DEV Community

carlosortet
carlosortet

Posted on

How I built a proactive personal AI assistant based on Moltbot for 10€ a month

An open source agent with persistent memory, command execution, real proactivity, and WhatsApp as its interface. Full architecture, real costs, and lessons learned.


Why Build Your Own Assistant

The idea of having an AI assistant that actually knows you is nothing new. Anyone who uses ChatGPT, Claude, or Gemini daily has felt the same frustration: every conversation starts from scratch, you have to re-explain your project context, and the assistant cannot do anything beyond the chat window.

At 498AS, we have been experimenting for months with language models applied to real workflows. Not just text generation, but agents that execute tasks, access tools, and maintain long-term context.

When we discovered Clawdbot (Moltbot), we saw an opportunity to close that gap. An open source project that lets you run your own AI agent with persistent memory, command execution, and the ability to message you before you message it.

We set it up. It works. And it costs 10 euros a month.


What Is Clawdbot

Clawdbot (Moltbot) is an open source project that provides the infrastructure for running your own AI agent. It is not a closed product or a subscription app. It is the technical skeleton on which you build your personalized assistant.

Its core capabilities:

  • Multichannel communication. It talks to you via WhatsApp, Telegram, Discord, or whatever platform you prefer.
  • Persistent memory. It remembers previous conversations, your projects, your preferences. It does not start from zero.
  • Command execution. It can run scripts, open applications, control your browser. Real actions on your computer.
  • Tool and API access. You configure which tools it can use, and the agent integrates them into its responses and actions.
  • Full control. It runs on your server, under your infrastructure. You decide what data it holds and who can access it.

The key difference from ChatGPT or Claude is that Clawdbot does not live in a third party's cloud. It lives wherever you decide.


Full System Architecture

The Central Server

The heart of the system is a VPS on Hetzner. Minimum specs: 2 vCPUs, 4GB of RAM. Model CX22. Cost: approximately 4 euros per month.

This server runs the Clawdbot gateway, which handles:

  • Receiving and sending messages
  • Communication with the AI model
  • Persistent memory management
  • Tool and node coordination

The AI Model

Subscription option. If you already pay for Claude Pro Max, ChatGPT Plus/Pro, or similar, Clawdbot can use that subscription directly. No separate API costs.

API option. If you prefer pay-per-use, you can connect the APIs from Claude, OpenAI, or any compatible model. For personal use it typically ranges between 10 and 30 euros per month.

WhatsApp as the Interface

The setup requires a dedicated prepaid SIM. Cost: 6 euros per month with minimal data. The agent has its own phone number. You message it like any other contact and it replies.

Security with Tailscale

The Clawdbot gateway is not exposed to the internet. It only listens on the Tailscale network.

Tailscale is a mesh VPN that creates a private network between your devices. End-to-end encryption, regardless of physical location. Free for personal use.

Our specific configuration:

  • The Hetzner VPS is on the Tailscale network
  • Our phones and computers are too
  • The gateway only accepts connections from Tailscale IP ranges
  • The firewall is explicitly configured: only Tailscale can reach the gateway

If someone scans the server from the internet, the gateway port does not even show up.

Distributed Nodes

Clawdbot lets you connect "nodes": computers that the agent can control remotely. My MacBook is connected as a node. I can be out on the street, message the agent via WhatsApp saying "open Chrome and look up X" and it does it on my Mac.

All traffic between nodes and the gateway goes through Tailscale.

Access Control

The agent does not talk to just anyone. We have an allowlist of authorized phone numbers. Permissions are granular: who can interact, which commands it can execute, which tools are available, whether execution requires prior approval.


The Proactive Agent: The Real Game Changer

This is the feature that turns a chatbot into an actual assistant. The agent does not wait for you to write. It messages you first.

  • Email alerts. The agent monitors your email and notifies you via WhatsApp when something relevant arrives.
  • Calendar reminders. It reads your calendar and alerts you ahead of meetings.
  • Custom alerts. Weather updates, price monitoring, GitHub repository activity.
  • Scheduled tasks. "Every Monday at 9, review my pending tasks and tell me what I need to do this week."

Proactivity completely transforms the relationship with the assistant.


Real Cost Breakdown

Item Monthly Cost
Hetzner CX22 VPS ~4 EUR
Prepaid SIM with data ~6 EUR
Tailscale 0 EUR (free tier)
Total infrastructure ~10 EUR/month

The AI model is separate. If you already have a subscription to Claude Pro, ChatGPT Plus, or similar, you pay nothing extra.


Lessons Learned After Weeks of Use

  • More useful than expected. What started as a technical experiment became a daily tool.
  • Proactivity is the differentiator. Having the agent message you when something important happens is a paradigm shift.
  • Security is non-negotiable. Tailscale, firewall, allowlist, granular permissions. All of it.
  • WhatsApp as an interface reduces friction. It is where you already are.

How to Get Started

Clawdbot is open source and available on GitHub with comprehensive documentation and an active Discord community.

  1. Get a cheap VPS (Hetzner, DigitalOcean, or similar)
  2. Install Clawdbot following the documentation
  3. Connect WhatsApp with a dedicated SIM
  4. Set up Tailscale on the server and your devices
  5. Define an allowlist of authorized numbers
  6. Configure tools and permissions

Security recommendations:

  • Do not expose the gateway to the internet. Use Tailscale.
  • Do not run as root.
  • Configure user allowlists.
  • Enable approval for command execution.

We wrote a detailed blog post with the full architecture and step-by-step setup: Read the full article on our blog


Article published by the 498AS team. Questions about the setup? Drop a comment or get in touch.

Top comments (0)