Which Gateway Is Handling Your WhatsApp Messages?
Run `openclaw channels login --channel whatsapp` on the serving Gateway, scan the QR code from WhatsApp Linked Devices, then verify routing and sender policy with a redacted message trace.
Follow one message across the three handoffs
Run `openclaw channels login --channel whatsapp` in the environment that runs your serving Gateway. The command offers to install the WhatsApp plugin if it is missing. Scan the displayed QR code from the phone’s Linked Devices screen, then send a harmless DM with a unique marker from an authorized test sender. Record when the Gateway received it, when the agent started, and when the sender saw the reply. If reception takes one second and the agent starts at second two but the reply arrives at second 19, the receive and dispatch path accounts for only two seconds. Inspect agent completion and outbound delivery next. If the serving Gateway never records reception, check which instance owns the linked session before changing model settings. OpenClaw’s documented path uses WhatsApp Web through Baileys. A Twilio configuration would exercise a different integration.
WhatsApp - OpenClawVerify which configuration governs the running service
When an interactive CLI and a service run under different users or launch environments, compare their effective configuration paths before editing sender policy. The documented default is `~/.openclaw/openclaw.json`, so identical command text can resolve to different files under different home directories. Record the service’s effective path and the destination your editor or config command will write. For a custom `OPENCLAW_CONFIG_PATH`, use the real file for writes. A snapshot read can follow a config-file symlink, so a successful read alone leaves the write path unverified. After applying the change, send separate marked DMs from an approved test identity and an unapproved one. Under the default pairing policy, record the second message’s pairing challenge and look for any corresponding agent start. That observation distinguishes an access-policy failure from a successful channel response.
Configuration - OpenClawChoose the number and Gateway
OpenClaw’s WhatsApp channel uses WhatsApp Web, and the Gateway owns the linked session. Decide which running Gateway will serve the channel before you link your phone. Open a shell on that host and prove the agent can answer through an interface you already use. That gives you a baseline for the model and agent before you introduce WhatsApp.
Use a personal number for a private pilot only if you are comfortable linking it to that Gateway and can test with controlled contacts. For a team inbox or customer pilot, use a dedicated number you control and recruit a consenting tester. Start with a direct message and keep groups and external action tools outside the first test.
WhatsApp - OpenClawLink the phone from the serving host
On the Gateway host, run openclaw channels add --channel whatsapp. Accept the prompt to install the WhatsApp plugin if it appears. Then run openclaw channels login --channel whatsapp to start linking. On the phone for the account you intend to use, open WhatsApp’s Linked Devices screen and scan the displayed QR code with its device-linking scanner.
Check that WhatsApp lists the newly linked device and that the login flow reports success. If the code expires, start a fresh login and scan the new code. If the scanner cannot read it, enlarge the terminal so the whole code is visible and check that the phone is using the Linked Devices scanner. Keep the QR private.
WhatsApp - OpenClawTest access with two senders
OpenClaw’s Gateway configuration controls who can message the bot. Check the installed version’s WhatsApp configuration for the direct-message policy and set it to admit only your test sender before inviting anyone else. Check that you are editing the serving Gateway’s active configuration. Use a harmless prompt that calls no tools so the test isolates access and reply delivery.
For an illustrative pilot, a team links a dedicated number and allows Priya’s test phone. Priya sends Reply with WA-42 and should receive one reply containing WA-42. A second consenting tester sends Reply with WA-43 from an unapproved number. The second prompt should produce no agent reply or task execution. If it does, close access and retest before adding users.
Configuration - OpenClawTrace a linked device that stays silent
If the phone shows a linked device and the permitted sender gets silence, repeat your baseline prompt through the working interface. If that fails too, investigate the agent or model before relinking. If the baseline passes, send a fresh WhatsApp test and check the serving Gateway for the inbound event. An absent event points you back to the session and Gateway. A received event with no agent run points you toward sender policy or routing.
Run openclaw security audit after changing access settings and review the findings before widening the pilot. Restart the Gateway during a test window and resend the permitted prompt. Require the same reply and access behavior after restart before relying on the connection unattended.
Security - OpenClaw