Set Up OpenClaw in 5 Minutes
This guide walks you through every step of configuring OpenClaw — from picking an AI model to having your bot live on Telegram or WhatsApp. With ClawMates, the whole process takes under 5 minutes and requires zero coding. If you prefer to self-host, the manual Docker instructions are included too.
In This Guide
What You'll Need
OpenClaw setup is deliberately lightweight. ClawMates removes all the infrastructure requirements so you don't need a server, Docker knowledge, or an API key — everything is included in the subscription. Here's what you actually need:
A Telegram or WhatsApp Account
Any personal Telegram account works out of the box. For WhatsApp, you need a phone number that's already registered on WhatsApp — personal accounts work via QR pairing, no business account required.
A ClawMates Account
Sign up at clawmates.net/setup — there's a 7-day free trial. No credit card needed to start. All AI API costs and server hosting are included in the $29.99/mo plan.
No Server Required
ClawMates handles all hosting on Fly.io with automatic redundancy. You never touch a terminal or SSH session.
No AI API Key Required
ClawMates provisions AI access for you. You choose the model; we handle the API billing. No OpenAI or Anthropic account needed.
Self-hosting? If you prefer to run OpenClaw on your own infrastructure, you'll additionally need: Docker, a Linux/Mac machine or VPS, and API keys for your chosen AI provider. See the OpenClaw installation guide for the full walkthrough.
Choose Your AI Model
OpenClaw is model-agnostic — it works as a bridge between your messaging app and whichever AI provider you prefer. Your choice of model affects response quality, speed, and the kinds of tasks your bot handles best.
Claude (Anthropic)
RecommendedClaude excels at nuanced reasoning, long-form writing, code review, and tasks requiring careful instruction-following. It has a large context window (up to 200K tokens) which means it remembers significantly more conversation history. Claude is less likely to hallucinate on factual questions and handles ambiguous instructions gracefully. Best for: personal assistants, writing aids, research companions, and customer support bots.
Available models: Claude 3.5 Sonnet, Claude 3.5 Haiku, Claude 3 Opus
GPT-4 (OpenAI)
GPT-4 is the most well-known model and arguably the most versatile. It handles a broad range of tasks from coding to creative writing and has excellent tool-use capabilities. GPT-4o is faster and cheaper than GPT-4 while maintaining most of the quality. Best for: general-purpose assistants, coding help, brainstorming, and tasks where you want the widest compatibility with existing prompts and templates.
Available models: GPT-4o, GPT-4o mini, GPT-4 Turbo
Gemini (Google)
Gemini is Google's frontier model with strong multimodal capabilities — it handles text, images, audio, and code within the same conversation. Gemini 1.5 Pro has a 1M token context window, making it excellent for analyzing long documents. Best for: bots that need to process images or PDFs sent through WhatsApp, or assistants working with very long context.
Available models: Gemini 1.5 Pro, Gemini 1.5 Flash, Gemini 2.0 Flash
Quick Decision Guide
- → Personal assistant / daily helper → Claude 3.5 Sonnet
- → Coding assistant → GPT-4o or Claude 3.5 Sonnet
- → Image / document analysis → Gemini 1.5 Pro
- → Budget-conscious / high volume → GPT-4o mini or Claude 3.5 Haiku
- → Long conversations / large documents → Claude 3.5 Sonnet or Gemini 1.5 Pro
You can always change your model later without redeploying. The ClawMates dashboard lets you switch models from a dropdown — no configuration file editing required.
Get Your Bot Token
A bot token is a credential that lets OpenClaw communicate with your messaging platform on your behalf. The process differs slightly between Telegram and WhatsApp.
Telegram: BotFather
Telegram bots are created through BotFather, an official Telegram bot that acts as a bot registry. The entire process takes about 60 seconds:
- 1
Open Telegram and search for
@BotFather(the blue verified checkmark is important). - 2
Start the conversation by clicking Start or typing
/start. - 3
Send
/newbotto create a new bot. - 4
BotFather asks for a display name (e.g. “My AI Assistant”) and a username ending in
bot(e.g.myai_assistant_bot). - 5
BotFather sends back a token that looks like:
7284019563:AAGt2kZ9_fJwEO4nPZF8QgLxYrMvHcI3bDsCopy this token — you'll paste it into the ClawMates dashboard. Treat it like a password; anyone with this token can control your bot.
