Where to find it
Every place that can be written by an AI has a Describe… item, first in its menu:
- New routine: the plus on Routines, then Ask an AI to write it.
- Change a routine: the routine's ellipsis, Describe changes…
- Change one step: the step's ellipsis, Describe changes…
- Add to a routine or a run: the plus on its page, Describe it…
- After a run: the run's ellipsis, Describe what to change…
- What a run collected: the review page, Sort these out
- Notes: the notes page, Deal with these notes
- Triggers: More › Triggers, Describe it…
They all open the same page.
The page
- Say what it is for. One sentence is enough: "A morning: water, ten minutes of stretching, a shower".
- Write it. The name beside the button is who writes it. Tap the name to choose another. See Who writes it for what it starts on.
- Or continue with an assistant. Sends the request to Claude, the ChatGPT app or any other assistant through the share sheet. Say what you want there if you did not say it here. When it answers, copy the whole answer and come back.
- Or paste. Paste its answer takes what you copied, words around it and all. When you come back with something new copied, the button says Paste what you just copied. To type a list of steps yourself, use Run something new… instead.
What comes back
The answer is never shown as code. You see what it made:
- A new routine: its name, the first steps, and three buttons. Keep it puts it in Routines. Start it walks it once, and the run asks at the end whether to keep it. Look it over shows every step, and every link or shortcut it would use, before you decide.
- Changes to a routine: how many, and Look over the changes, which shows each one before it is applied. A change can be undone.
- Answers about a run: what to keep and what was only for that run, and Look over the answers.
- Triggers: each trigger it wrote, to add one by one.
If the answer cannot be read, the page says why in a sentence. Try again sends the model the reason and what it wrote. Copy this for your assistant is for an answer from a chat, so you can paste the reason back to it.
What an assistant can and cannot put on a step
It can write names, descriptions, tips, tags, a colour and a symbol, links, shortcuts to run, and triggers. It can also write the things that reach past your phone: a webhook, a calendar event, a reminder, an alarm, a Home scene or a Health entry.
Those arrive switched off. Each one is marked orange on its step and does nothing, not on arrival and not as a button, until you open it, see where it goes and what it sends, and press Allow. Remove takes it off the step instead.
For Home Assistant, Slack, ntfy, PagerDuty and IFTTT it writes what to do there, a Home Assistant call like light.turn_off or a message for Slack, and the app knows the address. The step then says "Home Assistant: light.turn_off" rather than an address.
An assistant never sees a password or a token. It writes {secret:name} where one goes, and the step asks you to paste the value once. It is kept in this phone's keychain and never leaves it: not in an export, not in a shared link, and not in anything an assistant is shown.
A timer on a step you did not want
A length on a step makes it count down and ring. Models are asked to leave it off unless you give one, but if a step has one you do not want: the step's ⋯ → Take the timer off.