skills After everything is done, give me a table listing the service, connection method, current status, and what it can do for me.

Handle only one at a time,

because some require scanning a code, some require applying for a Key, and some just need a link dropped in.

Below I'll go through them one by one by configuration difficulty, from just scanning a code to needing to apply for a Key yourself. gogogo!

1. Mijia

The connection method is absurdly simple. Lingguang directly throws me a QR code, and I just scan it with the Mijia App and that's it.

It immediately found the 18 devices in my Mijia, 9 of them online, and also reminded me that the temperature and humidity meter in the living room was almost out of battery and needed a new battery first.

Then I raised the difficulty a bit: when I get up in the morning, the light and the plug (connected to the Mac mini) should turn on automatically, send me temperature and humidity every day, and also give me a reconnection tutorial for offline devices.

For turning on the light when getting up, Lingguang taught me step by step to create an automation for a wake-up mode in the Mijia App. Apple's Shortcuts next door can already be written with Apple Intelligence; here's hoping Mijia will later open it up so Lingguang can enter it directly.

For now, using ifelse to control a single appliance is still very simple.

2. Baidu Netdisk

Also a QR code scan, still no difficulty in use.

My pain point is that I'm too lazy to look through things one by one but I don't want to pay for a membership...

After scanning, it went through my netdisk and directly said: the bulk of your netdisk is basically course videos saved in earlier years; to free up space you don't need a membership at all.

So many memories—there was even the cat and dog dataset I used with OpenCV in 2017. Ridiculous...

3. DingTalk

My need is for it to summarize DingTalk group chats and emails for me every day.

For DingTalk, just click authorize. Lingguang also figured out the push method: the summary is posted in the conversation, and a notification is pushed to my phone. What I'm most satisfied with right now is that I can turn off DingTalk's nighttime message notifications; if there's really something urgent, Lingguang will push it to me.

Next I plan to also test Feishu and WeCom. WeCom itself has API interfaces, so logically connecting it shouldn't be a big problem. If Feishu can also be connected smoothly, then I can basically hand my daily office work over to it to watch.

4. QQ Mail

Usually, using a Personal Agent for daily summaries involves the four-piece set: public APIs, web pages, email, and RSS.

I very, very, very much recommend adding every mailbox you can connect. I once spent a week forwarding all my mailboxes to one fixed email address, thinking that from then on I'd only need to look at that one, but it still failed. Cross-platform compatibility is still insufficient; even setting aside time differences, messages still get sorted into the trash.

5. Meituan Errands

This one starts to get interesting. I only dropped a GitHub link, github. com/meituan/MT-Paotui-For-Client.

The addresses stored in errands are actually very useful. Later, when connecting some map-type apps, if you're too lazy to write the address again, you can even have it read the corresponding address from Meituan and remember it. For example, if I want to take a ride straight home, I can have it choose from Meituan.

This is also why I'm listing so many apps today: some apps' information can complement each other, which saves me a lot of configuration trouble.

I tried a very everyday need.

There's a really tasty Tokyo-style barbecue place in Hangzhou, and I wanted the errand service to grab a number for me first. It said it could handle it, and it got everything sorted—delivering the number ticket for that restaurant wherever I needed.

6 and 7, Luckin Coffee & McDonald's

Luckin also has an MCP, so I just pasted the whole MCP config block into the chat box.

Once connected, Lingguang immediately knew that the nearest Luckin to me was just over 300 meters away, was open for business, and could even order the latest items normally.

Next door, McDonald's also has its own MCP open platform. This time I didn't even copy the config—I just dropped a tutorial link and let Lingguang read the docs itself. After connecting, it helped me pick the nearest and most discounted two-person meal set, and check whether anything new had launched. Do you all know about McDonald's new All-Star Two-Person Sharing Meal eight-piece set? On Wednesdays I can treat it like Crazy Thursday too.

8. Didi Ride-Hailing

There's another blind spot: if you hail a ride through Amap, you can't choose an oil car or fresh air, but on an Agent you can.

My requirement is oil car ➕ fresh air, and to save home, studio, and Hangzhou Airport as trigger words, so that in the future I can hail a ride with one sentence.

This is exactly the scenario I mentioned earlier, where I can copy all the addresses from Meituan over and use them directly.

9. Amap

Still just drop a doc, plus the Web Key applied for on the open platform, and that's it.

Lingguang used this Key to try two paths. First it directly called the Web service interface to look up the coordinates of Xiaoshan Airport—it worked. Then it connected to Amap's officially hosted MCP, and pulled down route planning, POI search, and weather all at once.

10. Fliggy

Same as before: with a Personal Agent, as long as you know a service has an MCP, a Skill, or a GitHub project, configuring it is actually more convenient than an Agent on a computer. Fliggy also has an open platform, so as before, we connect everything that can be connected.

