Skip to main content
When you build a workflow that includes outward-facing or destructive actions — sending an email, deleting a file, posting a message to an external service — Atmet asks for your explicit confirmation before the workflow is activated. This design-time approval step ensures you always understand what an automated workflow will do on your behalf before it starts running unattended.

​
Why approvals happen at design time

Atmet separates the act of designing a workflow from the act of running it. The builder agent helps you plan and configure the automation; the runtime engine then executes it deterministically on every trigger event. Approvals fit into the design phase — once you have confirmed what a workflow will do, it runs automatically from that point forward without interrupting you again. This approach gives you the benefits of full automation (no manual steps on every run) while keeping you firmly in control of what your account is authorized to do.
Once you approve a workflow’s actions during the builder phase, subsequent runs execute unattended. If you need to change what the workflow does, edit it in the builder — the agent will ask for your sign-off again for any new or modified actions.

​
What triggers a confirmation request

The agent requests your explicit confirmation during workflow design for any action that falls into these categories:
  • Outbound messages — Sending emails from Gmail, posting messages to Slack, sending Telegram notifications, or publishing to social platforms
  • External record changes — Creating, updating, or deleting records in CRM systems (HubSpot, Salesforce), project management tools (Jira, Asana), or other connected platforms
  • Financial actions — Anything involving invoice data, payment records, or financial reporting
  • Destructive or irreversible actions — Deleting files, archiving records, or any action that cannot be easily undone
  • API credit consumption — Actions that use a third-party API key you have connected (such as generating images with your OpenAI key)

​
How the confirmation step works

1

Agent describes the action

As you build the workflow in chat, the agent identifies any actions that require your confirmation. It presents a plain-language description of exactly what it plans to do — for example: “This workflow will send emails from your Gmail account to contacts in HubSpot. Allow?”
2

Review the proposed action

Read the description carefully. The agent will specify which account or integration will be used, what data will be sent or modified, and under what conditions the action will fire.
3

Confirm or revise

Type your confirmation in chat or click Allow to approve the action. If anything looks wrong, tell the agent what to change — it will update the design and present the revised action for your review.
4

Workflow is activated

Once you have confirmed all actions in the workflow, the agent activates it. From this point, the workflow runs automatically on every qualifying trigger event without requiring further confirmation.

​
Inline confirmation in chat

Confirmation requests appear directly in the chat window as part of the workflow-building conversation. You do not need to navigate to a separate screen. The agent will display:
  • A plain-language summary of the action it wants to authorize
  • The specific integration and account involved
  • The conditions under which the action will fire
You can respond with natural language: “Yes, that’s fine” or “No — change it so it only sends to contacts tagged as active.” The agent interprets your response and adjusts the workflow accordingly.
If you want to modify the action before approving, describe the change inline: “Only send to contacts where the deal stage is ‘Proposal Sent’, not all contacts.” The agent will revise the design and ask you to confirm the updated version.

​
Setting standing approval rules in workspace memory

Rather than reviewing each action individually every time you build a new workflow, you can define standing authorization rules in workspace memory. The agent applies these rules automatically when designing workflows, surfacing them as pre-approved patterns. Example rules to store:
  • “Sending emails from Gmail to HubSpot contacts is approved for all sales workflows.”
  • “Never delete records from Salesforce automatically — always require explicit confirmation.”
  • “Posting summaries to the #reports Slack channel is pre-approved for reporting workflows.”
To save an approval rule, tell the agent: “Remember this as an approval rule” after describing the condition, or navigate to Memory → New Memory and set the category to Approval Rule.
Standing rules stored in workspace memory inform how the agent presents options during workflow design across your entire workspace. Review your active approval rules periodically to make sure they reflect your current policies.

​
Who manages approvals

By default, any workspace member with edit access can confirm workflow actions during the builder phase. You can restrict which team members are authorized to activate certain types of workflows by including that context in your workspace memory:
  • “Only workspace admins can activate workflows that send emails to external contacts.”
  • “Workflows that modify Salesforce records must be approved by the sales operations team.”
The agent surfaces these rules during the builder conversation and attributes the required confirmation to the right person.

​
Editing a workflow after activation

If you need to change what an already-active workflow does, open it in the builder by clicking Edit in Chat. The workflow is paused automatically during the edit session. Once you describe your changes, the agent will present any new or modified outward-facing actions for your confirmation before saving and reactivating the workflow.
Any time you add or modify an action that sends data outside your workspace or makes an irreversible change, the agent will ask for your sign-off before the updated workflow goes live. This applies even if you previously approved an earlier version of the same workflow.

​
Approval rules and workspace memory

Approval rules work hand-in-hand with workspace memory. The policies you store in memory shape what the agent flags for confirmation during workflow design, ensuring your team’s standards are applied consistently across every automation your workspace builds. See Workspace Memory for more detail on creating and managing memory items.