01 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

How I Run a 20‑Client Agency with AI Agents

Self-hosted. Open source. In your pocket.

Roger Yin

@daluoseo · HashSEO

docs.openclaw.ai ↑ ↓ Arrow keys to navigate
02 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

You're Drowning in Ops

  • 40%+ of your time goes to client comms, task management, and reporting
  • Scaling means hiring — but hiring kills margins
  • Context switching between 10-20 clients destroys focus
  • Your team does ops work that AI should handle
40%

of agency time spent on operations, not strategy or growth

03 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

OpenClaw in 30 Seconds

  • Self-hosted AI gateway that connects WhatsApp, Telegram, Discord, Slack to AI agents
  • Message from anywhere — phone, laptop, client meeting
  • Your data stays on YOUR machine. No SaaS vendor lock-in
  • Open source, MIT licensed. Free to use and modify
📱
20+ Channels
WhatsApp, Telegram, Discord, Slack, iMessage, and more
🔒
Self-Hosted
Runs on your machine. Your prompts, your data
🤖
Multi-Agent
Run specialized agents for different tasks
🧠
Memory
Agents remember context across sessions
04 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Email & Client Triage

  • Auto-scan inbox and prioritize by urgency level
  • Flag overdue client replies, payment issues, urgent requests
  • Classify: CRITICAL → HIGH → MEDIUM → LOW
  • Your agent tells YOU what needs attention — not the other way around
"What emails need my attention today?"

→ 🔴 Client X invoice overdue 14d
→ 🟡 Prospect reply pending 6h
→ 🟢 3 newsletters, 2 promos
05 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Technical SEO Audits on Demand

  • Message "run audit on clientsite.com" → triggers Screaming Frog crawl
  • Automated 73+ checks across 9 categories
  • Prioritized report: Critical → High → Medium → Low
  • Output as Google Doc or spreadsheet, client-ready
🔴 CRITICAL (3)
  • Mixed content (HTTP on HTTPS)
  • 12 broken internal links
  • Missing canonical tags

🟡 HIGH (7)
  • 34 pages missing H1
  • Duplicate title tags...
06 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Content Production Pipeline

  • SERP-driven article writing with semantic SEO methodology
  • Agent researches SERPs, builds outlines, writes drafts
  • AI detection + humanization built into the pipeline
  • Human approval checkpoints — you stay in control
01 SERP Analysis & Entity Extraction
02 Outline Generation ← You Approve
03 Full Article Draft
04 AI Detection Check
05 Humanization Pass
06 Final Review ← You Approve
07 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Client Reporting & Health Monitoring

  • Auto-pull GSC data, keyword rankings, backlink metrics
  • Generate monthly performance reports automatically
  • Daily client health checks — uptime, task status, payment
  • Anomaly detection: "Visibility dropped 20%" → instant alert
✅ 16/20 clients healthy
🔴 acmeco.com — payment overdue
⚠️ novatech.io — 2 tasks overdue
📊 brightpath.com ↗️ / zenleaf.co ⚠️
08 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Prospect Audits & Proposals

  • Message "audit prospect.com" → full pre-sales SEO audit
  • Competitor analysis, keyword gaps, technical health
  • Backlink profile, brand SERP control assessment
  • Output: client-ready proposal document
📋 Executive Summary
🔍 Technical Health Score
📊 Traffic & Keyword Analysis
🔗 Backlink Profile
⚔️ Competitive Landscape
🎯 Prioritized Recommendations
09 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Multi-Channel Client Management

  • Each client gets a Discord channel — agent monitors and responds
  • Press releases drafted and published via API
  • ClickUp tasks auto-created from conversations
  • One gateway handles all channels simultaneously
Telegram DM → Main Agent
Discord #brightpath → PM Bot
Discord #novatech → PM Bot
WhatsApp Group → CS Bot
Slack #internal → Ops Agent
10 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Personal EA & ADHD Productivity

  • Daily Top 3 priority system with focus reminders every 2 hours
  • Calendar scanning, meeting prep, notifications
  • Heartbeat system — agent proactively checks on things
  • ADHD mode: micro-steps, push don't ask, zero dead air
☀️ Good morning. Your Top 3:

1. 🔴 Brightpath audit due today
2. 🟡 Prospect call at 2pm — prep ready
3. 🟢 Review content drafts (3 pending)

⏰ Next meeting: 10:00 — Team sync
11 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Install in 5 Minutes

Three commands. That's it.

