Last week, someone sent a voice memo to their AI assistant. But they hadn't set up voice support.
Ten seconds later, it replied anyway.
Nobody told it to do any of that.
That's Clawdbot - an open-source AI assistant built by Peter Steinberger (@steipete), founder of PSPDFKit. It runs locally, connects to WhatsApp or Telegram, and unlike most AI tools, it messages you first.
The project hit 9K+ GitHub stars before Anthropic stepped in and asked for a rename. It's now Moltbot.
THE RESTAURANT CALL
Here's another moment that went viral:
Someone asked their Clawdbot to make a dinner reservation. OpenTable didn't work. So it used ElevenLabs to call the restaurant and complete the reservation by voice.
Same pattern. No instructions. Just problem-solving.
WHAT MAKES IT DIFFERENT
It runs on your machine. Mac, Windows, Linux — your hardware, your data.
It connects through the apps you already use: WhatsApp, Telegram, Discord, Slack. You message it like you'd message a friend.
It has full system access. Shell commands. Web browsing. Email. Calendar.
And unlike Claude Code or Cowork - it doesn't wait for you. It checks in. It sends updates. It reminds you about things before you ask.
That's the shift: from tool you use to assistant that works for you.
MY EXPERIENCE
I'm cautious with tools like this. I didn't want Clawdbot anywhere near my personal machine.
So I spun up a VPS, locked it down with Tailscale, and started small. Gradually giving it
harder tasks. Testing the boundaries.
What I like most: the proactive updates.
Every morning I get a digest. "Here's what's happening in Claude Code." "Here are three marketing angles for [project]."
I'm bullish on this. Background agents running 24/7, quietly improving things, surfacing
what matters. That's the future.
But I'm also paranoid about security. Every permission is intentional. Every tool it can
access, I've thought through.
It's early. It's exciting. Let's see where it goes.
THE SECURITY WARNING
Before you set this up, read this.
A lot of people are hosting it on servers with zero authentication. Open gateway ports.
Full access.
High-permission agents need high-permission security.
Here is a guide with the best security practices: https://lukasniessen.medium.com/clawdbot-setup-guide-how-to-not-get-hacked-63bc951cbd90
I personally used Hetzner with Tailscale to set it up. Here is an article with instructions to set it up: https://www.pulumi.com/blog/deploy-clawdbot-aws-hetzner/. I asked Claude Code to follow this guide and set up the VPS with Clawdbot installed for me.
THE MEMES
Show & Tell
I'm hosting a Claude Code show-and-tell tomorrow (Jan 30).
I'll demo my Clawdbot setup and how I'm actually using it day-to-day.
If you've built something with Clawdbot or anything else with Claude Code and want to share, sign up to present. Or just come watch.
I'll share the recording for anyone who can't make it live.
That's the week.
Abhishek