Sleep and wake, explained
What happens when a cloud computer goes to sleep: memory snapshots, deep sleep, waking from a preview link, and what you pay while it sleeps.
A computer you only use a few hours a day shouldn't cost you money the other twenty. That is the idea behind sleep. On homemade, a computer that sits idle goes to sleep, stops costing running time, and picks up exactly where you left it when you come back: same files, same terminals, same processes.
This post explains what actually happens when a computer sleeps and wakes, the two kinds of sleep, how long a wake takes, and what you pay.
When a computer goes to sleep
A computer sleeps after it has been idle for a while. Activity, like typing in a terminal, pushes that moment back. The default is 30 minutes. You can pick 10 minutes, 30 minutes, 1, 2, 4 or 8 hours, or never (on Pro), per project or per computer. In a repo you can set it with sleep_after in .homemade/config:
{
"sleep_after": "2h"
}You can also send a computer to sleep yourself: from its page, with homemade vm sleep, or through the API.
What "asleep" means
A sleeping computer is paused, not shut down. When it goes to sleep we save two things:
- Its disk. Every file, including things outside your repo like installed packages and shell history.
- Its memory. Everything that was running: your dev server, a half-finished test run, the shell you had open, an agent in the middle of a task.
That is why a wake isn't a reboot. Your terminals come back with their scrollback, and a process that was running keeps running from the same point. If a dev server was serving on port 3000, it still is.
The first time a computer sleeps, it writes a full copy of its memory. Later sleeps write only what changed since the last one, which is much smaller and faster.
Light sleep and deep sleep
There are two stages.
Asleep. For the first hour, the computer stays on its host, paused, with its memory snapshot right there. A wake from this state is the fastest. In our measurements, a wake after a short sleep reached a typed echo in the terminal in 0.58 to 0.86 s, and after 31 minutes asleep in 0.74 to 0.81 s.
Deep asleep. After an hour asleep, the computer's disk moves off the host into storage, and the computer gives up its place on the host completely. It no longer counts toward your plan's limit of computers at once. Waking it brings up a fresh computer with the same disk on whichever host has room. Because it starts from the disk, running processes from before the deep sleep are gone, and the services your project defines start again. Your files are all there.
A deep wake takes a little longer. On 19 November 2026, deep wakes on our staging environment took a median of 1.80 s from the wake request to a typed echo, with the slowest 10% at 4.0 s or under (30 wakes over three projects). The dashboard tells you when a wake is taking longer than usual for that project.
How a computer wakes
A sleeping computer wakes when:
- you press Wake VM on its page, run
homemade vm wake, or call the API; - someone opens one of its preview links and the project has Wake on visit switched on.
That second one is what makes sleep easy to live with. Send a teammate the preview link of your branch, go to lunch, and the computer sleeps. When they open the link, they see a short "Waking this app up" page, and the app opens by itself when it is ready.
A few rules keep this sane:
- Link previews in chat apps and bots never wake a computer.
- One computer wakes from visits at most once every 2 minutes.
- If Wake on visit is off, visitors only see that the app is asleep. Members of your team see a Wake it up button.
More about preview links in the 63-character hostname post and the previews docs.
The one slow case
There is one case where a wake can be slow: waking a computer right after it went to sleep, while it is still writing its memory snapshot. The wake waits for that snapshot to finish. For a large computer whose first sleep writes all of its memory, that can take a while.
We tried to skip that wait by resuming a computer that had only just been paused, without a snapshot. In our lab, a computer resumed that way never answered on its control channel again. Writing the snapshot is what resets that channel. So a wake still waits for it, and then resumes from memory. In practice, someone coming back after a minute never sees it, because the snapshot is done by then.
What you pay while a computer sleeps
Nothing for running time. You pay for the time a computer runs, per second, by size. A sleeping computer is not charged for running time and doesn't use up the hours included in your Hobby plan.
Its disk is kept for a limited time:
| Plan | Sleeping computers are kept |
|---|---|
| Hobby | 7 days |
| Pro | 30 days |
| Pro with Unlimited sleep ($6 a month) | Until you delete them |
Unlimited sleep includes 200 GB of sleeping disk, then about $0.04 per GB-month. After the keep time, a computer that hasn't woken is deleted. You get an email first (2 days before on Hobby, 7 days and 1 day before on Pro), and waking the computer starts the count again. Your project always stays, so you can make a new computer from it in seconds.
We wrote about the engineering that makes free sleep possible for us in how we keep sleeping computers cheap.
Why not just keep everything running?
You could leave a computer on all the time. On Pro, "never" is a valid sleep setting, and for a long agent task or a demo server that is the right choice. But most dev environments sit idle most of the day. A medium computer (8 vCPU, 16 GB) is $0.0648 an hour. Running all month, 730 hours, that is $47.30. Used four hours a day on workdays and asleep the rest, it is about $5.70.
And because sleep keeps memory, you don't trade that saving for a cold start every morning. You open the tab, the computer wakes, and your terminal is where you left it.
Tips
- Long agent runs. If an agent will work unattended for hours, give that computer a longer sleep time, or never on Pro, so a long task is never paused halfway.
- Shared previews. Turn on Wake on visit for projects where people outside your team open links.
- Many branches. Make a computer per branch and let the idle ones sleep. They cost nothing while asleep, and each wakes with its own state.
The full reference is in the sleep and wake docs, and per-size prices are on the pricing page.
Try homemade
Every project gets its own cloud computers, with Claude Code and Codex signed in. The Hobby plan is free.
Related posts
- How we keep sleeping computers cheapWhy a sleeping computer costs you nothing on homemade, and the engineering that keeps it cheap for us: diff snapshots, sparse memory and deep sleep.
- Preview URLs and the 63-character hostname limitWhy every preview link is one DNS label, how project, computer, port and team share 63 characters, and how we trim names without collisions.
- How we got click to terminal under 4 secondsFrom pressing Create to typing in a terminal: what took the time on a new cloud computer, what we cut, and the numbers before and after.
- All postsEvery post