Actions, WhatsApp and Email

An action is what Qlar does with the result once the task is finished. There are three, and an automation can use any of them together.

ActionShown to users asWhat it does
Create conversationNew conversationPuts the result in a new conversation on this agent and notifies the user.
Send WhatsApp messageWhatsApp messageSends the result to a phone number as a WhatsApp message.
Send emailEmailSends the result to an email address.

Which of them users may choose is set by Allowed Actions on the policy. There is no SMS or other channel; if a user asks for one, the agent says it is not available.

The KG-2041 check uses Create conversation. In the preview Atlas wrote it as "Create a new conversation in Qlar and notify the user", because the operations lead asked to be given a summary and named no other channel.


Before you start

  • Read Create an Automation.
  • To deliver results by WhatsApp, the agent needs its own WhatsApp number on the public or private server in its Published configuration. See Connect to WhatsApp.
  • Email needs nothing to connect, but Send email must be ticked under Allowed Actions on the policy. A policy saved before email existed does not have it.

Create conversation

The result becomes a new conversation on the same agent, owned by the user who created the automation.

Every weekday 07:30 β†’ KG-2041 delayed or on hold? β†’ Summarize the project β†’ Create conversation β†’ Notification
What happensDetail
A conversation appearsIn the sidebar list, marked as unread and with an automation icon so it is recognisable.
A notification appearsOn the bell in the top bar, with an unread badge. Opening it goes to the conversation.
The user can continueIt is a normal conversation. The agent knows which automation produced it and what its task was, so a follow-up such as "Who do I contact to speed up the CFO approval?" just works.

A conversation is created only when the run got as far as the action. Runs that failed or whose condition was not met leave nothing in the list.

Send WhatsApp message

The result is delivered as a WhatsApp message to the number stored in the automation. A typical Atlas use: a project manager asks, "Every day at 16:00, send the KG-2041 status to the site supervisor's WhatsApp, 0812…".

PointDetail
Sent fromThe agent's own WhatsApp number: the channel on the public or private server in its Published configuration. Meta Cloud, Coster and Qontak can only start a conversation with an approved message template, so they cannot send it; without a channel that can, the run records "No WhatsApp channel that can send".
RecipientAny valid phone number the user gives, such as a site supervisor's number. There is no verification that the number belongs to the user. The channel's contact rules still apply: in private mode, only numbers on its list receive results.
RepliesA reply reaches the agent like any other WhatsApp message to its number. The message itself ends with the agent's name, the automation's name and a link to the user's Automations page in Qlar.
How the number is readQlar normalises it to international format (a leading 0 becomes the country code) and writes it back in the preview, for example +62 812…, so the user can check it. A number that cannot be valid is refused.
If no number is givenOn WhatsApp, the number the user is chatting from is offered. Otherwise the agent asks.
Never sent twiceIf Qlar cannot tell whether a message arrived, it does not try again, so nobody gets a duplicate. The history shows "outcome unknown".

Because there is no verification, two protections exist:

  • The user sees the number before agreeing, and must confirm it.
  • Limits. The policy's Maximum WhatsApp Messages per Day caps the whole agent, and the agent's number sends automation results to the same number at most 10 times an hour, and at most 30 times an hour in all; a message over either limit waits for the next hour.

Consider the risk for your agent. A typo sends a private result, such as a project's budget, to a stranger, and an agent open to the public could be used to send unwanted messages β€” from your agent's own number, whose standing with WhatsApp they affect. If that is not acceptable, allow only Create conversation on the policy, set a low daily WhatsApp limit, or keep the channel in private mode.

Send email

The result is delivered as an email to the address stored in the automation. A typical Atlas use: the operations lead asks, "Every Monday at 8, email me last week's summary of all projects that are delayed or on hold."

PointDetail
Sent fromThe agent's own automation address, {agent}-automation@qlar.ai, with the agent's name as the sender. For Atlas, whose chat address is app.qlar.ai/atlas-ops, that is atlas-ops-automation@qlar.ai, shown as Atlas. There is nothing to connect or set up.
RecipientThe address the user gives, such as a colleague's. If they give none, the email address of their own Qlar account is used, and the preview says so. There is no verification that the address belongs to the user. One address per automation.
SubjectThe automation's name and the agent's name, for example "KG weekly delay summary Β· Atlas".
RepliesNobody reads the automation address, so a reply goes nowhere. The email links to the user's Automations page in Qlar (/{agent}/automations, for Atlas https://app.qlar.ai/atlas-ops/automations); to follow up, the user continues there.
FormattingTables, lists and emphasis are kept, unlike WhatsApp. A chart or canvas cannot be shown in an email, so it is replaced by the note "(grafik tersedia di Qlar)" ("chart available in Qlar"); the user can see it in Qlar.
Never sent twiceIf sending fails, Qlar does not try again, so nobody gets a duplicate. The history shows why it failed, and the next run sends as usual.

As with WhatsApp, there is no verification, so the same two protections exist:

  • The user sees the address before agreeing, and must confirm it in their next message.
  • Limits. The policy's Maximum Emails per Day caps the whole agent (default 100), and a user's automations may send at most 10 emails an hour to one address. That hourly limit is checked when the automation is created or changed, so an automation that could exceed it is refused with an explanation instead of failing later.

Consider the risk for your agent. A typo sends a private result to a stranger, and an agent open to the public could be used to send unwanted email under your agent's name. If that is not acceptable, leave Send email unticked on the policy or set a low daily email limit.

Using more than one

An automation may use any combination: for example "…create a conversation for me and also email it to me", or a WhatsApp message to the site supervisor plus an email to the operations lead. Each action is recorded separately in the history, so a failed WhatsApp message or email does not hide that the conversation was created, and the reverse.

When an action is not allowed any more

Actions are checked on every run. If you remove one from the policy, automations that use it skip it from then on. If none of an automation's actions is allowed any more, it is stopped with an explanation.


Check the result

  1. In the automation preview, the Action line names the channel the user wanted. For WhatsApp, the number is written in international format; for email, the address, with a note when it is the user's own account address.
  2. After a run with Create conversation, a new unread conversation appears in the chat sidebar and the bell shows a badge.
  3. After a run with Send WhatsApp message, the recipient gets the message, and the automation's history shows the action as sent.
  4. After a run with Send email, the email arrives from the agent's -automation@qlar.ai address under the agent's name, and the history shows Send email: Sent. If it is not in the inbox, check the spam folder.

Next steps