The Ultimate Guide to OpenClaw
How to build, deploy, and command your own AI agent swarm with OpenClaw.
ClawdBot, later renamed to MoltBot (due to complaints from Anthropic) and later to OpenClaw, has made quite a fuss in the tech community.
Thousands of excited posts on X or LinkedIn, a craze over Mac minis, hundreds of hype-fueled YouTube videos with wild use cases.
The culmination of this bandwagon is Peter Steinberger (the programmer behind it) giving an interview to Lex Fridman.
Now he’s in talks with OpenAI and Meta about selling the whole project as open-source 🤯.
My Initial Hesitation
I followed the project curiously, but with caution.
For one, I was concerned about giving a third-party bot unrestricted access to my computer.
Secondly, I didn’t want to buy a Mac mini just for the sake of running the bot.
Thirdly, I already have a bunch of AI tools that I pay a hefty monthly fee for.
At the same time, I desperately needed a personal assistant (PA).
Booking meetings across multiple time zones, searching for flight tickets, making reservations - all of those minor everyday tasks demanded my energy and time.
As a CPO, I’ve always felt that I could invest this resource into my work with a much higher ROI.
The PA Problem
I tried solving this problem in an old-fashioned way - by hiring PAs, twice.
Both times it felt like they were making my life harder.
I had to over-explain, handhold, and do damage control for their slip-ups.
I was reluctant to give it another shot.
Then I thought: Why not vibe-code my own assistant?
I’m proficient at Claude Code.
The OpenClaw (MoltBot) architecture is open-source.
Why not just copy it?
But do it on my terms.
I wanted it to run on a cost-efficient (read: cheap) and isolated VPS server.
I want to have stronger control over the tools I give it access to.
Additionally, I’d prefer it using Claude Code as a primary brain for everything since I already have a Max subscription.
But most importantly, this seemed like a fun challenge to tackle.
The more I dug into the project, the more in awe I was at how elegant the OpenClaw architecture is.
👋 Meet Saul
Here’s what I ended up with.
Meet Saul.
He’s a persistent PA with a personality (OpenClaw calls it Soul, play of words is purely accidental) and a sense of humor.
He has access to my Google Workspace - emails, calendars, everything.
He’s great at browsing websites, doing whatever I ask him to, and digging for whatever information is needed.
Book this hotel? Bam. Find me flight tickets? Done.
Check me in on my upcoming flight? Sure thing, sir.
He talks with me over Telegram.
He’s a self-improving bot; he can code and upgrade himself - all from a single Telegram text message from you.
Need access to Slack? Sure, on it, would be baked in three minutes.
He has memory - everything we talk about is stored and elegantly compacted when the context runs dry. You don’t need to remind him of your passport or phone number twice.
But most importantly, he’s proactive. You don’t have to push him.
He will reach out first at the cadence that we both have agreed on is non-disruptive.
Such a level of human PA intelligence costs multiple thousands of euros, easily.
I pay 24€/month.
This is how much a VPS droplet costs.
In this article, you’ll learn how to build your own PA on your own terms without the risk of giving any sensitive information OR overpaying for a Mac mini.
We’ll cover everything - from message processing to memory management, persistence, and the tool ecosystem.
Most importantly, you don’t need to be a techie or even understand any of the concepts below.
You can just copy-paste this article into your Claude Code and watch the magic unfold.
Not only will you understand the full spectrum of use cases for Clawd/MoltBot, but you’ll also be able to spin up endless persistent agents tailored to your liking.
