Triggers

A trigger starts an agent when something new happens in a connected app. Each run opens a new chat in your Chats, titled with the trigger’s name followed by the item’s title. Triggers live under When This Happens on the Schedules page (/schedules).

What can start a trigger

Trigger	Needs	Starts when
When a meeting ends	Fireflies	A meeting’s transcript and summary become available in a shared area.
When an email arrives	Gmail	A message arrives in a shared label. Mail you sent yourself never starts one.
When a file is added	Google Drive	A file appears in a shared folder, by being added or moved in. Edits to a file already there do not.

A trigger only sees what is shared with the brain: the company’s connection and your own. An item from someone else’s own connection never reaches your triggers.

Triggers start only for new items. When an app is first connected or a folder or label is first shared, the brain reads everything already there without starting any trigger.

Making a trigger

Pick one of the ready-made triggers:

• When a Meeting Ends, Draft the Follow-up sums up the decisions and action items, then drafts a follow-up email to the external attendees as a Gmail draft for your approval.
• When a Client Emails, Summarise and Draft a Reply says what the email needs from you and drafts a reply in your voice as a Gmail draft for your approval.
• When a File Lands in a Folder, Summarise It sums up a new file and anything in it to act on.

A ready-made trigger is offered only while its app is connected. Otherwise the card says which app to connect, with a link to Integrations.

Each trigger has a name, the agent that answers, what to do each time, and optional filters. Leave a filter blank to run on every item:

• a meeting’s title, or a domain one of its attendees is at, such as client.com
• an email’s sender (an address, or a domain for everyone there), its subject, or the label it arrived in
• a file’s folder, or words in its name

What a run can do

A run uses your model choice, your department access, your connected apps and the same tools as a chat you typed, and its cost counts toward your budget and your organization’s credits. While the organization has no plan or too few credits, a run is skipped with that reason, and one that runs out partway stops with it. The agent is told what happened (the item’s title, id, sender or attendees, and link) and reads the item with its own tools.

Nothing is ever sent for you. When a run wants to save an email draft or a document, it leaves an approval card in its chat and finishes. Nothing is written until you approve it there.

Managing triggers

Each trigger shows its state in words with an icon: On with when it last ran, Paused, Running now, or Failed with the reason. Open the Result goes to the last run’s chat.

• The switch beside its name turns it On and Off. Items that arrive while it is off never run, and the row says Paused.
• Run Test runs it once on the newest item that matches it, even one it already ran on, so you can see what it does.
• Edit changes anything except what it listens for, and the trash button deletes it after asking.

Limits

• Each person can have up to 20 triggers.
• An item runs a trigger once at most. It never runs twice, even if the app reports it again.
• One trigger runs on at most 6 items in an hour. Items past that are passed over, and the trigger says so.
• A run that fails is recorded with the reason and is not tried again.
• An item that is more than two days old when the brain first sees it starts nothing.

Admins can change these limits; see environment variables.

Triggers are private. Nobody else, admins included, sees or changes yours.