# 1. Install OpenClaw curl -fsSL https://openclaw.ai/install.sh | bash # 2. Run the setup wizard openclaw onboard --install-daemon # 3. Pair a channel (Telegram is fastest) openclaw channels login # ✅ Done. Message your bot.
12 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Make It Worth Your Time

📱
Start with Telegram
Simplest setup — just a bot token. WhatsApp needs QR pairing.
🎭
Write SOUL.md
Give your agent personality, rules, and style. This is what makes it yours.
🧩
Install 2-3 Skills
Browse clawhub.com. Pick skills that match your workflow.
🎯
Nail ONE Use Case
Email triage? Reporting? Pick one, perfect it, then expand.
13 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Everything Is a File

Plain Markdown. No database. No vendor lock-in. git it if you want.

~/workspace/ ├── SOUL.md ← Agent personality & rules ├── USER.md ← About you (timezone, prefs) ├── MEMORY.md ← Long-term memory ├── TOOLS.md ← API keys, tool configs ├── CLIENTS.md ← Client roster ├── memory/ ← Daily logs (auto) │ ├── 2026-03-05.md │ └── 2026-03-04.md └── skills/ ← Installed skills ├── sf-technical-audit/ └── semantic-article-writer/
14 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Why Multiple Agents?

  • One agent can't do everything well — context window fills up
  • Specialized agents = better system prompts = better results
  • Isolate client contexts for security + relevance
  • Each agent has its own memory and workspace
You don't hire one person to do sales, accounting, writing, and IT.

Same principle applies to AI agents.
15 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Example: 5-Agent Agency

🧠
Main (Dali)
Personal EA. Operations chief. Orchestrates everything.
✍️
Content (Luna)
Article writing, social media, brand voice content.
📋
PM Bot
Client task management, ClickUp integration, status updates.
🔍
Audit Bot
Technical SEO crawls, Screaming Frog, automated reports.
🤝
Customer Success
Client-facing comms, WhatsApp/Discord monitoring.
🔧
+ Sub-agents
Spawned on-demand for specific tasks, then cleaned up.
16 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Channel → Agent Routing

One Gateway process. Multiple channels. Each routed to the right agent automatically.

📱 Telegram DM Main Agent
💬 Discord #client PM Bot
📞 WhatsApp Group CS Bot
🖥️ Slack #internal Ops Agent
🌐 Web Dashboard Any Agent
17 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Sub-Agents & Background Tasks

  • Main agent spawns background tasks while you keep chatting
  • Coding agents (Codex, Claude Code) for building tools
  • Results auto-announced back to the requester
  • Each sub-agent has its own context — no interference
You: "Audit site.com while I prep
      for my 2pm call"

Agent: ✅ Spawned audit sub-agent

...30 minutes later...

Agent: 📊 Audit complete!
  Score: 72/100. Report ready.
18 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

The Amnesia Problem

  • AI agents wake up with amnesia every session
  • Without memory, you re-explain everything every time
  • Memory is the difference between a tool and a team member
  • OpenClaw uses plain Markdown files — simple, portable, yours
Day 1: "I run an SEO agency with
  20 clients, we use ClickUp..."
Day 2: "I run an SEO agency with
  20 clients, we use ClickUp..."
Day 3: "I run an SEO agency with
  20 clients, we use ClickUp..."

🤦 Every. Single. Day.
19 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Memory Layers

📁
MEMORY.md
Long-term: goals, client roster, key decisions, lessons learned. Curated manually.
📅
memory/YYYY-MM-DD.md
Daily logs: what happened today. Auto-appended. Read today + yesterday on boot.
🔍
Vector Search
Semantic recall across all memory files. Find related notes even with different wording.
💾
Pre-Compaction Flush
Auto-saves important context before memory cleanup. Nothing gets lost.
20 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Memory Tips That Actually Matter

  • Write decisions down IMMEDIATELY — don't trust "I'll remember"
  • Keep MEMORY.md curated — it's your agent's permanent brain
  • Use event-driven logging (append-only, compile periodically)
  • Enable pre-compaction flush so context isn't lost
  • Separate runtime state from deliverables
The #1 mistake: treating memory as "nice to have."

Memory is what turns a $200/month API bill into a $200/month employee.
21 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

OpenClaw Isn't Another AI Chatbot

It's your agency's operating system.

🔒
Self-hosted = you own your data
Start small → scale to a full AI team
🏆
Your prompts = competitive advantage
22 / 22
INTRO USE CASES SETUP MULTI-AGENT MEMORY

Resources

Everything you need to get started today.

📖
docs.openclaw.ai
Full documentation
💻
github.com/openclaw/openclaw
Source code (MIT license)
💬
discord.com/invite/clawd
Community Discord
🧩
clawhub.com
Skills marketplace
Roger Yin — @daluoseo — HashSEO Thank you 🦞