.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
.svg)
You installed OpenClaw. The onboarding wizard finished. Now comes the part that decides if you keep going or abandon it: your first message.
This is not a feel-good milestone. It is an end-to-end test. If your first message works, you have proven that:
- The Gateway is running and reachable
- Your model provider is configured correctly
- Your chosen channel is connected
- OpenClaw can process a message and reply back through the same channel
If you are still getting oriented, start with what OpenClaw is and how it works. If you want to compare model providers before you commit, use Claude vs OpenAI model choice for OpenClaw. If your plan is to run OpenClaw across multiple chat apps, keep OpenClaw multi-channel setup bookmarked.
Pick the easiest path for your first message
There are three common ways to send your first message. Don’t overthink it. Pick the one with the least friction for your setup.
- WhatsApp if you want mobile-first access and you are fine with WhatsApp Web pairing
- Telegram if you want the cleanest official bot API path
- SSH port forwarding to the local web UI if you want a secure way to confirm the Gateway UI without exposing ports
Email integrations can be great later, but they are a terrible “first message” path. Get one channel working first.
Path 1: WhatsApp first message
WhatsApp is popular because it feels natural. You already know the interface, so OpenClaw feels less like software and more like a contact.
To link WhatsApp, you typically run the channel login command, scan the QR code, then message the linked account. The exact prompts can vary by release, but the concept stays the same: it is WhatsApp Web style device pairing.
After linking, open WhatsApp and start a chat with the number you linked for OpenClaw. For quick testing, many people message themselves, then let OpenClaw intercept that conversation.
Send something simple:
Hello
Wait a few seconds. Your first response can be slower than later replies because the session is cold and the model call is happening for the first time.
If WhatsApp does not reply
Check these in order, because this is where most first-time setups fail:
- QR code expired. Generate a new one and scan immediately.
- Gateway is not running. Confirm it is active before trying to message again.
- Model provider is not configured. If your API key is missing or invalid, OpenClaw may receive the message but fail to generate a reply.
- Allowlist rules block you. If you configured an allowlist and used the wrong number, OpenClaw will ignore you.
One real security note for WhatsApp: the ecosystem has seen credential-stealing lookalike packages in the past. Be strict about installing official dependencies only. This matters more once you start using OpenClaw daily.
Path 2: Telegram first message
Telegram is the fastest “it just works” option because it is built for bots.
Create a Telegram bot via @BotFather and copy the token. Treat that token like a password. Add it to your OpenClaw config. Restart the Gateway if needed.
Many Telegram setups also use a pairing step. The first time you message the bot, it replies with a pairing code. You approve that code on the server. Only after approval does OpenClaw start processing your actual commands. This prevents random people from discovering your bot username and talking to it.
Once paired, send the same simple test message:
Hello
If Telegram does not reply
- Bot token typo. Copy it fresh from BotFather.
- Pairing not approved. If pairing is enabled, you must approve the code first.
- Gateway not running. Verify the process is active.
Telegram is also a clean stepping stone into team workflows. You can later add OpenClaw to a group and configure it to respond only when mentioned so it does not spam the chat.
Path 3: Confirm the local web UI over SSH port forwarding
If you are running OpenClaw on a VPS, the safest way to check the web UI is to keep the Gateway bound to 127.0.0.1 and forward the port over SSH.
We have a step-by-step guide for that here: OpenClaw quickstart onboarding over SSH.
The idea is simple: you create a tunnel from your laptop to the VPS, mapping local port 18789 to 127.0.0.1:18789 on the VPS. Then you open your browser to localhost on your own machine. The Gateway stays private and you still get a UI for debugging and verification.
If the UI does not load, do not expose port 18789 publicly. Fix the tunnel, confirm OpenClaw is listening on localhost, then try again.
What your first message should be
Keep it boring. You are testing plumbing, not intelligence.
Good first messages:
- Hello
- Hi OpenClaw
- Are you running
If you get a reply, you have already won. Now you can test one controlled “action” prompt.
Your second message should test one safe tool
After you get a reply, test tool execution with a low-risk request.
Example:
List the files in my home directory
If OpenClaw returns a directory listing, you have confirmed that the agent can execute a basic command and return grounded output.
If you are hosting on a VPS, keep the safety mindset. Restrict tools. Use allowlists. Avoid giving OpenClaw full shell access until you have a real reason. If you want a hardened setup, read how to host OpenClaw securely on a VPS.
How to verify your first message worked
You want to see evidence in at least two places, ideally three:
- Your chat app shows OpenClaw’s reply
- Logs show the inbound message and outbound response
- The local web UI reflects the session activity
If you only see the message in the chat app but nothing in logs, you may be talking to the wrong bot identity. If you see logs but no reply, you likely have a model provider error or a channel send failure.
Common first-message mistakes
- Gateway not started after onboarding
- Telegram pairing not approved
- API key missing, invalid, or rate-limited
- WhatsApp QR code expired before scanning
- Allowlist rules blocking your own account
- Expecting instant replies on a cold session
A quick warning about model authorization
If you are using Claude, do not try to route automation through consumer subscription credits in a way that violates provider terms. Use proper API access for agent automation. It is boring advice, but it saves headaches.
If you are unsure what model to run for your use case, use the OpenClaw Claude vs OpenAI model guide and pick a default you can live with for a week. You can always change it later.
