Know when your OpenClaw agent stops responding
When an OpenClaw agent stops responding, the usual reason is that its Gateway isn't running or isn't connected: the computer slept, lost its network or is mid-update. Less often the agent's turn failed, for example because its AI account needs attention. OpenClaw's health commands tell you which, if you check. Octopus Prime Health Watch checks for you: it shows every Gateway as Healthy, Warning or Offline and alerts Owners and Admins in the app when one goes offline.
Why an agent goes quiet
- The Gateway is down. The OpenClaw process stopped, or the machine it runs on is asleep or off.
- The Gateway is up but not connected. The network dropped, or a channel's connection didn't come back.
- An update changed something. OpenClaw updates often, and an update can break a plugin or a channel until it's fixed.
- The turn failed. The model provider refused the request, the account ran out of credit, or the agent's setup has an error.
- It's still working. A long task can take minutes before the first word comes back.
Check it yourself with OpenClaw
On the Gateway host, OpenClaw's health checks answer the first question quickly:
openclaw status: a local summary of whether the Gateway is reachable, its channels and recent activity.openclaw health: the running Gateway's own health snapshot (--jsonfor scripts)./statussent as a chat command: a status reply without running the agent.openclaw logs --follow: what the Gateway is doing right now.
For alerts, OpenClaw exposes HTTP probes (/healthz for "the process is live", /readyz for "ready to take work").
An outside monitoring service can poll them, but only if it can reach a Gateway address, which usually means exposing one.
Trade-off: these work when you're at the machine or have set up monitoring. Nobody is told automatically, and the people who use the agent can't tell "down" from "thinking".
Let Octopus Prime watch it
With Octopus Prime, Health Watch watches every connected Gateway for you, and the people who use an agent can tell "down" from "thinking". Nothing on your network is exposed: you open no ports and give your computer no public address.
| What happens | What you see |
|---|---|
| The Gateway's computer sleeps, loses its network or stops | Health Watch shows the Gateway as Warning, then Offline |
| It stays offline | Owners and Admins get an alert in the app, one per outage, not a flood |
| An agent stops answering messages | Owners and Admins get an alert naming the agent |
| Your message gets no answer | The app says the agent didn't respond, in words that fit the cause, and you can tap Retry |
| The Gateway comes back | Waiting messages arrive, nothing is sent twice, and the Gateway shows as Healthy again |
When an agent is stuck but its Gateway is still connected, Restart Gateway restarts OpenClaw from the app. It's off until the Gateway's owner turns it on for that computer, and the app shows the step. By default only the person who connected the Gateway can use it. And when an OpenClaw release breaks connections, affected organizations get a short notice, then another when it's fixed.
Health Watch alerts are in the app only, not push or e-mail. It watches; it doesn't start agents or repair Gateways, so a machine that's switched off still needs a person.
A short checklist
- Is the Gateway's computer awake and online? Check Health Watch, or run
openclaw statuson it. - Did OpenClaw just update? Restart the Gateway so the update loads.
- Did the agent post a notice about its AI account or setup? Fix that on the Gateway, then Retry.
- Still stuck? The troubleshooting reference lists symptom-first checks your own agent can follow.
