24 Hours with OpenClaw: The AI Setup That Made Me Feel Like I Finally Have a Chief of Staff

It’s been less than 24 hours since I set up OpenClaw, and I’m writing this from the airport wondering why I didn’t do this sooner.

This morning, my new AI assistant:

• Processed 15 consulting requests and categorized them by fit

• Drafted an email response about a Microsoft Teams expert call (I approved it in 30 seconds)

• Found my wife’s text about a dental appointment, blocked my calendar, and texted her back

• Researched the cheapest flights to Australia with stopover options

• Created a watchdog system so it can alert Amy if it goes down while I’m traveling

I didn’t have to code these things. It just… did them at my request.

The bottom line: OpenClaw turns Claude from a tool you visit into an assistant that works alongside you. If you’ve been frustrated with copy-pasting into ChatGPT or Claude, this is the unlock you’ve been waiting for.

---

What Is OpenClaw?

OpenClaw is an open-source framework that lets Claude (or other LLMs) run continuously on your machine with:

• Persistent memory: It remembers context across conversations

• Messaging integration: I interact via Telegram, but it supports WhatsApp, Discord, Slack, iMessage

• Proactive behavior: Cron jobs let it check email, monitor things, reach out to you

• Full computer access: Shell commands, file access, calendar, email via CLI tools

Think of it as giving your AI a body. It can read your files, send emails, check your calendar, and message you when something needs attention.

A Note on Security

I know what you’re thinking: full computer access sounds terrifying. Here’s how I think about it:

1. It runs locally. Your data never leaves your machine unless you explicitly configure external tools.

2. You define the boundaries. The SOUL.md file lets you set guardrails on what the AI can and can’t do.

3. Sandbox mode exists. For the paranoid (reasonably so), you can run it sandboxed. I do for development work, but I can’t for my executive assistant since it needs real access to be useful.

4. Trust is earned. Start with read-only access, then expand as you build confidence.

Is it riskier than a web-based chatbot? Yes. Is it riskier than the random npm packages you already run? Probably not. Your call.

---

My Setup (The 4-Hour Version)

Hardware: Mac mini (M2, 16GB). Runs 24/7 in my home office. You could use any Mac or Linux box. If you don’t have dedicated hardware, Cloudflare’s Moltworker lets you run OpenClaw on their Sandbox SDK for $5/month with no Mac Mini required, though the assistant capabilities will be more limited.

Installation:

curl -fsSL https://openclaw.ai/install.sh | bash

The wizard walks you through everything:

1. Connecting your LLM provider (I use Claude via Anthropic API)

2. Setting up a messaging channel (I chose Telegram)

3. Configuring your workspace

I’ll be honest: it wasn’t seamless. I asked my AI assistant a lot of questions about where to find API keys and how to answer the wizard’s questions. I had issues getting my Claude OAuth credentials set up. My assistant thought we needed to try a Claude Max gateway proxy, but no: OpenClaw natively supports your Claude Max subscription directly. Just be aware that if you consistently max out your subscription every month, Anthropic may cancel it.

After setup, you’ll find these files in your workspace that you can customize over time:

• SOUL.md: Defines the AI’s personality and boundaries

• USER.md: Everything about you, your bio, preferences, work context

• TOOLS.md: Local notes like calendar IDs and contact info

• HEARTBEAT.md: What to check periodically

For API keys and secrets, I recommend using environment variables or a secrets manager rather than storing them in plain text files. The OpenClaw docs cover secure credential handling.

Once you’ve written these files, you have a personalized AI assistant that knows your context.

---

What’s your biggest frustration with current AI assistants? Drop it in the comments. I’m curious what problems people are trying to solve.

---

What Made It Click for Me

I’ve tried a lot of AI setups. Claude Code, Cursor, custom GPTs, n8n workflows. They all require me to initiate. I have to remember to ask. I have to context-switch.

OpenClaw flipped that.

Example: I was in a Lyft heading to the airport when a consulting request came in. My AI:

1. Saw the email (via hourly cron job)

2. Matched it against my expertise profile

3. Sent me a Telegram message with the opportunity summary

4. When I said “yes,” it drafted a response

5. I said “add more availability.” It checked my calendar.

6. I said “send.” It replied to the email.

Total time from my end: maybe 2 minutes of typing on my phone.

Most consulting requests don’t pan out. When they do, they’re worth it. But the real win is streamlining this to near-zero cost so I never miss a good one.

I’m writing this now from an Alaska flight with nothing but free texting. I’m still working with my AI through Telegram. That’s the unlock.

Another example: My wife texted about a dental appointment. My OpenClaw saw it (iMessage integration), created a calendar event with 30-minute driving buffers, invited her, and texted her “Got it! 🦷”

I didn’t ask it to do that. It just knew that’s what I’d want.

---

The Neurodivergent Unlock

I’ll be direct: this is the best ADHD accommodation I’ve found in AI.

My brain drops things. Important emails sit unread. Calendar conflicts happen. I forget to follow up. The mental load of tracking everything is exhausting.

I manage my personal calendar, my business calendar, and my client calendars. Doing this as a human is impossible at scale. I’m just one person with a handful of clients. The cognitive overhead was killing me.

OpenClaw acts as my external executive function:

• Hourly email scans so nothing urgent slips through

• Morning briefings with my calendar and tasks

• Proactive reminders delivered where I’ll see them (Telegram)

• Memory files so context persists even when my brain doesn’t

The best part? Setting up new automations is as simple as “can you check X every day at 9am and let me know?” Done.

It’s like having a chief of staff who never judges you for forgetting.

---

The Real Talk: What It Can’t Do (Yet)

• No magic. It’s still Claude underneath. It makes mistakes. It needs guardrails.

• Setup isn’t trivial. You need comfort with the terminal and writing markdown files.

• Trust boundary. It has access to your machine. Think carefully about what you enable.

• Cost. Claude API isn’t free. Heavy use could run $50-100+/month.

Budget option: OpenClaw supports Google Gemini, which has a generous free tier. You won’t get Claude-level reasoning, but for basic monitoring and reminders, it works. I’d start with Gemini to test the setup, then upgrade to Claude when you’re ready to go deeper.

---

Should You Try It?

Yes, if you:

• Already use Claude or ChatGPT daily

• Are comfortable with command line basics

• Want AI to work with you, not just for you when asked

• Have repetitive tasks that need monitoring (email, calendar, messages)

Maybe wait, if you:

• Are new to AI tools

• Don’t have a machine that can run 24/7

• Aren’t ready to trust an AI with shell access

---

Getting Started

1. Read the docs: https://docs.openclaw.ai

2. Join the Discord: https://discord.com/invite/clawd

3. Install it: curl -fsSL https://openclaw.ai/install.sh | bash

4. Start with SOUL.md. Define who your AI is before giving it power.

The community is active and helpful. I went from zero to “holy shit this is incredible” in an afternoon.

---

Tried this approach? Share your results below. I respond to every comment.

---

If you want to see how I’ve configured my OpenClaw setup, including my SOUL.md, USER.md, and cron jobs, that’s coming in a future post for paid subscribers. Subscribe to make sure you don’t miss it.