Customer Support Bot

An intelligent AI support agent that handles customer inquiries across DMs and group channels with knowledge base integration and ticket escalation.

Overview

The Customer Support Bot is a versatile AI assistant designed to provide 24/7 customer support across multiple messaging platforms. It combines a searchable knowledge base, conversation memory, and intelligent escalation to deliver professional, consistent support experiences.

Unlike simple chatbots, this agent:

• Learns from interactions — Builds institutional knowledge over time
• Handles multiple channels — DMs for private issues, groups for public Q&A
• Escalates intelligently — Knows when to loop in human support
• Maintains context — Remembers customer history and preferences
• Provides accurate answers — Searches knowledge base before responding

Key Features

🎯 Multi-Channel Support

• DM sessions: Private 1:1 support with full conversation history
• Group channels: Public Q&A in Discord/Telegram servers
• Pairing mode: Approve DM users before engagement (prevents spam)

📚 Knowledge Base Integration

• Pre-seeded with your product/service documentation
• Semantic search over knowledge articles
• Updates knowledge base as new issues are resolved
• Cites sources in responses

🎫 Smart Escalation

• Recognizes when issues require human intervention
• Creates structured escalation reports
• Tracks unresolved issues in memory/escalations.json
• Notifies operator of urgent cases

💬 Natural Conversations

• Warm, professional tone (not robotic)
• Asks clarifying questions when needed
• Provides step-by-step troubleshooting
• Follows up on resolved issues

🔒 Privacy & Security

• Sandboxed DM/group sessions (no shell access)
• Customer data stored separately per user
• Option to anonymize transcripts
• Secure credential handling

Quick Start

1. Install via Hatchery

hatchery run @openclaw/customer-support-bot my-support-bot

2. Configure Environment

Copy .env.example to .env and configure:

# Required: Anthropic API key
ANTHROPIC_API_KEY=sk-ant-xxx

# Required: Messaging platform credentials
# Choose one or more:
TELEGRAM_BOT_TOKEN=123456:ABC-DEF...    # From @BotFather
DISCORD_BOT_TOKEN=MTk4NjIyNDgzNDc...   # From Discord Developer Portal
SLACK_BOT_TOKEN=xoxb-...                # From Slack App settings

# Optional: Business hours (for escalation urgency)
BUSINESS_HOURS_START=09:00
BUSINESS_HOURS_END=17:00
BUSINESS_TIMEZONE=America/New_York

# Optional: Escalation webhook
ESCALATION_WEBHOOK_URL=https://hooks.slack.com/services/xxx

3. Customize Knowledge Base

Edit MEMORY.md to include your product's knowledge base:

## Product Knowledge

### How to reset password
1. Visit https://yourapp.com/reset
2. Enter your email address
3. Check email for reset link (valid 1 hour)
4. Create new password (min 8 characters)

### Common Error Codes
- **E101**: Database connection failed → Check internet connection
- **E202**: Invalid API key → Regenerate key in dashboard
...

4. Start the Agent

cd ~/.hatchery/instances/my-support-bot
docker-compose up -d

5. Approve DM Users (Pairing Mode)

When a new user messages the bot, you'll receive a pairing request in your main session. Reply:

Approve DM access for @username

Configuration

Session Policies

DM Policy (default: pairing):

• pairing: Require operator approval for each new DM user (recommended)
• open: Accept DMs from anyone (use for public bots)

Group Policy:

• Responds only when @mentioned (prevents spam)
• Read-only access (cannot modify knowledge base from groups)
• No access to customer-specific data

Escalation Triggers

Edit AGENTS.md to customize when the bot escalates:

## When to Escalate
1. **Immediate**: Security issues, data breaches, payment failures
2. **Urgent**: Service outages affecting multiple users
3. **Normal**: Feature requests, complex technical issues
4. **Low**: General feedback, questions for product team

Response Templates

Add common response templates to memory/templates.md:

### Refund Request Template
Thank you for contacting support. I've escalated your refund request to our billing team.

Reference number: {TICKET_ID}
Expected response: 1-2 business days

Is there anything else I can help with?

Usage Examples

Example 1: Simple Q&A (Group Chat)

User: @SupportBot How do I export my data?

Bot: You can export your data from Settings → Privacy → Export Data.

The export includes:
• All your posts and comments
• Profile information
• Account settings

The export is delivered via email as a .zip file within 24 hours.

Need help with anything specific in the export process?

