DEV Community

Six Dusts Lab
Six Dusts Lab

Posted on • Originally published at sixdustslab.xyz

How I Built an AI Agent That Handles WhatsApp, Telegram & Slack Customer Messages — For €264/mo

Every small business owner knows this pain: you wake up to 30 unread WhatsApp messages, 15 emails, and a Telegram group that exploded overnight. Half of them are asking the same three questions.

"What are your prices?"
"Are you open on Sunday?"
"Do you have availability next week?"

You spend 3 hours answering. By the time you reply to the last one, the first person has already gone to a competitor.

The Problem With Existing Solutions

I tried everything:

  • Chatbot builders (ManyChat, Chatfuel) → too rigid, customers hate decision trees
  • Hiring a VA → €800-1,200/month, inconsistent, limited hours
  • Auto-replies → "Thanks for your message, we'll get back to you" → customers hate this too

What I actually needed was something that could read my business context (pricing, services, FAQ, policies) and hold real conversations — not follow a script.

Enter HappyClaw

HappyClaw is an open-source AI agent system built on Anthropic's Claude Agent SDK. It's not a chatbot builder — it's an autonomous agent that:

  1. Connects to your channels — WhatsApp, Telegram, Slack, email, CRM
  2. Reads your business context — FAQ, pricing, product catalogue, policies
  3. Answers customers naturally — in any language, 24/7, in ~6 seconds
  4. Qualifies leads — scores them on budget, timeline, and fit
  5. Takes actions — books appointments, logs to CRM, drafts proposals
  6. Knows when to stop — escalates to a human when confidence is low

The Technical Architecture

For the developers reading this:

Customer Message (WhatsApp/Telegram/Slack)
    ↓
HappyClaw Core (Claude Agent SDK)
    ↓
Per-user isolated Docker container
    ↓
MCP Tools (12 built-in) + Custom Skills
    ↓
Response → Channel
Enter fullscreen mode Exit fullscreen mode

Key technical decisions:

  • No agent reimplementation — wraps Claude Code directly, auto-benefits from upstream improvements
  • Container isolation — each user gets a Docker container with persistent memory
  • Skill system — extensible SKILL.md files for custom workflows
  • Multi-provider — Claude Opus 4.6 primary, GLM failover

The core is MIT-licensed: github.com/riba2534/happyclaw

Real-World Results

After deploying for a property agency:

Metric Before After
Response time 6 hours 6 seconds
Messages handled automatically 0% 70%
Lost leads (slow reply) ~5/week ~0
Staff time on messaging 3h/day 45min/day

What It Costs

HappyClaw (managed by SixDustsLab)

Plan Monthly What You Get
Solo €29 + Claude (~€185) = €214/mo 1 user, 1 channel
Team €79 + Claude (~€185) = €264/mo 5 users, 3 channels
Business €349/mo (Claude included) 10 users, 5 channels, SLA

One-time setup: from €1,950 (we configure everything).

vs. Hiring

Part-time hire HappyClaw Team
Monthly €800-1,200 €264
Hours/day 4-5h 24h
Languages 1-2 All
Weekends No Yes
3-year total €28,800-43,200 €12,454

Founding 10 Program

We're selecting 10 businesses to be our founding customers:

  • 20% off setup
  • Lifetime price lock
  • Direct line to the team
  • Priority features

If you run a small business and spend too much time on repetitive messages, try the Telegram demo bot or visit sixdustslab.xyz.


Built by SixDustsLab. Open source core on GitHub.

Top comments (0)