In my case, I have set up and use separate builds for company use + personal use.

First, when it comes to company use,

Many people say they can't really use it due to security concerns, but on an isolated server (built in a NAT environment with no external IP, while allowing outbound), I'm using it only through isolated channels. To improve company efficiency, I implemented it with the most restricted task permissions possible and kept it thoroughly isolated from the outside.

1. The inconvenient manual work of looking up syslog logs by application -> then finding sentry logs

I just gave it search tips and made it analyze the error flow and do the analysis all at once. Normally when this issue occurs, it takes at least 5 minutes to look up, but now you only have to wait 10 seconds and it comes out...

Of course, instead of directly accessing the syslog server, I made it accessible only through a somewhat restricted API, and gave sentry only API read permission.

2. Looking up only the necessary information from the backup DB and sending it to the partner companies we collaborate with

There's a data format that the collaborating company requests, which I used to look up annoyingly,

but now, by giving only restricted read database access, it looks up in real time from the hot backup DB, writes the content, and only goes as far as creating the mail-send queue,

and I made it so a person checks at a set time and sends the mail. It's not a DB subject to iSMS, and since they usually only request access logs, I made it look up only the log DB and send.

3. I made a bot that analyzes the code style and rules currently stored in the repository, creates relevant guidelines in an md file, receives events via webhook when code is pushed from gitlab, analyzes the changed code in the commit log, and automatically does a code review for the person who pushed it.

The code review still lacks a lot, but it catches and notifies you of minor syntax errors and code with different rules. What's scarier is that it gives feedback on security elements that weren't filtered out in validation....

This is now one of the bots that my team members are most satisfied with.

4. Issues come in from other teams via Slack, and I made a bot that automatically registers the issues into youtrack. It analyzes past issue done information to figure out what kind of work each person mainly handled, creates guidelines in md, and registers issues automatically based on that information. It filters out duplicates amazingly well.

The downside is that, perhaps because the absolute amount of learning data is small, there are cases where the registered issue isn't assigned to the proper person in charge.

Besides these, there's more stuff to build, but I only built and deployed these for now, and the team members are quite satisfied. Because of this, everyone has become more interested in openclaw, so everyone bought a Mac mini on installment and, while waiting, set up VMs for personal use and are testing.

For personal use....

1. Sending reports on US market, Korean market index information, and analysis of each sector and sharply falling/rising stocks at set times

Actually, on Reddit right now there are a lot of people attempting automated stock trading programs (mostly day trading), but it seems no one has seen results yet. I'm just building up insight about the market with the reports alone.

I'm also getting advice on various risk signals related to stock fluctuations and investment direction.. but predicting the future is hard since even AI can't do it, while it reads technical charts well. I set rules like hearing about the 1st defense line and 2nd defense line info when prices fall and deciding the buy timing, and getting advice on the 1st resistance line and 2nd resistance line to partly decide the sell timing, but I didn't really see much fun from it...

I also made it scrape US-related news in real time and alert me to news that seems important, and it's very useful for seeing the market direction.

2. Recording coin fluctuation rates and successfully predicting win rates related to coin fluctuation rates on Polymarket

There were people in a certain community who put money on Polymarket and made money by predicting win rates with AI, and after hearing that among those the win rate was coin fluctuation rate, I'm testing it, but when I invested virtual money, I've already become a beggar loser(?).. Besides this I'm also trying to approach other parts, but since this really is a gambling den, predicting win rates is really hard. It seems like a great way to ruin yourself if you carelessly approach it without insight.

3. Registering info on products I want to buy and getting lowest-price alerts

This is for getting an alert when the lowest price changes for something I want to buy. I implemented it using the CDP method, so it even avoids bot detection. It's working quite well.

4. Scraping the direction in which openclaw is being utilized

Since this area is moving actively these days, I'm collecting information from some foreign communities and checking what is mentioned the most.

There's a lot of various discussion, and the most mentioned thing is still mostly "how to install it," so there's nothing noteworthy yet, but it seems most people are making a lot of attempts as personal projects.

5. Making it possible to do vibe coding anytime on my phone wherever I go

In the past you had to bring a laptop to code and check,

but now I give instructions from my phone, run the test code, and view the results on a webpage,

and at the start of a project I can set the planning direction, receive the planning content and prototype as HTML, and view it according to the flow.

At this point, along with the urge to view it on a big screen.... the Galaxy Fold temptation came.

What's even more amazing is that to run all of this, the usage of the codex I bought via the Kakao special deal for the openai pro that I bought as a limited batch of 5 this time is very, very plentiful.

In the case of gemini, when I first implemented things, using pro was a bit insufficient, and even when using Claude's x20, I could see the opus usage rapidly decreasing, but codex didn't decrease at all.

So I build new features with codex, do planning and development orchestration with opus, and handle miscellaneous schedulers and responses with gemini pro3.

This is the extent to which it's being used now, and I'm both afraid and excited about how many usage methods and even more features will come out in the future...

I really feel like we are now living in an era of great upheaval.