Spend caps, fail-closed
Four firings a day per automation, fifty across your account — and if the check itself errors, the run doesn’t happen.
Automations
Tell an agent what to do and when, in one sentence. From then on it wakes on schedule, does the work with its own tools, and leaves the finished draft in your chat — with a notification telling you it’s ready.
Drafts, never sendsYour timezone, by constructionFree to create and manage
No builder, no node graph, no cron syntax. You say it to the agent you want to own it — and it’s even allowed to push back on a cadence it thinks is unwise.
Every morning at 7, go through my inbox and draft a reply to anything that actually needs one.
Every day at 9, prepare a post about our new collection — caption, hashtags and image.
Every evening at 6, pull the day’s numbers and write me the two lines that matter.
The ceiling
An automation prepares — you send. There is no unattended publish and no setting that unlocks one: “auto” is coerced back to prepare-and-wait in two separate places in the code, and the background agent’s own instructions end with “never send, publish, or take any irreversible action unattended.” Your 7am automation can’t embarrass you at 7am.
Two schedule shapes exist today: daily at a time — in your timezone, daylight-saving correct by construction — or every N minutes, thirty minimum. No weekday selection yet, no cron, no monthly trigger. We’d rather you know before you build your week around it.
Four firings a day per automation, fifty across your account — and if the check itself errors, the run doesn’t happen.
Three consecutive failures and it disables itself and tells you why, instead of burning credits forever.
Out of credits overnight? The firing is skipped and labelled so. Nothing is disabled.
A run that promised a deliverable must produce it — or state the blocker in one sentence. Every run records why it ended.
Repeat-call guards, a stall detector, and a hard time cap with a graceful final turn.
An integration dropped mid-run? Reconnect within 24 hours and the paused call replays itself.
Everything else starts from a sentence — a new account ships with zero pre-installed automations, so you get exactly the routine you described, not the nearest template. These three are toggles.
What’s scheduled, what performed, what to post next.
Your inbox categorised: Priority, Work, Bills, Deliveries, Security, Personal.
“Where’s my order?” gets the order looked up and a reply drafted. 10 a day, drafts only.
No — not with a setting, not with a plan upgrade. Every firing ends in a draft waiting for your click. “Auto” is coerced back to prepare-and-wait in two independent places in the code.
Not natively — the schedule model is daily-at-a-time or an interval of 30 minutes or more. Weekly works as a 7-day interval. Weekday selection is a real limitation today, not a hidden checkbox.
Tell the agent: “move my morning automation to 8”, “pause the daily post”, “delete that”. The /automations page can pause, resume and delete; creating and editing happen in conversation.
The one you asked — nine of the ten can own automations, each with only its own toolbox. Maya’s triage can read your inbox; Ashley’s daily post cannot. The boundary applies at 3am exactly as in a live chat.
After three consecutive failures it pauses itself and shows why, rather than retrying forever on your credits. You re-enable it deliberately once the cause is fixed.
It’s told not to: if there’s genuinely nothing worth doing, it must say so in one short sentence rather than invent a deliverable. An empty morning reads “nothing needed you”.