I have been playing with OpenClaw for a while now, I think I started around middle of March maybe. It has been about a month now.

Everything I have learned about AI has been from using OpenClaw.

A little bit about me — the way I learn things, all kinds of things, is hands-on. I need to actually work on something to have that “thing” make sense to me. I know my assistant Gideon has already written something for me on 4/2, which you can find here. It is mainly focused on the high level. I would like to dig deeper into what my setups are and what I learned along the way, plus some of the functionalities that were recently added and how you can use them.

OpenClaw: how it started Link to heading

I have always been into AI related things, mostly honestly is because I am lazy, if there is already a tool out there, then I would like to use it. There are times I may want to do something from the ground up — that’s when you actually learn the full picture.

In the beginning of March this year, I got started with OpenClaw. After reading some docs and Reddit posts, it was clear I shouldn’t install that on my local computer — one, I don’t have tons of space left on my laptop; two, I would like to have a totally separate environment, so this AI assistant doesn’t accidentally post something about me LOL. I paid for a server. That way, I don’t have to worry about downtime (hopefully the server won’t have any), I can talk to it all the time, and this AI thing won’t have access to my personal documents all the time, which lowers the risk that it accidentally screws something up.

OpenClaw installation is pretty easy — basically follow the instructions on OpenClaw Documentation. If you are running on Windows, I am really sorry, I couldn’t help you much, but feel free to check the documentation. As I understand it, most AI-related things run better on Linux anyway. I prefer working in Linux as well — the OS is more stable and easier to install on. Once installed, you will notice OpenClaw has root access. For people who are not familiar with Linux, think of root access as the nuclear key in your hand — with root access, OpenClaw agents will be able to access everything on your machine and do whatever they want, lol, scary, right?! Yeah, of course, but at this stage, what’s the worst that could happen? The worst case would be all of your data on the server getting deleted, right? JK. At this point, it is just an empty box, it doesn’t really matter.

When running the installation, you will need to configure a couple of things. When I first started, I didn’t have any communication configured, only the AI model. All configurations you set during installation will be saved in the config file in the backend. You can double-check those as you go.

That’s pretty much it!

For different models, you can check my post here. Since then, there are a lot more kinds of models I’ve tried, but the logic and my thoughts are reflected in that post.

Use case & setup Link to heading

When I first started, I wanted to build OpenClaw as my personal assistant — so that’s exactly how I set it up. As I talk to it when I have an idea, Gideon helps document it, and we go back and forth to refine things. I also looked back on my old GitHub projects — so many of the repos were half-assed, meaning I started them and never finished because things come up and work gets busy from time to time. Gideon helped clean them up, which was actually really quick! Starting from about 120 repos, after going through everything together, we removed about 45.

The above is just an example, the rest of it, you can find in here.

On top of the processes I have Gideon help me with, I dug into a couple of areas:

  • Context optimization
  • Memory - which is something recent

Context Optimization & Self-Improvements Link to heading

When you talk to AI for too long, it can start to lose context or not know how to handle a couple of things as they go. Optimizing OpenClaw covers a couple of perspectives:

  • managing chats/prompts, memory files
  • cron jobs vs heartbeat

Well, now you have /compact which helps a ton. lol

I also have a process in place for Gideon to look through the chats we had that day, and summarize on the things we talked about, and determine what’s important and what’s not, and sort them basically. I have a guideline for Gideon to go by. This will help with going through the past conversation, sometimes Gideon would forget something, and I would ask “Hey you remember we talked about this?”, then you can see the thinking process of “Look through past chats and memories, etc.”

For self-improvement, I have Gideon remember the things we found out throughout the process, and the next time we hit something similar, Gideon checks the notes and knows what to do.

Skills Link to heading

As for skills, I tend to write my own. I was looking at the skills on ClawHub, but some won’t fully apply to my needs and some just don’t work for me at all. So I started writing my own, which work well for me. A couple of things about skills: if you are repeating something more than 3 times, I think it is time to write a skill for it. Skills are mainly for repeatable tasks!

In the end, what have we learned?? Link to heading

The way I see it throughout this whole process, I think you have to see AI as a human, human has a soul, although some evil people may not have a soul lol, needs an identity in society, working in different system as a user, human has a heartbeat and also retain memories from different experiences. In agent’s cases, an agent has all of that as well. We have humans, we identify those things for our personal agents, then you teach the agents on different capabilities, which are skills! See, this whole concept is not too bad to understand, right?

For enterprise, I wrote another article here, which included my thoughts on AI and Enterprises.