There was a small interlude in the middle: the automatic installation kept timing out when connecting to GitHub, so Lingguang switched to manual installation itself, wrote the Key into the config, and then tested it once with "Hangzhou West Lake hotel," which returned real hotel packages and booking links.

Then I asked: I often fly between Beijing, Shenzhen, and Hangzhou—is there a suitable unlimited-flight pass?

It recommended Loong Air's year-round unlimited-flight card; Loong Air's home base is right in Hangzhou. It also reminded me afterward that this kind of card usually requires paying the airport construction and fuel surcharge separately for each trip. Next year I'll tell everyone how much I saved.

11. Meitu

This time I first asked: for generating ultra-HD images, blurring/watermarking images, and removing passersby, does Lingguang need to rely on the Meitu CLI, or can it do it natively?

OK—text-to-image and simple blurring, Lingguang can do itself; poster generation, person removal, and ultra-HD upscaling can be supplemented through Meitu.

Notice Lingguang's reply in the right image above: the key has been stored separately in the sandbox—that's the Ant APASS trusted agent infrastructure mentioned earlier at work.

Let me show everyone the cat poster I made.

12. Alipay

I also went to Alipay with questions. It's the end of the year, and I really hope Alipay can have an Agent develop invoices for me and tally up what I spent this year.

However, the Skill Alipay previously made public—for personal daily transfers and payments, and viewing bills—Alipay has not opened a general interface; it's more for receiving payments, and one of them specifically supports personal scenarios.

During my testing, Lingguang 2.0 iterated very quickly, and later it directly natively connected the entire Alipay ecosystem.

I tried the most practical money-management scenario: having it sort out my Alipay bills for the last three months to see where the money went, focusing on those automatic-renewal assassins.

And because it's set up as a scheduled task, I can just give Lingguang 2.0 a budget, and it will tally on a schedule—much more useful than ordinary budgeting apps. Once it finds I'm at risk of exceeding the budget, it notifies me.

13. WeRead

I've been using WeRead for a long time; I previously used it to tally the books I read each week and turn them into action cards.

The native weread-skills can read the bookshelf, reading duration, and how far I've gotten in each book.

But I haven't had time to read lately, so the first piece of data connected in—the bottom of the card says this week's reading duration is 0. That's a public execution...

14. NetEase Cloud Music

Bet you didn't expect it—NetEase Cloud Music has one too...

github. com/NetEase/skills. I also brought along a question I've been unable to figure out lately: I've been feeling the songs I hear are getting more and more boring—why does NetEase Cloud Music recommend songs in three languages to me at the same time? But this config is a bit more troublesome; you need to set up a self-built app to complete the configuration.

15. Taobao

The last one: Taobao.

Of the 15, it's the only one that doesn't even require binding an account—just get straight to work. It was also the most "Personal" segment of the day.

I used voice to throw out a long, rambling message, roughly: I heard the iPhone 18 Pro dropped in price because of some coupon—help me check, but I only trust reliable ones like Apple's official flagship store or the 10 Billion Subsidy.

Lingguang's last paragraph was written like this: I've already set up price tracking for you—from the whole Double 11 through November 13, checking once every 12 hours.

I casually asked again about the iPhone Duo—indeed there are stores but no stock, preorders sold out in a minute, and the schedule says you can't get one until January 2027. It also checked the several news items I'd scrolled past one by one.

15 of them—all connected by now.

Looking back on this day, at the very beginning I actually had a bias.

Isn't a Personal Agent just the original Agent with a new shell? Configure a few working Agents, and you can get a similar effect.

After connecting these 15, I changed my mind.

The form on mobile is much richer than I thought.

Scan-code ones, MCP ones, Skill ones, CLI ones—all can be installed on it, like building blocks. What's more comfortable is that after installation, it proactively comes to me. Many scheduled tasks—I just said one sentence, and it arranged the time points itself, kuku.

In the past, we scrolled through a dozen information sources every day, and the cognitive pressure was huge.

Now that we've switched to an Agent, logically it should decrease.

But not at all.

I have a project where I let an Agent run for two or three days straight. It finished the work, but I had to spend a long time having it recount where it got to and whether the plan file was actually updated.

A Personal Agent is exactly the opposite: I don't have to chase it with questions. When the time comes, it can't wait to push everything in front of me.

So I did a bolder design.

Data anxiety, information anxiety—I'll just hand them both to it. I used an Apple Watch all-day recording device to record a whole day, uploaded it into Lingguang, and had it divide 10 alarms for me by duration, starting from 2 a.m. to check whether there are AI news updates, all the way through every node of the day.

Let me release another one and take a look—it really is the great way is simple. Calendars, reminders, email notifications are all useless; only the annoying alarm clock works...

The only thing left for me to do is one thing.

Trust it, and trust my own design.

Feed the data collected on schedule back to it,

and the Personal Agent will grow into the shape that suits me best.