Overview
Your support inbox is full. Most questions are the same 10 things. Deploy an OpenClaw agent on AgentClaw to handle the repeatable work automatically โ freeing your team for the cases that actually need human judgment.
What You'll Build
An AI customer service agent that:
- Resolves common issues instantly (order status, account questions, how-tos)
- Collects structured info (order number, account email) before escalating
- Routes complex issues to the right team member via DM
- Provides consistent, on-brand responses 24/7
Architecture
Customer โ Telegram/Discord
โ
OpenClaw Agent (on AgentClaw)
โ
Instant answer OR structured handoff
โ
Your support team
Prerequisites
- AgentClaw account
- Telegram Bot Token or Discord Bot Token
- Anthropic or OpenAI API key
- A written FAQ / knowledge base (even a Google Doc works)
Setup Steps
1. Document Your FAQ
Before building, write down the top 10-20 questions your customers ask. This becomes your agent's knowledge base.
2. Create Your Instance
Sign up at agentclaw.app and create a new agent instance for your business.
3. Write a Detailed System Prompt
The quality of your system prompt determines the quality of your bot:
SYSTEM_PROMPT="You are the customer service assistant for [Company Name].
KNOWLEDGE BASE:
- Shipping: Orders ship within 2 business days. Tracking via email.
- Returns: 30-day return policy. Email returns@company.com.
- [Add your specific FAQs here]
ESCALATION:
If a customer has a billing dispute, account suspension, or complex issue:
1. Apologize for the inconvenience
2. Ask for their order number and email
3. Say: 'I'm connecting you with our support team. Expected response: 4 hours.'
Always be empathetic and solution-focused."
4. Connect Your Channel
Go to Integrations โ Telegram (or configure Discord via env vars) and connect your support bot.
5. Set Up Notifications (Optional)
Add your email to receive alerts when the agent escalates a conversation:
ESCALATION_EMAIL=support@yourcompany.com
6. Start and Test
Click Start in your dashboard, then send some test messages through Telegram to verify responses.
Measuring Success
Use the Logs tab to review conversations. Track:
- % of questions resolved without escalation
- Response time (should be < 3 seconds)
- Customer satisfaction (ask at end of conversation)
Tips
- Iterate on the system prompt โ check logs weekly and add missed cases
- Be honest about AI โ OpenClaw can be configured to disclose it's AI when asked
- Restart to update โ changes to
SYSTEM_PROMPTenv var require an instance restart