every

Schedule anything on your computer. Actually know it ran.

_{launchd on macOS · systemd on Linux (beta) · Task Scheduler on Windows · zero dependencies}

Install · Schedules · Commands · vs cron

---

cron never tells you it silently skipped your backup. launchd wants 40 lines of XML before ignoring you too. every is one human phrase — and a straight answer to "did it run?".

every day 9am -- brew update
every 30m -- '~/bin/sync-notes.sh'
every monday 10:00 -- './weekly-report.sh'

$ every list
NAME           SCHEDULE      LAST          STATUS   NEXT
brew           day 9am       24 Jul 09:00  ok       25 Jul 09:00
sync-notes     30m           24 Jul 14:30  ok       24 Jul 15:00
weekly-report  monday 10:00  21 Jul 10:00  FAIL(1)  28 Jul 10:00

$ every log weekly-report     # exact output of the run that broke
$ every doctor                # plain-language diagnosis

vs cron · vs raw launchd

	cron	raw launchd	every
Add a job	30 9 * * 1 in crontab -e	~40 lines of XML + launchctl	every monday 9:30 -- cmd
Did it run?	silence	silence	every list → ok / FAIL
What did it print?	a local mailbox nobody reads	wire log paths yourself	every log <name>
Mac was asleep at 9am	run lost forever	runs on wake	runs on wake — and you can verify it
PATH	minimal, brew tools "not found"	minimal	your login shell, as in your terminal
Working directory	$HOME, always	configure it yourself	the directory you added the task from
A run fails	nothing happens	nothing happens	macOS notification + FAIL in list
When it breaks	Console.app archaeology	Console.app archaeology	every doctor tells you why

Apple deprecated cron on macOS years ago. every is launchd — with a human interface and a memory.

Install

macOS and Linux — one line, no sudo (installs into ~/.local):

curl -fsSL https://raw.githubusercontent.com/serhiileniv/every/main/install.sh | sh

Windows — one line, PowerShell:

irm https://raw.githubusercontent.com/serhiileniv/every/main/install.ps1 | iex

Re-run either to upgrade. The installer verifies the download against the release checksums, and puts the binary where your scheduler already expects it, so an upgrade reaches tasks you scheduled with an earlier version.

Prefer Homebrew on macOS? It upgrades in place.

brew tap serhiileniv/tap && brew install every

Both install the same binary. Homebrew gives you brew upgrade; the one-liner works anywhere, including machines without Homebrew.

Native Windows tasks use the Windows Task Scheduler and store data under %LOCALAPPDATA%\every. Interval schedules on Windows require at least one minute; use WSL if you need the Unix backends or sub-minute intervals.

Options: system-wide, a pinned version, uninstall

# system-wide
curl -fsSL …/install.sh | sudo sh -s -- --prefix /usr/local

# somewhere else, or a specific release
curl -fsSL …/install.sh | sh -s -- --prefix ~/opt --version 0.4.0

# from a checkout (same flags; builds what's in front of it if Go is present)
git clone https://github.com/serhiileniv/every.git && ./every/install.sh

# uninstall (tasks and logs are kept; it won't strand a live timer)
curl -fsSL …/install.sh | sh -s -- --uninstall

Re-running the installer upgrades in place. Scheduled tasks keep working across an upgrade — units point at <prefix>/bin/every, which stays put.

Zero dependencies, literally: every is a single static binary with nothing to install alongside it — no runtime, no interpreter, no shared libraries. The installer verifies the download against the release checksums, and sets up man every and tab completion for bash, zsh, and fish.

Schedules

You type	It means
90s · 15m · 2h	fixed interval
hourly	every hour
day 9am · day 17:30	daily at that time
day 9am,6pm	daily, several times
weekdays 9:30 · weekends 11am	Mon–Fri / Sat+Sun
monday 10:00 · monday,thursday 6pm	weekly on those days

Commands

every <schedule> [--name NAME] [--quiet] -- <command>   schedule it
every list [--json]                           status of everything
every log <name> [-n N]                       output of recent runs
every run <name>                              run it right now, see the output
every pause / resume <name>                   stop / start scheduling
every rm <name>                               remove (logs are kept)
every doctor                                  why isn't it running?

Failed runs pop a desktop notification (silence it per task with --quiet).

every list --json prints one object per task for scripting — pipe it to jq:

every list --json | jq -r '.[] | select(.status | test("FAIL")) | .name'   # failing tasks

Exit codes

Follows sysexits.h, so scripts can branch on $?:

Code	Meaning
0	success
64	usage error (bad arguments)
66	no such task / no logs yet
1	other failure

every run (and scheduled runs) exit with the command's own code, or 124 on --timeout, or 128+signum if a signal killed it.

Fine print

• The command is a shell line (like cron): tokens after -- are run through your login shell, so env prefixes, pipes, &&, and globs all work. Your outer shell strips quotes first, so quote args with spaces or metacharacters as you would at a prompt — wrap the whole thing in one quoted string when in doubt: every day 9am -- 'pg_dump db | gzip > ~/backup.gz', every 1h -- 'touch "my file.txt"'.
• Timeouts: add --timeout 30m to kill a run that overruns — otherwise a task that hangs will block its own next run (the OS won't start a second copy of the same task). The kill takes the whole process tree with it.
• Output is captured but bounded (first + last 32 KB per run), so a chatty command can't blow up memory or fill the disk.
• Tasks live in ~/Library/LaunchAgents/com.every.<name>.plist; runs are recorded under ~/.local/share/every/ (logs rotate at 5 MB). launchd can't execute from TCC-protected folders, so agents run a copy of every from the data dir — see DECISIONS.md for design notes.
• Linux (beta): systemd user timers, same commands; units live in ~/.config/systemd/user. Timers stop at logout unless you run loginctl enable-linger $USER (the installer tells you if it's off). Failed runs notify through notify-send. Field reports very welcome — the units are tested, months of real desktop uptime aren't.
• Windows: native Windows Task Scheduler tasks, same commands; tasks live under the \\every\\ task path and invoke the binary through the command wrapper. Calendar tasks use StartWhenAvailable; interval tasks require at least one minute. Tasks use the current interactive user by default, so they run while that user is logged in. The default shell is cmd.exe (set EVERY_SHELL to a PowerShell executable when needed). Failure notifications use best-effort msg.exe and are always recorded in the run log.
• Where things live: on macOS/Linux, tasks/logs/ledgers live under ~/.local/share/every by default; native Windows uses %LOCALAPPDATA%\every. Honors $XDG_DATA_HOME (data) and $XDG_CONFIG_HOME (systemd units on Linux); EVERY_HOME overrides the data dir entirely. NO_COLOR is respected. If you set XDG_DATA_HOME after creating tasks, the old ones stay in the old dir.
• Uninstall: every rm each task, then rm -rf ~/.local/share/every.

Roadmap

Where it's going and what's still rough: ROADMAP.md. Issues and PRs welcome.

MIT © Serhii Leniv