Notifications

The notifications system keeps you informed about relevant events in real time.

The notification bell

There is a bell icon in the top-right corner of the screen. The number on it tells you how many unread notifications you have.

Click the bell to open the notifications panel.

Notification types

Type	When it is generated	Who receives it
New task	A task is assigned to you, including when you claim it yourself	whoever received the task
Task delegated	Someone delegates a task to you	whoever received the delegation
Process error	An instance fails on an automated step	whoever started the instance
AI agent failure	An agent does not finish a step's analysis	whoever started the instance and whoever completed the step that led to the AI step; if neither is an Administrator or Manager of the tenant — including when neither is a person (started by API, by e-mail or by a timer) — the tenant's Administrators and Managers too
Robot failed or timed out	A robot execution ends badly	whoever triggered the execution
Message from an agent	An AI agent sends a notice during the process	the user the agent names
Overdue task	An open task passes its due date	the assignee; when the task has none, its candidates and whoever holds the candidate role
Instance over target	A running instance passes the process completion target	whoever started the instance and whoever holds its open tasks
Delay escalated	A task or instance is still late past the process escalation time	every active Manager and Administrator of the company — also by e-mail

The AI agent failure notice arrives once per step: if the platform tries again and fails again, nobody gets a second notice. Clicking it opens the instance, with the failed step marked.

Only Administrators and Managers can retry or dismiss the failure, and the notice text follows that: whoever operates the tenant reads that the step can be run again (and where Failed Jobs is); everyone else reads that the failure shows on the instance and that an Administrator or Manager has been told and decides whether the step runs again.

Deadlines: overdue and escalation

The bell warns when a deadline passes, not before. A task that is due soon only gets the "due soon" badge on the list (24 hours ahead) and shows up in the daily digest — there is no "almost due" notice.

1. Overdue — the task passed its due date, or the instance passed the process completion target. Whoever holds the work gets a notification (table above). It arrives once: nobody gets the same notice again every minute.
2. Escalation — when the process sets escalate after and the delay lasts that long, every Manager and Administrator of the company gets Delay escalated in the bell and by e-mail (when SMTP is configured). To stop the e-mail, turn off Receive escalated delays by e-mail in your profile; the bell notice still arrives.

If someone moves the due date later, the delay is reset and the notice applies to the new deadline. Test runs never raise a deadline notice.

The notice is for a deadline that passes while the platform is watching. Whatever was already late when the watch began — on the first start of the platform with deadlines, or when someone shortens a process's target or escalation — enters silently as late (or escalated) and shows on the Monitor, but sends no notification and no e-mail. That way a target change never drops hundreds of notices into anyone's inbox at once. The target and the escalation are set per process, on the definition's Config tab — see Definitions. Deadlines count wall-clock hours: weekends and holidays count like any other day.

WARNING

Deadlines are checked every minute, so a notice can arrive up to a minute after the deadline.

Completion raises no notification

A completed process generates no notification. An instance that finishes well fires a webhook, if one is registered for instance.completed — see Webhooks.

The daily digest e-mail

Once a day you get one e-mail with the work waiting for you — not one per task. A busy process alone would produce hundreds of messages a day; the digest makes the volume depend on the number of people, not the number of tasks.

If you belong to more than one company, it is still one message, with a section per company. Each section carries, in this order:

1. Waiting the longest — oldest first, with how many days they have waited
2. Overdue — what is past its due date, kept apart from what is merely old
3. New in the last 24 hours
4. That company's total

Every row links straight into the task. At most five rows per block: the total says the rest.

The digest is always your own work, whatever your role. A manager or an administrator sees the whole company on the Tasks screen, which is what that screen is for; the e-mail is not. It carries what is assigned to you, or a candidate for you and your groups — a "what is waiting for you" summary with the whole company inside it would be exactly the noise it exists to avoid.

On a day with nothing waiting, nothing is sent. A company with no tasks produces no section, and someone left with no sections gets no message — a daily message with nothing in it would lead you to ignore the ones that do have something.

The hour is the platform's (07:00 by default, adjustable by the Super Admin in the server configuration), not your own timezone. To stop receiving it, turn Receive the daily digest off in your profile.

The digest is not for urgency

A step that needs somebody within two hours is not served by an e-mail arriving tomorrow morning, and raising the frequency does not fix it — it only rebuilds the volume problem. For that, watch Instances or your task inbox.

Marking as read

• Click a notification to mark that one as read
• Use the Mark all as read button to clear them all at once

Real-time notifications

Flowi Agentic uses a real-time connection (WebSocket). New notifications appear on their own, with no need to reload the page, and the bell counter updates the moment something arrives.