You want an AI assistant that actually runs around the clock — not one that dies when your terminal closes or your laptop sleeps. This guide takes you from zero to a fully operational 24/7 AI agent using OpenClaw on a cloud server. No prior experience with AI deployment required. Just follow the steps.
By the end of this guide, you'll have:
Total time: 30–45 minutes for a first-time setup.
Running OpenClaw on your personal machine is a bad idea for three reasons: uptime (your laptop sleeps), security (the agent has file and network access), and reliability (your home IP changes). A Tencent Cloud Lighthouse instance solves all three.
Recommended specs: 2-core CPU, 4 GB RAM. This handles single-store workloads comfortably. For multi-channel or high-volume setups, go with 4-core.
Region selection tip: If you're integrating with Discord or Telegram, choose an overseas region for better network connectivity. If you're connecting to domestic platforms, choose a region in mainland China.
Once the instance is provisioned (takes about 30 seconds), access it via the Lighthouse console:
The OpenClaw template comes with everything pre-installed. Start the configuration:
# Launch the onboarding wizard
clawdbot onboard
# The wizard will walk you through:
# 1. Selecting your LLM provider
# 2. Entering your API key
# 3. Configuring your agent's persona
# 4. Binding a messaging channel
Security reminder: When entering your API key, the wizard stores it in a secure local config. Never hard-code API keys in shell scripts, Dockerfiles, or any file that might be shared or committed to version control. If you need to set it as an environment variable:
export OPENCLAW_API_KEY="sk-your-key-here"
You can also configure the API key through the visual management panel in the Lighthouse console: navigate to your instance > Application Management > Models > API Key > paste and apply.
OpenClaw supports multiple LLM providers:
For custom model configuration, see the Custom Model Tutorial.
This is where your agent comes alive. Pick your platform and follow the corresponding guide:
| Platform | Guide | Best For |
|---|---|---|
| Telegram | Setup | Global reach, developer-friendly |
| Discord | Setup | Community management, gaming |
| Setup | Business communication, e-commerce |
Each guide covers bot creation, token generation, and webhook configuration. Expect about 10 minutes per channel.
Skills are what transform OpenClaw from a chatbot into an agent. From the chat interface:
# Install a customer-service skill
# "Please install a skill for me using Clawhub; its name is ecommerce-cs-assistant."
# Install the browser automation skill (if not already present)
# "Please install a skill for me using Clawhub; its name is agent-browser."
# Verify your skill stack
# "Check which skills you have currently installed."
For the complete skill management guide: Installing OpenClaw Skills.
This is the step most tutorials skip, and it's the one that makes or breaks your 24/7 setup. Without daemon mode, your agent dies when you close the terminal.
# Enable user-level systemd services
loginctl enable-linger $(whoami)
export XDG_RUNTIME_DIR=/run/user/$(id -u)
# Install and start the daemon
clawdbot daemon install
clawdbot daemon start
# Verify it's running
clawdbot daemon status
# Expected output: "running"
Once you see "running," you can safely close the terminal. Your agent will continue operating in the background and auto-restart on server reboot.
Send a test message through your connected channel:
If any test fails, check the logs:
# View recent logs
clawdbot daemon logs
| Symptom | Likely Cause | Fix |
|---|---|---|
| Agent doesn't respond | Daemon not running | clawdbot daemon start |
| "Invalid API key" error | Wrong key or expired | Re-enter via clawdbot onboard or the visual panel |
| Slow responses | System prompt too long | Trim to under 150 tokens |
| Channel not connected | Webhook URL misconfigured | Verify the URL matches your server's public IP |
You now have a fully operational 24/7 AI assistant. From here, you can:
If you haven't deployed yet, start now:
Thirty minutes from now, you'll have an AI assistant that never sleeps. That's a pretty good ROI on half an hour.