Tip: You can also customize your bot's profile picture and description via BotFather using /setuserpic and /setdescription.
WhatsApp: Meta Business Setup
WhatsApp integration with OpenClaw works differently from Telegram. Rather than a bot token, WhatsApp uses QR code-based account linking — similar to WhatsApp Web. This means your existing personal WhatsApp number becomes the bot's number. No Meta Business account is required for the personal (QR pairing) method.
If you want a dedicated business WhatsApp number (separate from your personal phone), you can use the Meta Business API method:
- 1
Go to
developers.facebook.comand create or log in with a Meta account. - 2
Create a new app, select Business type, and add the WhatsApp product.
- 3
Under WhatsApp > API Setup, copy your Phone Number ID and generate a Permanent Access Token via System Users.
- 4
Paste both values into the ClawMates WhatsApp configuration fields.
For most users, the QR pairing method (Step 3 below) is simpler and requires no Meta developer account. The Meta Business API is only needed if you want a dedicated business phone number.
Connect WhatsApp
WhatsApp uses a QR code pairing flow — the same technology behind WhatsApp Web. Your phone must remain connected to the internet for the bot to stay active (your number is the bot's identity). Here's how to complete the pairing:
- 1
Open the ClawMates Dashboard
Navigate to /setup after logging in. Under the Channels section, select WhatsApp and click Connect WhatsApp.
- 2
A QR Code Appears
The dashboard displays a QR code that expires in 60 seconds. The QR code is generated fresh each time you initiate pairing. If it expires before you scan it, click Refresh QR.
- 3
Scan with Your Phone
On your phone, open WhatsApp and go to Settings > Linked Devices > Link a Device. Point your camera at the QR code in the dashboard. WhatsApp will confirm the link within a few seconds.
- 4
Pairing Confirmed
The dashboard status changes to Connected. Your WhatsApp number is now linked to the OpenClaw gateway. Messages you receive on WhatsApp will be routed through the AI model you selected.
Phone Requirements
- → Your phone must stay connected to the internet (Wi-Fi or data) at all times.
- → WhatsApp on your phone must remain installed and active.
- → The linked device session lasts until you manually unlink or the session expires (~30 days of inactivity).
- → OpenClaw uses your number as the bot identity — the same number cannot be used on two separate WhatsApp Web sessions simultaneously.
Note: OpenClaw's WhatsApp integration uses the unofficial WhatsApp Web protocol (the same as WhatsApp Web in your browser). This is the most accessible option for personal use. For production business deployments, consider the Meta Business API method described in Step 2.
Configure Personality
The system prompt is the most important configuration in OpenClaw — it defines who your assistant is, what it knows, how it speaks, and what it should or shouldn't do. Getting this right dramatically improves the usefulness of your bot.
Writing an Effective System Prompt
A system prompt is a set of instructions given to the AI before any user message. Think of it as a job description for your assistant. Here are a few example prompts to get you started:
General Personal Assistant
You are a helpful personal assistant named Alex.
You help with scheduling, research, writing, and daily tasks.
Keep responses concise — 3 sentences or fewer unless asked for detail.
When you don't know something, say so rather than guessing.Coding Assistant
You are a senior software engineer specializing in TypeScript,
Python, and React. Provide working code with brief explanations.
Always include error handling. Prefer simple solutions over clever ones.
If a question is ambiguous, ask for clarification before answering.Research Companion
You are a research assistant with expertise in science,
history, and technology. Provide accurate, well-sourced information.
Cite specific facts and note when information may be outdated.
