Overview
Replace your "we'll get back to you" auto-reply with an AI agent that actually answers questions. This use case walks you through deploying an OpenClaw agent on AgentClaw that handles your Telegram support channel around the clock.
What You'll Build
A Telegram bot that:
- Answers common questions from a knowledge base you define
- Handles multiple conversations simultaneously
- Escalates to a human when needed (via DM forwarding)
- Runs 24/7 without you keeping a server alive
Prerequisites
- An AgentClaw account
- A Telegram Bot Token (from @BotFather)
- An Anthropic or OpenAI API key
Setup Steps
1. Create Your AgentClaw Instance
Sign up at agentclaw.app/signup and create a new agent instance. Choose a subdomain that matches your product.
2. Configure Your AI Model
In your instance dashboard, go to Integrations → AI Model and add your Anthropic or OpenAI API key. Claude Haiku works great for support bots — fast and cost-effective.
3. Connect Telegram
Navigate to Integrations → Telegram and paste your bot token from BotFather. Your bot will immediately be connected to your OpenClaw instance.
4. Customize Your Agent
Use environment variables to set your agent's persona and knowledge:
AGENT_NAME=SupportBot
SYSTEM_PROMPT="You are a helpful support agent for [Your Product]. Answer questions about..."
5. Start Your Instance
Click Start in the Overview tab. Within 60 seconds, your bot is live and responding on Telegram.
Tips
- Keep system prompts concise — Claude and GPT work best with clear, specific instructions
- Test with
/start— OpenClaw handles the Telegram webhook automatically - Use the Logs tab to see exactly what your bot is processing in real time
Cost Estimate
- AgentClaw hosting: $9.99/month
- Claude Haiku API: ~$0.001 per message (very affordable at scale)
- Total for 10,000 messages/month: ~$20