Does the automation do everything on its own?

It performs the task on its own when triggered. But it only does what the chosen agent can do, and any action classified as sensitive stops and becomes a pending approval. A pending item without a decision expires and the run is closed without acting. Autonomy here is about execution, not decision.

With whose permissions does it run?

Those of whoever created the automation. If the trigger comes from an external system, those of the API key's owner. There is no special automation access: if the person cannot see a document or use a function, neither can their automation.

What kinds of triggers exist?

Three. By schedule: every day, every week, every N hours or minutes, in the chosen time zone. Through the “Run now” button. And from an external system: in on-demand mode, your helpdesk, CRM or ERP calls an address with an API key and sends the event's data. There is no trigger on a published document or on a received message.

Where do I see the result?

In the automation's history, with status, duration and the “View generated conversation” link; in the notification bell, which reaches the owner and the collaborators; and in the conversation itself, which can be continuous across runs or new on every trigger.

Can I share an automation?

Yes. Whoever creates it is the owner and invites people as viewers or editors. Collaborators follow the history and get the notices of every run. Sharing the automation does not share its agent: each has its own scope.

How much does a run cost?

Each run that ends as finished consumes credits like a conversation at the chosen agent's AI level; a run that fails, expires awaiting approval or is cancelled is not charged. Credits are shared by the company, and the history shows when and why each run happened. Allowances and prices are those of the company's plan.

Platform · Automations

The routine runs on schedule. The result arrives with an owner.

An automation is an agent, a team or a workflow performing a task you wrote — at the time you set, when you press the button or when another system says so. Every run has a history, becomes a conversation you can open, notifies whoever shares it and, if the task needs to act on a system, stops and asks for approval.

Try it free Book a demo

See how it works

SCHEDULE · BUTTON · EVENT

three ways to trigger; none of them skips an owner and a history

NEXT 5 RUNS

the preview shows when it will run before you save

STOPS AND ASKS FOR APPROVAL

a sensitive action in an unattended run becomes a pending item, not an action

NOTICE TO WHOEVER SHARES

finished, failed or awaiting approval: the owner and collaborators are told

Where this shows up in your day

What repeats every week and depends on someone remembering.

Routines that exist in every company and that today start with a person opening the same documents, on the same day, to produce the same summary — when they remember.

Finance

The Monday summary

Every week someone gathers the closing's open items, checks them against the procedure and sends a summary to the manager. When that person is out, the summary does not go out.

Every Monday, 7am

Customer service

The ticket that arrives after hours

The ticketing system gets a request at 10pm. Triage based on the approved documentation only happens the next morning, when someone opens the queue.

Triggers when the ticket comes in

Quality

The check nobody wants to do

Comparing the published procedure with the week's records is repetitive work, so it gets left for later. Later becomes never.

Runs on its own, tells the owner

How it works

From a written task to a run with a history.

Five steps. The task text is the same you would type in chat; the rest is choosing who performs it, when, and who follows along.

01
Describe the task
Write the request as you would in chat — or answer three questions and Skyller drafts the task, suggests the frequency and picks the most suitable agent for you to review.
02
Choose who performs it
An existing agent, team or workflow. The automation uses that target's knowledge, tools and AI level.
03
Set the trigger
Every day, every week, every N hours or minutes, with a time zone and a preview of the next five runs. Or on demand: through the “Run now” button and, if you wish, from an external system.
04
Adjust the follow-up
Continue in the same conversation on every run or open a new one; time limit and retries on failure; people or groups who get the notice.
05
Follow the history
Each run shows status, start, duration, attempt, who triggered it and the “View generated conversation” link. Failed or asked for approval: it stays visible, with the reason.

Live proof

The run stopped. Someone decided. It finished.

This is the difference between automation and autonomy: the routine runs without anyone asking, but the action that touches a system waits for a person.

Starting point
A “Weekly collections” automation, every Monday at 7am, performed by the finance agent, which checks the connected order system and has “record payment” marked as an action with approval.
What happens
1. 01At 7am Skyller runs it: the agent reads the published collections policy, checks overdue orders and drafts the summary — all without anyone at the screen.
2. 02On reaching “record payment for order 4471”, the run stops: the history marks the run as paused and the manager gets the “awaiting approval” notice.
3. 03The manager opens the pending item, approves in the conversation and the run resumes; the history closes as finished, with the conversation link.
Result
The routine went out on time, the sensitive action waited for a person, and everything — who created it, when it ran, who approved — was recorded in the same place.

What this scene does not promise

An automation only does what the chosen agent can do: published documents and enabled functions. Without a connected system, it looks up and writes; it records nothing.
The run uses the permissions of whoever created the automation — or of the API key's owner, when triggered by an external system — not a special access.
A pending item without a decision does not become an action: it expires and the run is closed without executing.

Scene with a fictional company and orders; the schedule builder, the history and the approval pause are the product's.

To evaluate calmly

How it triggers, what it keeps and who follows along.

Select a topic to see its capabilities, conditions, and limits.

Clock with preview: Presets for every day, every week, every N hours or minutes, plus an advanced mode. Explicit time zone and a preview of the next five runs before saving.
Run now: Any automation runs right away through the button. The run happens in the background and the result shows in the history and in the notification bell.
Trigger from another system: In on-demand mode, the automation gets an address your helpdesk, CRM, ERP or monitoring calls with an API key. The event's data enters the task as fields — and the text flags a field that did not arrive.

Skyller · Team pending-task summary

Team pending-task summary

1 · Identity2 · When to run3 · What to run

Frequency

Time

Schedule preview

Every day · 09:00 · America/Sao_Paulo

Time	Status	Result
09:00	Preview	The run has not been simulated yet

Demonstration · illustrative data

Real application

Ticket triage, built by the people who answer tickets.

The gain is not replacing the queue: it is the first reading going out the minute the ticket arrives, based on what the company already documented, with the person deciding whatever falls outside the pattern.

01Support analyst
Creates the on-demand automation “Ticket triage” with the support agent, whose fixed base is the published documentation.
No clock: the automation only runs when someone or some system calls it.
02Company IT
Creates the API key in Settings › Integrations and configures the ticketing system to call the automation's address on every new ticket, sending title and description.
The run uses the permissions of the key's owner.
03Skyller
On every ticket, performs the task with the title and description as fields: classifies, points to the applicable procedure with its source and drafts the first reply.
A field that did not arrive stays visible in the text instead of disappearing.
04Support analyst
Gets the notice, opens the generated conversation, adjusts the reply and sends it. Whatever falls outside the pattern, she handles directly.
Replying to the customer remains the person's decision.
05Coordination
Follows the history: how many runs, how many failed, how long each took — and adjusts the task when the pattern changes.
Editing the task applies to the next runs; the previous ones stay in the history.

Comes ready

Schedule builder with presets, time zone and a preview of the next five runs
Run now button and on-demand trigger from an external system with an API key
History per run with status, duration, attempt, origin and generated conversation
Notice to owner and collaborators and a pause for approval on sensitive actions

The company decides

Which agent, team or workflow performs each automation and at which AI level
Which actions ask for approval — and therefore where a run may stop
Time limit, retries and whether each run continues the conversation or opens another
Who is invited as viewer or editor and receives the notices

Questions from whoever decides

What people ask before letting something run on its own.

Next step

Pick the routine someone forgets to do.

Write the task as you would ask in chat, set the time and see the preview of the next runs. On Monday, the summary lands in the bell with the conversation open. Or schedule a demo and build the first automation with our team.