Hello
I recently purchased a Mac mini M4 64GB and have been running Openclaw.
It's been about 2 weeks since installation.
With the Mac mini M4 64GB environment, I now understand why many people use the 16GB model.
It's so slow and frustrating that I can't use it...
At first, I was using qwen3.5 35b with ollama, but the configuration got messed up and it stopped responding
It kept looping infinitely, so recently I paid for Xiaomi's MIMO V2 API and tried it for about a week.
If I had continued using local LLM, I probably wouldn't have thought about making an app today.
Below are the bots I'm running.
The one at the top is the bot I created first. It handles the brain (mimo v2 pro) and I asked ChatGPT about the personality (soul) implementation and attached it well.
Two days ago, through Link (bot name), I first created a simple stock price and international news alert service using cron jobs.
As I kept giving it tasks, the waiting time became boring, so I created a separate briefing bot.
Link Briefing doesn't have an LLM connected; it's a bot that just delivers stock market and international news through Openclaw when the time comes (cron job).
At 12, I also send out a 2-minute podcast in conversation format hosted by two hosts (female, male) (subscribers are me and my wife)
Doing this for the first time, I realized how amazing it is to get real-time data. How many APIs did I connect just to create one briefing material...
Mario has the DeepSeek API connected and the DeepSeek chat model attached.
Mainly, if there are videos or documents to schedule or take notes on, I pass them on and it automatically categorizes and stores them in Calendar, Notion, and Obsidian. I'm not naturally a note-taker, but it's so convenient that I use it. It automatically summarizes, organizes, and tags everything.
Mario just receives the Telegram bot API from Link and if I tell him a few roles, he creates soul.md and scripts I've implemented so far (with Google Calendar API, etc.) himself and even connects the LLM, so it's easier.
Then today I had some time, so I moved Teslamate (a web service that integrates with Tesla vehicles and accumulates and displays data) that I had created on AWS to the Mac mini
and collaborated with ChatGPT (my money is precious...) to create an app.
At first, I made it as an iPhone app, but since I'm not going to deploy it and will only use it personally, I switched to a web interface. (I accept it coolly, really...)
I have no development experience so I don't understand everything they're saying
But looking at how well it does the tasks I've given it so far (mimo v2 pro), I'm confident.
In less than 10 minutes, it converted and completed testing...
The app that came out
Clean UI... I just entered a prompt and it did this much
Simple feature testing...
At first, I didn't explain in detail and just requested through chat, so it created an app that queries the DB connected to Teslamate and returns results, and I put that into a prompt for ChatGPT, so the result is a bit messy.
So I suggested attaching a local LLM as originally planned.
It checks the model on my PC and recommends it.
It was completed in about 15 minutes and testing was done.
Looking at the results, it's better than I thought.
(The actual mileage is higher, but something seems to have gone wrong when I backed up and migrated the data.)
The response speed is within 10 seconds, which satisfies me. Initially it took about 17 seconds, so I asked why
It optimized itself and completed testing.
Really amazing.
I have no development experience, just network programming assignments in school
At the company, I've only developed server scripts
In the past, I had ideas but lacked development skills, had no money, and was too lazy to execute, so I didn't do it
But I paid $50 in API costs and spent $10 out of that to create this
It's so fun that I'm losing track of time. (It's not a dopamine joke.)
Now what features should I add... Tesla is a moving computer and provides data generously
There's a lot to do ㅎㅎ
Simple conclusion
Don't spend money on memory purchases; spend money on smart model API costs
Openclaw configuration is annoying and difficult, but everything is solved the moment you attach a smart model.
The car is definitely Tesla
Thank you for reading this long post.
