Automations — when this happens, do that

You tick the checkbox on a task. The Stage changes to done and the card slides out of your board — and you never clicked the Stage.

That's an automation: a small "when this happens, do that" rule that lives on a Collection. You didn't wait for us to build it. You asked your AI for it, and it wrote the rule.

The three parts

Every rule has the same three parts. Read them out loud and it's a sentence:

Part	Means	Example
when	the thing that sets it off	when Done becomes ticked
if (optional)	an extra check before it runs	…and only if the Stage isn't already done
then	what happens	set Stage to done

when — the trigger

Two things can set a rule off:

• changed — a property's value changed. You can narrow it to a particular new value, a particular old value, or leave it open so any change fires the rule.
• created — a brand-new entry was just made. Handy for "every new task starts in do today".

A useful thing to know: dragging a card between columns on a Board is a change to whatever property the board groups by. So a rule like "when Stage becomes done, tick the Done box" works from a drag with nothing extra.

if — the optional gate

A check that has to pass before the rule runs. Same set of comparisons you already use for conditional card styling: is / is not, greater / less than, contains, empty, has a value. You can require all of them, or any of them.

Most rules don't need one. Reach for it when you want "…but only when".

then — the actions

What actually happens. A rule can do up to eight things in a row:

• set a property to a value
• set now — stamp a date or time property with this moment
• clear a property
• toggle a checkbox
• increment a number (counting attempts, streaks, retries)
• add tag / remove tag — one tag, leaving the others alone

Why this is safe

Everything an automation does is something you could have done yourself by tapping. It writes values, exactly like the app does when you edit a field — no hidden powers, nothing that reaches the internet, nothing that can delete an entry.

Three more things keep it calm:

• You always get a toast. When a rule fires you see "Stage → done · by 'Close out a finished task'" with an Undo button. Undo puts everything back, including the tick that started it.
• Loops are rejected before they save. If rule A would set off rule B which would set off rule A, Kaizendex refuses to save it and tells you exactly which action closes the loop.
• Rules cascade only a little. One rule can set off another — ticking Done sets the Stage, which stamps a date — but the chain stops after three steps, always.

Writing one

The easy way is to ask:

"When I tick Done on a task, set the Stage to done."

Your AI will describe the rule back to you in plain words and ask before saving it — an automation changes your data quietly and repeatedly, so it checks first.

To see or edit your rules by hand, open Automations from the sidebar's rail. Every rule you have is one row in a table, whichever Collection it belongs to: an On checkbox, its name, what starts it, its Collection, a one-line plain-English summary of what it does, its status, and when it last ran and runs next. The summary always uses names — the preset, agent, property or tag it points at. If one of those no longer exists, it says so (for example "a missing preset") instead of showing a code.

The bar at the top of the page stays put while the table scrolls. It holds the search box (it shows how many rules you have, and "3 of 12" while you type), three filters — Collection, Trigger and Status (on or off) — and New automation. Each filter choice shows how many rules it would leave; a filter that is set shows its value, and the ✕ beside it clears it. It's the same bar on a phone. Click a rule's name to open it.

You can also reach the same list from inside a Collection — open it, click the ⋯ menu in the top bar, and choose Automations. It is the same list, just filtered to that Collection: everything you can do on the page you can do there.

Edit on a rule opens its JSON. The full grammar is in Automations reference, and there are ready-made rules to copy in Automation recipes.

Rules that run on a clock

Not every automation waits for you to do something. A schedule trigger fires on the clock instead — every day at a time, on chosen weekdays, or every so many minutes. These run on our servers, so they happen whether or not the app is open on any of your devices, and the time is your time: "08:00" means 08:00 where you are, and it keeps meaning that when the clocks change.

Your recurring AI check-ins are exactly this. "Send me a motivational message every morning at 8" is a schedule trigger with a runAgent action — it hands your instruction to one of your agents and lets it answer in its chat. That is why your AI tasks live on the Automations page alongside everything else: they were never a separate feature, just an automation with a clock for a trigger and an agent for an action.

Recurring entries

A schedule can also createEntry — mint a real entry in the collection, dated the day it ran. This is how "did I do it today?" tracking works: the entry is created by our servers whether or not you open the app, and an entry you never tick stays in your history as proof you skipped that day. There are no ghost rows and nothing is invented at the last minute.

By default it skips the day if you already logged that thing yourself, so logging at 05:50 doesn't leave you with a duplicate at 06:00.

It starts the day, unless you say otherwise. A recurring entry with no time set runs just after midnight in your timezone, so today's entry is already there when you first open the app — you should never have to wait until mid-morning for the day to exist. Set a time on the schedule whenever you want a different one; a time you chose is always kept.

