OpenClaw Multi-Agent System — Build Guide
Build a personal AI agent system on macOS that runs 24/7, managed via Telegram.
This guide documents real-world experience building and operating a multi-agent system with OpenClaw on a Mac Mini. Every lesson comes from production use — not theory.
What You’ll Build
3 AI agents that automate your daily life, split by domain of concern:
- Personal Agent — email triage, calendar, meeting prep, expenses, Goodreads, Typefully
- Research Agent — web research (Crawl4AI), Reddit/YouTube/Facebook digests, critical writing
- Kioku Agent — emotional companion, health check-ins, long-term memory (knowledge graph)
All controlled via Telegram. All running on your Mac. ~$11/month API cost.
Guide Structure
Getting Started
- Prerequisites & Installation — Hardware, software, first agent
- Architecture Overview — How OpenClaw works, key concepts
- Your First Agent — SOUL.md, TOOLS.md, Telegram bot setup
Core Skills
- Google Workspace Integration — Gmail, Calendar, Drive, Sheets, Tasks
- Browser Automation — Playwright for Goodreads, Facebook
- Cron Jobs & Scheduling — Automated workflows, delivery, troubleshooting
Advanced
- Multi-Agent Design — 3-agent split by domain of concern
- 3 Agent Profiles — Detailed tools, workflows & diagrams for each agent
- Model Selection — Unified M2.7, fallback chains, local models
- Memory System — Kioku-lite, tri-hybrid search, knowledge graph
- Operational Playbook — Debugging, common errors, maintenance
Recipes
- Recipe: Morning Briefing
- Recipe: Email Triage
- Recipe: Content Digest
- Recipe: Book Tracker
- Recipe: Expense Tracker
Quick Start (5 minutes)
# Install OpenClaw
brew install openclaw
# Create your first agent
openclaw init
# Start the gateway
openclaw gateway start
# Chat with your agent on Telegram!
Read Prerequisites & Installation for the full setup.
Built with lessons from 2+ months of daily production use. Contribute on GitHub.