Triggers, outputs and delivery
Every action is the same four parts: an instruction, a trigger, an output and a delivery. This page covers each, and is explicit about what works today and what is still being built.
Actions live under Actions in the left navigation.
The instruction
Write it as you would ask Aktis in chat. It is handed to the same assistant, with the same access to what your organization has ingested.
Summarise what enterprise customers said last week and flag anything that contradicts the current roadmap.
Two things make an instruction work better:
- Say what you want done, not what you want searched. "Summarise and flag contradictions" produces something to read. "Enterprise feedback" produces a list.
- Name the window if it matters. "Last week" and "since the last cycle" are understood, and without one the answer covers everything.
An instruction that Aktis cannot answer from what you have ingested will say so, rather than inventing an answer. If that happens repeatedly, the gap is in what has been connected, not in the wording.
Triggers
| Frequency | Fires |
|---|---|
| Daily | Every day at the time you set |
| Weekly | On the day and time you set |
| Monthly | On the day of the month you set |
| End of cycle | When your current cycle closes |
The time is your local time, in the zone your browser is in when you write the action. It is stored as a wall-clock time rather than a fixed instant, so "every Monday at 8:00" stays 8:00 through a daylight-saving change instead of drifting by an hour twice a year.
Two details worth setting expectations on:
- Actions run within five minutes of their time. An action set for 08:00 runs somewhere between 08:00 and 08:05. Actify checks for due actions on a short cycle rather than holding a timer per action, which is what keeps schedules reliable when they are edited or paused.
- Monthly on the 29th, 30th or 31st falls back to the last day of a shorter month.
End of cycle needs an active cycle with an end date, set under Settings → Signals. Without one there is nothing to fire against, and the action says so on its card rather than sitting silently.
Outputs
What the action produces, and therefore where the durable result is kept.
| Output | Produces | Kept in |
|---|---|---|
| Report | A written answer | Artifacts |
| Strategic artifact | A written answer | Artifacts |
| Alert | A heads-up with the detail behind it | The action's own history |
A report or artifact is filed as a draft artifact with the date in its title, so a weekly action builds a series you can read back through rather than twelve documents with the same name. It is a draft like any other: edit it, publish it, or delete it.
An alert deliberately files nothing. A daily heads-up saved as a document every day would fill your library with notices nobody opens — it lives on the action's history instead.
Delivery
Where the result is kept follows from the output above. Delivery is a separate question: who gets told.
| Channel | Status |
|---|---|
| Actify | Always on. The result appears on the action's history. |
| Available. Sent to the person who created the action. | |
| Slack | Not yet — see below. |
These are independent of each other and of the output. A report can be filed perfectly and still fail to reach an address, so the two are recorded separately: a delivery problem shows as a delivery problem, and does not make a working action look broken.
:::note Why Slack delivery is not available yet Actify's Slack app asks a workspace for permission to read the channels you select and to react to a message — deliberately not for permission to post. A bot that can post anywhere it can see is a much larger thing to grant, and we did not want to ask for it in exchange for a feature that did not exist yet.
Posting an action's result into a channel needs that permission. Enabling it means asking every connected workspace to reconnect and approve it, so it is a change we will make deliberately rather than quietly. :::
Running one by hand
Run now on any action queues it immediately. The answer usually arrives within a few seconds and appears in the action's history.
It works on a paused action too. Pressing the button is a deliberate choice, and Actify honours it without resuming the schedule — a paused action stays paused.
An action that is already running will not start a second time. A run costs a retrieval and a generation, so pressing the button again while the first is still working would buy a second answer to the same question; Actify says so and points you at the history, where the answer is on its way.
History
Every run is kept: when it ran, whether it was scheduled or run by hand, the full answer it produced, and a link to whatever it filed.
The whole answer is shown rather than a preview. For a report or an alert the answer is the output, and hiding it behind a link would hide the only thing the run produced.
When a run fails
A failed run is recorded with the reason and the action carries the message on its card. Failed runs are not retried automatically. Every retry is another model call against an instruction just as likely to fail again, so Actify tells you instead of quietly spending on it. Press Run now once the cause is fixed.
Two failures look similar and are not:
- The run failed — Aktis could not produce an answer. Nothing was filed.
- Delivery failed — the answer exists and was filed; sending it did not work. The result is still on the action's history and in Artifacts.
What actions cannot do yet
One thing you might reasonably expect, and where it stands:
Watching a metric. An action runs when the clock says so, not when a number moves. A watch — "tell me when negative feedback about billing doubles" — is a natural extension and is not built. The nearest thing today is a daily or weekly action whose instruction asks the question; it will tell you what changed, it just asks every day rather than only when something happens.
Editing, pausing and deleting
Editing changes an action in place — its name, instruction, schedule, output, delivery or tags. Everything it has already produced is kept, and so is its history: correcting a typo in an instruction does not cost you the record of what it has been telling you. A changed schedule takes effect from the next run; the slot the old schedule was pointing at is not honoured.
Pausing stops an action running and leaves everything else intact. Its history and anything it filed stay exactly as they are, and resuming it picks up the next scheduled slot — not the ones missed while it was off.
Deleting removes the action from the list. Reports and artifacts it already produced are kept: Actify does not delete work you may have built on.
Questions about actions: [email protected]