Where the experiment came from

Recently, a friend saw an architecture diagram someone had shared on Threads and asked whether I could build something similar.

The basic idea was to pull messages out of a LINE group, store text in Google Sheets, save images and files to Google Drive, and then connect Gemini to organize and analyze everything.

My first reaction was: Hey, this actually looks useful.

If you translate the idea into everyday life in Taiwan, you notice something quickly:

A lot of people’s work information does not live in a formal system. It lives in LINE.

Company discussions are in LINE. Family travel plans are in LINE. Friend gatherings are in LINE. Customers send last-minute information in LINE. Vendors send quotations in LINE…

A lot of important information starts as something someone “just drops in the chat,” and later that becomes the only copy anyone has.

Then two days later, everyone starts asking:

• “Who sent that image?”

• “What conclusion did we reach last time?”

• “Does anyone still have that file?”

• “Can you post it again?”

LINE is extremely convenient, but it has one major weakness:

It is very good at receiving messages, but not very good at organizing them for you.

So I thought: why not build my own version and see what happens?

What I built was very simple

I started with a small experiment:

Connect a LINE group to Cloudflare and see whether chat messages can become data that is easier to organize, search, and extend.

The flow looks roughly like this:

LINE Bot
→ LINE Messaging API
→ Cloudflare Worker
→ KV / R2 / D1
→ 後續再接 AI 分析

The process itself is straightforward.

When someone sends a message, the Worker receives it first. If it is text, the content is recorded. If it is an image or file, it gets stored. Other things such as state, settings, and cache can be placed in different storage layers.

For today’s coding AI agents, building this part is basically a piece of cake.

The interesting part is this: once messages become structured data, many things that were previously difficult start to become possible.

For example:

• Automatically summarize what a group discussed today

• Classify important images and files

• Identify questions that keep being asked repeatedly

• Extract action items from discussions

• Help recall the conclusion from the previous conversation

In other words, LINE stops being only a chat tool. It starts to look more like an information entry point.

I think this has a lot of potential

I believe many people in Taiwan use LINE constantly, and this is a very real pain point.

A lot happens inside LINE, but it becomes difficult to find later.

You clearly remember that someone sent that image. You clearly remember that the group discussed that issue. You clearly remember seeing a link, a file, or an address in the chat.

But you still cannot find it, especially as LINE itself has started removing some features or moving them behind paid plans.

I think a lot of AI still feels somewhat disconnected from real-world scenarios. AI is very well suited to repetitive, messy work like this.

For example:

• Find the screenshot someone sent last time

• Organize travel information from the family group

• Remember restaurants recommended by friends

• Extract the important points from a group discussion

None of these tasks looks especially difficult. But making them smooth and genuinely useful still takes quite a bit of work.

Why Cloudflare?

I think Cloudflare is very convenient for this kind of small experiment.

You do not need to decide everything up front:

• Where should the server run?

• Where should storage live?

• What happens when traffic arrives?

• How should background jobs run?

• How should AI be connected?

Cloudflare already provides a full set of building blocks that you can combine gradually:

• Workers: receive webhooks

• R2: store images and files

• D1: store structured data

• KV: hold settings and cache

• Queues: process background jobs

• Workers AI: handle early summarization, classification, and organization

It lets me validate an idea very quickly, starting from the free tier.

Where this experiment could go next

Right now, this is still just a small experiment. It is not a product yet, and it is not a complete tool.

But I think it could grow in several directions:

1. Automatically organize group information

Summarize, classify, and extract the important points from group conversations.

2. Help you find information again

For example:

• Who sent what

• What was discussed on a particular day

• Where a specific image is

• What a particular file contains

3. Extract action items

Find “who needs to do what” and “when it needs to be finished” directly from the conversation.

4. Turn LINE into a memory entry point

This is the part I find most interesting.

Instead of treating LINE as the final destination for a conversation, treat it as an entry point into memory.

A lot of information is not missing. It is simply buried inside chat logs, and nobody actually organizes it.

If a system can capture, classify, and understand that information, then it becomes more than a backup.

Of the ideas above, I think 4. has the most value. Once you have enough data, you could potentially build AI agents that distill your friends’ or colleagues’ preferences and experience, and maybe even turn them into Skills.

Conclusion

Many people do not like LINE, and I am not a big fan either. But it is still the most widely used messaging app in Taiwan. From a technical perspective, I also think some parts of its design are interesting and naturally suited to integrations and secondary development. Next, I want to try connecting LINE LIFF and see what else can be built. If you are interested in this article or the experiments I am working on, feel free to reach out and discuss them.