A personal agent that keeps its word

Say it once.
Dooly keeps at it.

Most assistants answer and forget. Dooly holds on to what you asked for — checks on it, decides when it matters, and has the result ready before you come looking.

No cron jobs. No workflow builder. You say it in your own words.

Today running on its own
  • 09:00You: Send me five Spanish words every 20 minutes.

Nothing was posted into your chat. It waits until you open it.

The gap

An assistant that only answers is a tool you have to remember to use.

You ask a chatbot to help you learn vocabulary. It writes five words, beautifully. Then nothing — until you remember to come back and ask again. The intent was yours to carry the whole time.

The work was never the hard part. Staying with it was.

The model

Two things to say. Everything else is Dooly's job.

Goals where you want to get to

IELTS 7.0 by December. 10km without stopping. $10,000 saved. No schedule, no steps — and optional, because plenty of what you want help with is not a project.

Follow-ups what Dooly keeps caring about

“Every 20 minutes send me five words.” “Tell me if a customer has not come back in three days.” One sentence in your own words, kept whole — never compiled into a checklist you then maintain.

A follow-up does not need a goal, and a goal does not need a follow-up. Naming a goal later attaches it to work already running — nothing recreated, nothing lost.

While you are away

It works the way a person would: look, judge, then act.

01

A moment comes

A clock, a signal from your data, or a condition you stated — “if three days pass with no visit”.

02

It looks first

It reads the current state before deciding anything. A trigger is permission to look, never an order to act.

03

It decides

Do the work, tell you, wait — or nothing at all. Deciding nothing is worth saying is a complete outcome.

04

It leaves the result

Finished work waits in Updates, written when the moment came — not when you tapped.

The order matters. A schedule that fires an action is a cron job, and a cron job breaks the first day the world changes — the news source moves, the customer already rebooked. Dooly looks before it decides, so it can notice that.

Restraint

Most of the work is knowing when to stay out of your way.

Silence is a real answer

A follow-up that wakes and finds nothing new says nothing. Not a failure, and not a “no updates!” message — the absence of one.

Your chat stays your chat

Results arrive in Updates, never appended into a conversation you were not reading. Open one and it becomes its own thread, already written.

It knows when not to reach you

Active hours, quiet hours, rate limits, cooldowns — “only between 8am and 10pm” is something you say once, in a sentence.

In practice

You describe it once, the way you would tell a person.

Learning

Send me five words every 20 minutes, and ask me each morning what to focus on.

Two rhythms, one intent — the morning choice decides what the 20-minute one sends.

News

Every afternoon at three, round up the tech news and put it in a page.

A page a day. When a source dies it finds another, instead of failing quietly.

Customers

Tell me if Sarah has not come back in three days.

Checked on a cadence, raised only while it is still true — it can see she already booked.

Money

If anything over $500 goes out, take a look for me.

Wakes on the transaction, not on a timer, and only above the number you named.

Learning, money, customers, news — the same follow-up underneath. No separate builder per use case, and nothing to wire together.

Why not just…

You have tried the other three. This is what they leave out.

Can it…A reminderA chatbotA workflow builderDooly
Fires at a time you set
Does the work, not just the nudge
Looks at the world before deciding
Stays silent when nothing changed
Survives the source changing
Set up by saying one sentence

A workflow builder can do most of this — after you build the workflow, and again every time something about it changes. The difference is not capability. It is who does the maintaining.

Authority

It only goes as far as you let it.

Each follow-up carries how far it may act on its own — and you can name exceptions without narrowing everything else.

1ObserveLooks and keeps track. Says nothing, changes nothing.
2NotifyMay reach you when it finds something worth your attention.
3ActMay do reversible work on its own — and still asks first for anything you listed.

“Keep an eye on it for me, but ask before you send any email.” One named exception, and it holds — whatever else the follow-up may do.

The first minute

Setup is one sentence long.

There is no builder to learn, no triggers to wire, and nothing to maintain afterwards.

Start free

No card required

  1. 1Say the thingOne sentence, in your own words. “Every Friday, tell me what I spent this week.”
  2. 2It asks only what it mustUsually nothing. If a detail is genuinely missing — which day, which table — it asks once.
  3. 3Close the tabThat is the setup. The next time it matters, the work is done and waiting in Updates.

Straight answers

The questions people actually ask.

A wake that finds nothing new says nothing — that is a complete outcome, recorded as one. You can also give any follow-up quiet hours, active hours, a rate limit, or a cooldown, in the same sentence you set it up with.

It acts only as far as you allow

Observe, notify or act — set per follow-up, with named exceptions that always ask first.

Stop it in one tap

Pause or end any follow-up. What it already did stays; it simply never wakes again.

Free until it is doing real work

The limit is daily capacity, not features. The same agent on every plan.

Tell it one thing you keep forgetting.

That is the whole setup. It will be there tomorrow — and the day the answer changes, it will be the one who noticed.