In partnership with
Have you set up an AI agent that runs 24/7 on your own hardware?
Tech Twitter has lost its mind over Clawdbot this weekend. 18,000+ GitHub stars in three weeks and a Discord server of 5,000 people swapping configs at 2am.
The hype is through the roof, so I had to set it up and talk about it.
Dev Tools of The Week
1. Blinky: Open-source AI debugging agent for VSCode that uses LLMs to identify and fix backend bugs through print statement debugging, LSP navigation, and runtime analysis. Inspired by SWE-agent, it lets you describe bugs, specify repro steps, and watch it debug autonomously.
2. CodeSquire: AI code assistant specifically for data scientists working in Google Colab, BigQuery, and JupyterLab. Turns comments into code, translates natural language into SQL queries, explains existing code, and tailors suggestions to your coding style, through a Chrome extension.
3. Code to Flow: Converts any code into interactive flowcharts, sequence diagrams, or class diagrams using AI. Supports all major languages (Python, JavaScript, Java, C++), helps visualize complex logic for debugging, and exports as SVG/PNG/PDF, has no code storage, completely safe.
What is Clawdbot?
Clawdbot is a self-hosted AI agent that lives inside your messaging apps like WhatsApp, Telegram, Discord, etc. Unlike ChatGPT, Clawdbot texts you back from the apps you already use.
Remember Poke? Yeah, rip.
Three capabilities separate it from every consumer chatbot:
Persistent memory across all conversations
Proactive messaging (it reaches out to you)
Full computer access via the terminal commands, browser automation, and file manipulation.
The Mental Model
Think of Clawdbot as three components working together:
The Architecture
Your phone sends a WhatsApp message → Gateway receives it → Gateway calls Anthropic API → Response returns through Gateway → Appears in WhatsApp.
For remote access, Tailscale creates a private mesh network so your laptop can reach the Gateway running on a VPS in Germany without exposing public ports.
Installation: Two Paths
Path 1: Testing Locally (10 minutes)
Good for experimentation. Your assistant sleeps when your laptop sleeps.
The wizard walks you through model authentication (Anthropic API key recommended), channel connections, and workspace setup. Your Gateway runs at 127.0.0.1:18789.
Path 2: Production Setup on Hetzner VPS (30 minutes)
For 24/7 availability at ~$5/month.
The $3.90 CAX11 (2 vCPU ARM, 4GB RAM) handles chat comfortably, and you can upgrade to 4GB+ if you want browser automations.
Memory Warning: 1GB droplets throw OOM errors during npm install. Either provision 2GB or add swap:
Tailscale Setup for Secure Remote Access
Tailscale creates an encrypted tunnel without exposing public ports:
Access your Gateway dashboard via http://<tailscale-ip>:18789/.
For HTTPS, Clawdbot auto-configures Tailscale Serve:
Connect Your Messaging Channel
Telegram (recommended as it uses official bot APIs):
Message @BotFather on Telegram, send
/newbot, follow promptsCopy the token into
clawdbot onboardwhen promptedMessage your bot and you'll receive a pairing code
Approve:
clawdbot pairing approve telegram <code>
WhatsApp: Scan QR via Settings → Linked Devices.
Warning: use a dedicated number. WhatsApp has no "bot" concept and may flag unusual activity.
Troubleshooting, Security, and Limitations
What NOT to Do
Browser automation is unreliable. LLMs misinterpret pages, click wrong elements, and struggle with dynamic content. One user's agent "accidentally started a fight with Lemonade Insurance" through an overly aggressive automated email.
Token costs add up fast. Federico Viticci burned 180 million Anthropic tokens in one week building automations. Running three agents costs ~$300/month in API fees. Use cheaper models (Gemini Flash, MiniMax) for routine tasks; escalate to Opus only for complex work.
Onboarding remains rough. Windows users must use WSL2 as native Windows is "untested and problematic." OAuth credential sync between Claude Code and Clawdbot frequently breaks.
Security Hardening
The documentation is blunt: "Running an AI agent with shell access on your machine is... spicy."
Essential measures:
Enable pairing: unknown senders receive a code and require manual approval:
clawdbot pairing approve <provider> <code>Keep Gateway on loopback (127.0.0.1): use SSH tunnels or Tailscale Serve for remote access
Set API spending limits in your Anthropic/OpenAI dashboard
For group chats, enable sandboxing:
agents.defaults.sandbox.mode: "non-main"runs non-main sessions inside Docker containers
Common Errors
Run clawdbot doctor to surface misconfigurations.
Run clawdbot status --all for a pasteable debug report.
My Take
Clawdbot is the first widely-adopted implementation of the "personal AI agent" concept that tech futurists have predicted for years.
But the tradeoffs are risky. You're granting an LLM shell access and browser cookies.
The security model assumes you understand what you're enabling, and token costs punish experimentation.
The right user for this right now is someone technical, with risk tolerance, who has specific automation needs that justify the setup cost.
Start with Telegram on a local machine and get a morning briefing working.
Feel the difference, then decide if the production setup is worth it.
Until next time,
Vaibhav 🤝🏻
If you read till here, you might find this interesting
#AD
The best marketing ideas come from marketers who live it.
That’s what this newsletter delivers.
The Marketing Millennials is a look inside what’s working right now for other marketers. No theory. No fluff. Just real insights and ideas you can actually use—from marketers who’ve been there, done that, and are sharing the playbook.
Every newsletter is written by Daniel Murray, a marketer obsessed with what goes into great marketing. Expect fresh takes, hot topics, and the kind of stuff you’ll want to steal for your next campaign.
Because marketing shouldn’t feel like guesswork. And you shouldn’t have to dig for the good stuff.