Never speculate or present opinions as facts.Temperature Setting
Temperature controls how creative vs. deterministic your assistant's responses are. It ranges from 0.0 to 1.0 (some models support up to 2.0):
Best for: factual Q&A, coding, structured data extraction
Best for: general assistants, conversation, summaries
Best for: brainstorming, storytelling, idea generation
Bot Name and Avatar
For Telegram bots, you can set a display name and profile picture through BotFather (covered in Step 2). For WhatsApp, the bot appears under your phone number's contact name — you can change this in your contacts app on recipients' phones. Consider setting the bot's name in the system prompt as well, so it introduces itself consistently.
Deploy
With your model chosen, token ready, and personality configured, the final step is deployment. ClawMates handles this in one click — but if you're self-hosting, the manual Docker command is included below.
Deploy with ClawMates (Recommended)
- 1
In the ClawMates dashboard, review your configuration summary — model, channels, and system prompt.
- 2
Click Deploy Bot. ClawMates provisions a Fly.io machine, injects your configuration, and starts the OpenClaw gateway.
- 3
The status indicator changes from Deploying to Live within 60–90 seconds. OpenClaw's gateway takes about 5 minutes to fully initialize on first boot (WASM compilation) — your bot will start responding after that.
- 4
Test by sending your Telegram bot a message (find it via its username) or messaging the linked WhatsApp number.
Deploy Manually with Docker
If you prefer to self-host, here's the minimal Docker command to run OpenClaw with a Telegram bot token and Claude:
docker run -d \
--name openclaw \
--memory 2048m \
-e TELEGRAM_BOT_TOKEN=your_token_here \
-e ANTHROPIC_API_KEY=your_api_key_here \
-e NODE_OPTIONS=--max-old-space-size=1536 \
-p 18789:18789 \
alpine/openclaw:latest \
node /app/openclaw.mjs gatewayNote: 2048 MB RAM is required. The gateway takes ~5 minutes to start (WASM compilation). Check docker logs openclaw to monitor startup. For a complete guide including config file setup, see the OpenClaw installation guide.
Verifying Your Deployment
After deployment, verify everything is working:
# For self-hosted: check logs for successful gateway init
docker logs openclaw --tail 50
# You should see lines like:
# [gateway] OpenClaw gateway starting...
# [gateway] WASM sandbox compiled (took 4m 32s)
# [gateway] Telegram polling started
# [gateway] Ready on port 18789Troubleshooting Common Issues
Most setup problems fall into a handful of categories. Here's how to diagnose and fix the most common ones:
Bot is not responding on Telegram
This is the most common issue. Check these in order:
- 1. Make sure you sent
/startto the bot first — Telegram bots ignore messages before the initial /start. - 2. Verify the bot token is correct. Regenerate it via BotFather (
/revoke) and update it in ClawMates. - 3. Check the ClawMates dashboard — if the status is not “Live”, redeploy.
- 4. Wait 5–6 minutes after deployment. OpenClaw's WASM compilation takes time on first boot.
- 5. Check that your AI API key (if self-hosting) has available credits.
WhatsApp QR Code Expired
QR codes for WhatsApp pairing expire after 60 seconds. If you see “QR code expired” in the dashboard:
- → Click Refresh QR in the ClawMates dashboard to generate a new code.
- → Have your phone ready before clicking Refresh — scan immediately once the new QR appears.
- → Ensure your phone's WhatsApp app is updated to the latest version; older versions may not display the Linked Devices option.
WhatsApp Bot Disconnects After a Few Days
WhatsApp linked device sessions expire if the phone is offline for extended periods or if WhatsApp is reinstalled. To fix:
- → Go to the ClawMates dashboard and re-pair via a new QR code scan.
- → Ensure your phone stays connected to the internet regularly.
- → Avoid force-quitting WhatsApp on iOS; it can interrupt the session.
Invalid Bot Token Error
If you see “401 Unauthorized” or “invalid token” in the logs:
- → Double-check you copied the full token including the numeric prefix before the colon.
- → Tokens look like
1234567890:AABBCCddEEffGGhhIIjjKKllMMnnOOppQQrr - → If you previously revoked the token in BotFather, generate a new one and update ClawMates.
Bot Responds Slowly or Times Out
Slow responses are usually caused by:
- → AI provider latency: GPT-4 Turbo and Claude can take 5–15s for long responses. Use a faster model variant (GPT-4o mini, Claude Haiku) if speed is critical.