Example 2: Troubleshooting (DM)

User: I'm getting error E202 when I try to connect

Bot: Error E202 means your API key is invalid or expired. Let's fix that:

1. Go to https://yourapp.com/dashboard/api
2. Click "Regenerate API Key"
3. Copy the new key
4. Update it in your app settings

⚠️ Important: The old key will stop working immediately after regeneration.

Did that resolve the error?

User: Yes, working now!

Bot: Great! I've noted this solution in case other users encounter the same issue.

Example 3: Escalation (DM)

User: I was charged twice for my subscription

Bot: I'm sorry to hear that! Double-billing is a serious issue that requires immediate attention from our billing team.

I've created an escalation ticket:

**Ticket #1523**
Issue: Double charge on subscription
Priority: Urgent
Expected response: Within 4 business hours

Our billing team has been notified and will reach out to you directly at the email associated with your account.

In the meantime, can you tell me:
1. Which payment method was charged?
2. What were the charge amounts and dates?

This will help expedite the resolution.

Example 4: Knowledge Base Update (Main Session)

Operator: Add to knowledge base: New feature launched - dark mode toggle in Settings → Appearance

Bot: ✅ Knowledge base updated.

Added to MEMORY.md under "New Features":
- Dark mode toggle now available in Settings → Appearance
- Recorded: 2024-01-15

This will be included in future support responses.

Architecture

How It Works

┌─────────────────┐
│  User Message   │
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│ Memory Search   │ ← Searches knowledge base
│ (Semantic)      │   for relevant articles
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│ Claude Sonnet   │ ← Generates response
│ (LLM)           │   with context
└────────┬────────┘
         │
         ▼
    ┌────┴─────┐
    │ Resolved? │
    └────┬─────┘
         │
    ┌────┴────┐
    │   Yes   │   No
    │         │
    ▼         ▼
 Response   Escalate
            to Human

Memory Structure

memory/
├── escalations.json      # Active and resolved escalations
├── customer-history/     # Per-customer conversation logs
│   ├── user-123.md
│   └── user-456.md
├── templates.md          # Response templates
└── knowledge-updates.md  # Running log of KB additions

Session Types

• Main: Full operator control, can edit knowledge base, approve pairings
• DM: Sandboxed support conversations, can search KB and create escalations
• Group: Public Q&A only, read-only KB access

Troubleshooting

Bot doesn't respond in DMs

Cause: Pairing mode requires approval

Solution: Check your main session for pairing requests. Approve with:

Approve DM access for @username

Responses are slow

Cause: Knowledge base semantic search takes time

Solution:

• Pre-index knowledge base (set memory.indexTranscripts: true)
• Use Claude Haiku for faster responses (edit hatchery.yaml model)

Bot gives outdated information

Cause: Knowledge base not updated

Solution: Update MEMORY.md with current information:

Update knowledge base: [new information]

Escalations not triggering

Cause: Escalation keywords not recognized

Solution: Check AGENTS.md escalation rules. Add keywords like:

• "charged twice", "can't login", "data loss", "security issue"

Advanced Customization

Multi-Language Support

Add language detection and responses in AGENTS.md:

## Language Support
- Detect user language from first message
- Respond in same language when possible
- Escalate to human for languages not supported (English, Spanish, French)

Integration with Ticketing Systems

Add webhook integration in skills/create-ticket/:

#!/bin/bash
# skills/create-ticket/create.sh
curl -X POST "$ZENDESK_API_URL/tickets.json" \
  -H "Authorization: Bearer $ZENDESK_TOKEN" \
  -d '{"ticket": {"subject": "'$1'", "comment": {"body": "'$2'"}}}'

Custom Analytics

Track support metrics in memory/analytics.json:

{
  "totalConversations": 247,
  "resolvedWithoutEscalation": 198,
  "averageResponseTime": "2.3 minutes",
  "commonIssues": [
    {"issue": "password reset", "count": 45},
    {"issue": "API key issues", "count": 32}
  ]
}

Best Practices

1. Keep knowledge base current — Review and update weekly
2. Monitor escalations — Review escalated tickets to improve KB
3. Train on real conversations — Add successful resolutions to MEMORY.md
4. Set clear boundaries — Define what the bot can/cannot do
5. Human handoff — Make it easy for users to reach humans when needed

Contributing

Found a bug or want to improve this template?

• Report issues: https://github.com/openclaw/templates/issues
• Submit improvements: https://github.com/openclaw/templates/pulls

License

MIT License - Free to use and modify for your support needs.