Most people have OpenClaw misconfigured - and that’s a real problem
If you’ve heard about OpenClaw and installed it without touching anything else, you basically have a glorified chatbot. Out of the box, that’s all it is: you ask it things, it answers, end of story. But properly configured, it’s a completely different animal. We’re talking about an agent that works while you sleep, sends you a notification when it finishes a task, remembers what you told it last week, and can coordinate other agents to handle more complex workflows.
In this post I’ll walk you through how to set it up properly from scratch. Whether you already have it installed or you’re starting fresh, this covers everything you need.
Before you install anything: where to run it
The first decision you need to make is where OpenClaw is going to run. Some people install it on their own machine, others on a Mac Mini, and then there’s the option I actually recommend: MyClaw.
A Mac Mini can easily run you $700 to $800, and on top of that you’d need to leave it on 24/7 if you want the agent always available. Running it on your own computer isn’t ideal either - the moment you shut it down or put it to sleep, the agent stops existing.
MyClaw is a specialized platform that spins up OpenClaw on a server for you - no terminal, no commands, full graphical interface. You configure everything from the browser, it comes with support, and the most interesting part: they have AI provider pricing that beats going directly to OpenAI or Anthropic yourself. You can bring your own API keys if you already have them, or use theirs at a discounted rate. I put in $30 to record the video and it worked perfectly. The only case where it might not make sense is if you already have a subscription that comes with credits, like Codex - but for most people, MyClaw is the more practical option by a wide margin.
The workspace: the files that change everything
Once you have OpenClaw running, the real work starts here. OpenClaw has a workspace - essentially a folder with config files that the agent reads every time it starts up. Without this configured, you don’t have an agent, you have a chatbot.
I’ve put together a repository with all of this already set up that you can clone directly: github.com/edunavajas/openclaw-workspace. It includes all the files I’m about to explain, already structured and with examples. I’ll cover how to install it in a bit - it’s simpler than it sounds.
Let’s go through each file.
SOUL.md: the agent’s personality
This is the file where you define how the agent talks and behaves. A lot of people fill it with stuff like “you are friendly and helpful” and then wonder why the agent responds like a user manual.
What actually works is being concrete: “have your own opinions,” “be direct and don’t beat around the bush,” “if you can handle something on your own, do it without asking me.” Things that genuinely change behavior. You need to rewrite it in your own voice - if you don’t, the agent sounds generic, and in day-to-day use that gets old fast.
USER.md: how it knows you
Every OpenClaw session starts from zero in terms of context. It doesn’t know who you are unless you tell it. The USER.md file is exactly that: you introduce yourself, tell it your name, your timezone, what projects you’re working on, what apps you use, and what hours you don’t want to be pinged.
The more complete this file is, the more useful the agent is from the very first moment of each session. It won’t take more than half an hour to fill out properly the first time, and after that you rarely need to touch it unless something in your context changes.
IDENTITY.md: who it is
This file defines the agent’s identity: its name, what it is, where it’s running. By default, OpenClaw tends to say things like “as a language model…” - and that’s exactly what we want to avoid. Give it a name, tell it what it is and where it runs, and you go from having a generic chatbot to something more concrete.
This also becomes especially important once you start having multiple agents coordinating with each other. If you’ve got a squad of agents and they all have the same undefined identity, chaos is guaranteed.
MEMORY.md: the most important one
Without this file, every session feels like meeting the agent for the first time. With it, the agent remembers your preferences, decisions you made together, things that went wrong last time, what worked. It’s the difference between a collaborator who knows you and one who always starts from scratch.
The trick here is that the agent doesn’t always save to memory automatically when it should. To make sure it does, add an instruction in SOUL.md or IDENTITY.md telling it to log new things to memory. That way it happens without you having to ask every time.
HEARTBEAT.md: from reactive to proactive
This is where OpenClaw makes the leap from tool to actual agent. The heartbeat is basically a periodic ping: every X minutes, the agent checks this file, and if there’s a task that needs attention right now, it acts.
You define the tasks and when to run them. “Every morning at 7, check the calendar and send me a summary.” “Every Monday, review the status of active projects.” “Whenever someone mentions me in Slack, notify me with a summary.” The agent stops waiting for you to say something and starts doing things on its own.
AGENTS.md: the operations manual
This is the file where you tell it what it can do on its own and what requires your sign-off. Think of it as a delegation contract. It’s also important to define behavior for group contexts like Telegram or Slack: when to speak up, when to stay quiet. If you don’t set this, the agent can become pretty annoying - jumping into conversations that have nothing to do with it.
How to install all of this
The github.com/edunavajas/openclaw-workspace repository already has everything structured: the six files I just described, a setup for a multi-agent squad, shared directories, and permission policies.
To install it you have two options. The easiest: from the OpenClaw or MyClaw chat itself, tell the agent to clone the GitHub repository and install the workspace (if you have the GitHub skill configured it’ll do it on its own - it’s included by default). Or manually from the terminal, which honestly isn’t that complicated either.
Once it’s installed, the agent walks you through filling in your profile. There’s a fair amount of information to enter - I know it feels like a chore - but you do it once properly and it’s set for good. You only need to revisit it if something significant changes in your life or projects.
Real-world use case: the morning briefing
To show you how all of this works together, here’s what I actually use myself. Every morning, before I’m even out of bed, OpenClaw checks my calendar, pending mentions, email, and news I care about - then sends me a summary over Telegram or WhatsApp.
I don’t have to ask for it. It happens because it’s configured in the heartbeat. I wake up and the context for my day is already waiting for me.
The repository includes a ready-made prompt for this in the prompts/ folder. It’s as simple as copying it and sending it to the agent. It’ll ask you for the credentials it needs as it needs them - you don’t have to dump everything upfront.
What’s coming next
The next video is going to be about giving the agent access to a GitHub account so it can make improvements to personal projects autonomously. If this already sounds useful, you’re going to like what’s next a lot more.
If you install the workspace and run into any questions, drop them in the comments or reach out directly. The repository also has its own README with more detailed instructions for each file. And if you want to get started as fast as possible, remember that the quickest path is through MyClaw - it handles all the infrastructure for you so you can focus on what actually matters: configuring it right.
What do you think?
Leave your opinion, question or suggestion. Comments are synced with GitHub Discussions .