DoubleX AgentesDocs

Guides › Start here

Alerts (proactive agent)

The proactive agent follows your email, calendar and what the assistant did for you, and only speaks up, inside the conversation, when it changes what you should do. Everything stays in Alerts.

Alerts is where the proactive agent in DoubleX Agentes works. It does not wait for a question or need a trigger: connect your email and calendar and it starts following along on its own. It does not repeat every email or calendar change, which you already see in your apps. It puts together what it knows and only speaks up when that changes what you should do, like a chief of staff:

One detail about the 4 pm meeting: Guilherme mentioned today that he has limited availability after 3 pm, and neither he nor Antônio has answered the invite. Want me to check with Guilherme?

The message lands in your conversation with the assistant (the same one where it scheduled the meeting, or the day's alerts conversation), and you reply right there. On the phone it arrives as a notification that opens the conversation. The history stays in Alerts: open it from the Alerts block on the Home page (the unread count shows next to Home in the sidebar).

How it works#

  1. Sensors, no AI. Every few minutes, the agent compares your email and calendar with the previous check and finds what changed: a new email, a meeting created, moved or canceled, an unanswered invite, overlapping times. Between 15 and 75 minutes before each meeting with other people, it also checks the meeting.
  2. Thinks with the whole context. When something is new, the AI looks at the next 24 hours of your calendar (including who answered each invite), recent emails, what the assistant did for you in the last few days (meetings scheduled, emails sent) and what the agent already said. Most of the time the conclusion is to stay quiet.
  3. Speaks once, connecting the dots. When it is worth saying, it is one message per situation: a reschedule, the clash it creates and the pending invite become one message, not three. Newsletters, automated notifications, everyday overlaps (dailies, long blocks) and anything it already said are left out.
  4. You reply in the conversation. "Yes, ask him" and the assistant does it. The alert also has Accept and Decline for invites and Open for the email or event.
  5. Learns. Don't tell me about this mutes that sender or topic, and what you dismiss weighs on later decisions. In Settings › What matters to you, write your priorities.

What the agent follows#

SourceWhen it speaks up
Gmail, OutlookWhen an email affects one of your commitments, replies to something the assistant sent, or someone is waiting on a decision and delay has a cost.
Google Calendar, Outlook calendarWhen a change gets in the way of something that matters, and before meetings with other people if there is something useful to know (who has not confirmed, a related email, a change).
Conversations with the assistantFollows what it did for you (meetings scheduled, emails sent) and comes back in the same conversation when something changes.
Any connected appWhatever you define in a monitor.

Your email and calendar accounts are included automatically. In Settings, turn off the ones you do not want followed, turn on quiet hours (urgent ones still come through) and urgent alerts by email.

Daily summary#

Every day at 8 am (change the days and time in Home › Daily summary), your day arrives in 30 seconds: as a notification if you turned them on, or by email. If the notification reaches no device, it goes by email. It is the one predictable alert of the day.

It reads your email and calendar when connected, files changed in Google Drive and what happened on DoubleX: conversations, routines that ran or failed, actions the assistant took and notes saved. Without email connected it still goes out, with what happened on the platform. If you are not using DoubleX yet, it brings ideas to start today: ask the assistant what it can do, hand over a task, create a landing page or talk to Lumi. It brings:

  • The essentials of the day in two or three sentences.
  • What I noticed: one observation that connects the dots across email, calendar and what the agent has been following.
  • Highlights with Handle it: one tap opens the assistant with the request ready (for example, a draft reply).
  • To make tomorrow better: one suggestion, such as automating something that repeats, which the assistant sets up in one tap.
  • Scoreboard for the last 24 hours: emails followed, interruptions and actions the assistant took for you.

Dismissing a highlight teaches what does not need to come back. If the AI is down or credits run out, a quick summary goes out instead, and Generate again makes the full one.

Monitors#

A monitor checks a connected app from time to time and alerts you when a condition happens. Examples: a campaign's CPA, stalled deals in the CRM, declined payments, Slack mentions or a database metric.

  • It only calls read-only tools: it never sends, posts, changes or deletes anything.
  • AI only runs when the result changes, and it doesn't alert twice about the same situation.
  • With Alert and run an agent, the chosen agent acts when the monitor fires, for example analyzing the campaigns and suggesting what to pause.

Create one under Alerts › Monitors › New monitor: pick the connection, the query and the condition, and use Test query to see the data. Or ask the assistant, for example "tell me if any campaign's CPA goes over $50", and it sets up the monitor.

Rules#

A rule makes an agent act on its own when an email arrives or something changes in the calendar that matches the condition. For example: "when a customer email asking for a quote arrives, the sales agent prepares the proposal and leaves a draft". The alert links to the agent's conversation.

From ChatGPT and Claude#

Through the MCP server, ChatGPT and Claude read your alerts (list_radar_alerts) and create monitors from the chat (list_monitor_tools and create_monitor).

Cost and limits#

  • Alerts only uses AI when there is something new (an email, a calendar change or a meeting about to start), and usage shows in Usage and credits as "Alerts".
  • Each person has a daily cap of AI checks. After it, and also without a plan or credits, only meetings moved or canceled in the next few hours still trigger an alert, with ready-made text, and the Alerts page shows the reason.