- → High temperature + long prompts: Lower temperature to 0.3–0.5 for faster, more focused responses.
- → Server cold start: If the Fly.io machine was idle, the first message may take 10–20 seconds while the container warms up. ClawMates keeps machines warm during business hours.
Bot Not Following System Prompt
If the bot ignores your persona or instructions:
- → Be more explicit. Instead of “be concise”, write “always respond in 2 sentences or fewer.”
- → Place important instructions at the beginning and end of the system prompt — models pay more attention to these positions.
- → Redeploy after changing the system prompt; changes don't take effect until the next deployment.
- → Some behaviors are safety-restricted by the AI provider and cannot be overridden by system prompts.
Next Steps After Setup
Once your bot is live and responding, here are the recommended next steps to get the most out of OpenClaw:
Advanced OpenClaw Tutorial →
Learn multi-channel routing, conversation memory, command handling, and custom integrations. Takes your bot from basic to production-grade.
What is OpenClaw? →
Deep dive into OpenClaw's architecture, how the gateway works, and why it's become the leading open-source bot framework.
OpenClaw Hosting Options →
Compare self-hosting on VPS, Fly.io, Railway, and managed ClawMates hosting. Cost breakdown and reliability comparison included.
Install OpenClaw Manually →
Full self-hosting guide with Docker Compose, environment variables, and config file reference. For power users who want full control.
Optimize Your System Prompt
The single highest-impact improvement you can make after initial setup is refining your system prompt. Spend 15–30 minutes testing different prompts and observing how the bot responds to edge cases. Prompt engineering is a skill — the OpenClaw tutorial covers advanced prompt patterns used by power users.
Add Multiple Channels
A single ClawMates deployment can run on both Telegram and WhatsApp simultaneously. Once your Telegram bot is working, revisit Step 3 to connect WhatsApp, or vice versa. Both channels use the same AI model and system prompt.
Monitor Usage
The ClawMates dashboard shows message volume, response times, and deployment health. Keep an eye on these to identify if your bot needs a higher-capacity model or a revised system prompt to reduce unnecessary verbosity.
Frequently Asked Questions
How long does it take to set up OpenClaw?
With ClawMates, setting up OpenClaw takes under 5 minutes. You choose your AI model, paste your bot token, and click deploy. Self-hosting OpenClaw manually takes 1–3 hours and requires Docker and Linux server knowledge.
Do I need coding skills to set up OpenClaw?
No. ClawMates provides a point-and-click setup with no coding required. If you choose to self-host, basic Docker and terminal knowledge is helpful but not mandatory.
Which AI model should I choose for OpenClaw?
Claude is best for nuanced reasoning, writing, and analysis. GPT-4 is the most versatile all-rounder. Gemini handles multimodal inputs (images) well. For most personal assistants, Claude 3.5 Sonnet or GPT-4o are recommended.
Why is my OpenClaw bot not responding on Telegram?
Most common causes: (1) you haven't sent /start to the bot, (2) incorrect bot token — regenerate via BotFather, (3) the bot hasn't finished initializing — wait 5 minutes after deployment, (4) your AI API key has run out of credits.
How do I set up OpenClaw for WhatsApp?
WhatsApp uses QR code pairing. In the ClawMates dashboard, go to the WhatsApp section and scan the QR code with your phone (WhatsApp > Linked Devices > Link a Device). The connection stays active as long as your phone remains online.
Can I use OpenClaw on both Telegram and WhatsApp simultaneously?
Yes. A single ClawMates deployment supports both Telegram and WhatsApp at the same time. Configure each channel separately — both will use the same AI model and system prompt.
Can I change the AI model after deployment?
Yes. In the ClawMates dashboard, update the model selection and redeploy. The redeployment takes under 2 minutes and your bot will use the new model immediately afterward.
Ready to Deploy Your AI Bot?
Skip servers, Docker, and configuration files. ClawMates handles everything — deploy OpenClaw to Telegram or WhatsApp in 5 minutes.
7-day free trial · No credit card required · Cancel anytime