An agent that is installed and forgotten is a risk, not a saving. Keeping OpenClaw standing comes down to four habits: update with a backup taken first, read the logs, save what you cannot lose, and watch memory, disk and spend. The exact update commands and the file locations are in the project’s official documentation, which follows each release; this article tells you what to do and when.
As with any unmanaged VPS, maintaining the application is up to you. The server, the network and the access are our side. See how far our support goes.
Updating
| 1 |
Read the release notes on the project before updating. Look for changes that break the configuration and for new Node.js requirements.
|
|
| 2 |
Take a backup first. A snapshot of the VPS and, separately, the configuration and data folder the documentation names. See backing up a VPS.
|
|
| 3 |
Update off-hours and with the agent stopped, following the official procedure.
|
|
| 4 |
Test with a harmless task. Only then reconnect the messaging channels.
|
|
| 5 |
If it goes wrong, go back to the backup and postpone the update until a calmer release.
|
|
|
Do not update blindly, or in a panic in the middle of the night. An agent that starts behaving differently after an update can delete or send things it did not do before.
|
Logs: what to look for
| Sign in the logs |
What it usually means |
What to do |
| Authentication errors with the provider |
Key expired, revoked or out of credit |
Check the provider’s console and replace the key |
| Requests repeating in a loop |
A failing task retrying |
Stop the task and fix the cause before switching it back on |
| Out of memory |
The VPS is too small for the load |
Reduce what runs at once or move to a bigger plan |
| Commands you did not ask for |
Malicious instruction or a model error |
Switch it off, revoke the keys and restore the backup |
Where the logs live depends on how you installed it and is in the documentation. If you run it as a system service, the service’s logs are read with the system’s own tools, such as journalctl on distributions that use it.
What to save and what to watch
| 1 |
Save the agent’s configuration and data (the folders the documentation identifies) off the server, and test that you can restore them. See making and testing your own backup.
|
|
| 2 |
Keep the keys in a password vault, not only on the server. If the VPS is lost the keys can be recreated, but it is a chore you can spare yourself.
|
|
| 3 |
Watch memory and disk now and then, with free -h and df -h. Logs that grow forever fill the disk.
|
|
| 4 |
Check the spend at the provider once a week. See what it costs.
|
|
|
Put a short monthly review in the calendar: version, logs, disk, spend and keys you no longer use. It is little, and it is what separates a useful agent from a problem waiting to happen.
|
|
Has the VPS run out of space or memory, or stopped answering? Open a ticket and we will check what we can see on the server side.
Open a support ticket
|
RECOMMENDED PRODUCT Web hosting with cPanel Domain and SSL included, daily backups and the panel you already know. from $6.60/mo (3-year plan, with coupon) See plans |
Angola
Brasil
Kenya
Moçambique
Nigeria
Portugal
South Africa
United Kingdom
United States
International