If a scheduled rule ever fails to run, it now says so: the rule shows a Last run failed badge on the Automations page and you get a notification, because a missing entry you never hear about is worse than one that arrives late. A rule that is on but shows Not scheduled has lost its next run — untick it and tick it again to re-arm it.

It can create the entry through a preset. Point the rule at one of the Collection's presets and the entry arrives already titled, already carrying its emoji and its Category tag — exactly as if you'd picked the preset by hand. Anything the rule sets itself wins over the preset's defaults.

That's what makes one shared Collection work for many things. In a single Habits Collection, the "already logged today?" check looks only at the habit that rule creates, not at the Collection as a whole — so ten habits with ten morning rules give you ten entries, instead of the first one at 06:00 silencing the other nine for the rest of the day. See Tracking habits for the whole pattern.

A preset brings its own checklist. If the preset has sub-entries, the scheduled entry gets exactly those — the same list you get when you pick the preset by hand — instead of the Collection's default sub-entries. A preset with no sub-entries still gets the Collection's defaults.

One rule can choose. "Make preset A on Monday, Wednesday and Friday, preset C on Sunday, and preset B the rest of the week" is one rule, not three. Ask your AI for it, or see Branching for the JSON.

If we were down, we backfill — up to 14 days. Missed days inside that window are created for you the moment the runner comes back, so an outage on our side never leaves a hole in your record. Anything older than 14 days is not backfilled: at that distance a missing entry means the same thing as an unticked one, and inventing rows for a month you never saw would be worse than the gap.

A scheduled rule shows its last run on the page, so you can see it working without waiting around for it.

Bringing in new videos from channels you follow

If you keep a collection of YouTube channels, a scheduled rule can save their new uploads for you — each one becomes an entry in your Content collection with its transcript and a short summary, exactly as if you'd asked for it by hand. The action is called channelIngest.

Two things to set up:

1. Tick the channels you want. The channels collection needs a checkbox property called Selected for ingestion. Only ticked channels are ingested — being in the collection is not enough, on purpose. (No checkbox at all? The rule stops and tells you to add one, rather than importing everything.)
2. Add a daily rule on that collection — ask your AI: "every morning, import new videos from my selected channels".

What to expect:

• A handful at a time. Up to 5 new videos per run by default, and never more than 20 — so switching this on doesn't dump a channel's back catalogue on you. Set sinceDays if you want to ignore anything older.
• Nothing twice. A video you already have is recognised and skipped.
• Nothing lost. If a run is interrupted, the videos it had already picked up are remembered and finished next time. A video that failed to import is retried before anything new is fetched.
• Quiet days stay quiet. No new uploads means no entries, no notification, nothing to read.
• Unavailable feeds stay visible. If a channel could not be checked, the run records that failure. It does not claim there were no new uploads. Videos saved from other channels remain saved.

This rule only saves the videos. If you also want your AI to fold what it learns into a guide, that's a second rule that asks your AI to do it — see Guides and where their advice came from.

Turning one off

Every rule has an On checkbox in the Automations table. Unticking it keeps the rule around without running it, which is the right move when you're not sure yet — much better than deleting it and rewriting it later. Your AI can do it for you too: "pause the close-out rule".

Duplicate (in a rule's ⋯ menu) copies a rule so you can vary it. The copy arrives off, so it never starts changing your data the moment you make it — tick it on when you're happy with it.

Delete (in the same menu) removes the rule at once. The message that appears has an Undo — it brings the rule back exactly as it was, still on or off, with the same schedule. Your AI can bring a deleted rule back too.

What it doesn't do (yet)

• Rules only run on your own edits — on any of your devices. A change your AI assistant makes in a chat doesn't set them off.
• Rules act on one entry — the one that changed. A rule can't reach into a different Collection.
• Rules can't send email, call a website, or do anything outside your own data.

Where to next

• Automations reference — every trigger, every check, every action, with copy-pasteable JSON.
• Automation recipes — worked examples you can lift.
• Tracking habits — a daily entry per habit, created for you.
• Customizing the card — the other half: what happens when you tap, rather than when a value changes.

Move unfinished items on a schedule

A scheduled rule can update properties on entries in its collection that match its conditions. For a daily list, select Stage = do today and Done ≠ checked, then set Stage to your unfinished-items option. All actions in this rule must be property updates; use a separate rule for creating recurring entries.

The JSON editor shows how many entries match now and a few example names. The run checks again at execution time. Entries changed after the scheduled time, or while the update is being prepared, are preserved. A delayed cleanup therefore may leave recently edited items in place. Errors appear in the automation's run history, rather than being reported as successful updates.

For repeatable routines, schedule Create entry from a preset. Each day gets a separate unchecked entry, keeping past checkmarks available in history.