An agent that only runs while your laptop is awake is not really always on. It works beautifully all afternoon while you watch it, and then you close the lid, and the brief you scheduled for 7am never arrives. Nothing crashed. The machine underneath the agent went to sleep, and the agent went with it.
This is the most common way a working OpenClaw setup stops working, and it has almost nothing to do with the agent, the prompt, or the model. It is a hosting problem wearing a disguise, which is why so much time gets spent debugging the wrong layer.
What actually ends the run
Three things kill a local agent, and none of them announce themselves.
- Sleep. Closing the lid suspends every process on the machine, your agent included. Both macOS and Windows treat this as correct behaviour, because a laptop is built to sleep, not to serve.
- Network changes. Moving between wifi networks, or a drop of a few seconds, cuts the agent off mid task. A long tool call that loses its connection does not always fail cleanly, and a half finished task is harder to spot than a missing one.
- Reboots. An operating system update restarts the machine without asking whether anything long running was in the middle of something.
What makes these hard to diagnose is that the failure is quiet. There is usually no error in the log, because there was no error. There was a process, and then there was no process. You find out when the output does not arrive.
Keeping the laptop awake is not the fix
The obvious workaround is to disable sleep and leave the machine plugged in. It does work, in the narrow sense that the process keeps running. But you have just turned a laptop into a server, and it is the worst server you own. It moves. It runs on battery when it moves. It reboots for updates on its own schedule. It is the same machine you need to carry to a meeting, and the moment you do, your always-on agent is off.
You have also tied your uptime to one device with no redundancy and no way to find out it stopped. A power blip at 2am and the morning brief is gone, and the first you hear of it is when you go looking for it.
The three honest places for an agent to live
There are really only three, and the trade is between what you get and what you maintain.
- Your laptop. Free, instant, and it stops when the lid closes. Fine for building and testing an agent. Not a home for one.
- A VPS you rent. It stays online as long as you keep it online. You take on the machine: provisioning, secrets, runtime updates, restarts after a crash, log storage, and some way to notice when the agent has quietly stopped. That is real work, and it is recurring work, not a one time setup.
- Managed hosting. The environment stays online by default and the operational chores belong to the platform. You configure the agent rather than the server. That is the model Qoren is built on.
The choice is not about technical ability. Plenty of people who could run a VPS perfectly well decide they would rather not run one more, because the maintenance never ends and it is never the interesting part of the job. Our guide on running OpenClaw without leaving your laptop on walks the same decision in more detail.
What moving it looks like
On Qoren, an environment is the host your agents run on, and it is a couple of commands from the Qoren CLI. You create an environment, deploy an agent onto it from a template, and read its log the same way you would locally.
qoren env create --name always-on --size standard
qoren agent create --env env_abc123 --template morning-brief --name "Morning brief"
qoren agent logs agt_def456 --limit 100The commands hide a twelve step install, and the CLI prints each step as it happens, so a failed deploy tells you which step failed rather than just failing. Everything the console can do is available here too, through the same account and the same plan limits.
Before you call it always on
Moving the agent off your laptop is necessary but not sufficient. Four questions decide whether it is genuinely unattended:
- Does it come back after the host restarts, without you logging in?
- Does the schedule actually fire while you are asleep, or only while a terminal is open?
- Will you find out when it breaks, or will the silence look exactly like success?
- Is there a hard spend cap, so a stuck loop overnight costs a known amount?
If the answer to all four is yes, the lid can close.