> ## Documentation Index
> Fetch the complete documentation index at: https://docs.quivly.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Actions

> An inbox of AI-drafted recommendations and workflow approvals. Review suggested next steps, edit drafts, and send emails, Slack messages, or calendar invites from one place.

The **Actions** page is your team's to-do inbox for AI output. Two kinds of items land here:

* **Recommendations** — AI-drafted next steps for an account, generated when a [signal rule](/product/signal-rules) fires (a health drop, a negative call, a usage change) or by an [agent](/product/agents).
* **Approvals** — agent runs paused at a Review step, waiting for a human decision.

The inbox has two lanes — **Strategic** (account-level recommendations) and **Operational** (approvals and system tasks) — with filters for status, a "My Accounts" toggle, and search.

## Working a recommendation

Each recommendation shows the customer, a Risk / Growth / Neutral chip, what changed (health bucket movement with per-category deltas) or what was noticed (the signals that fired), and a list of suggested next steps.

Steps are either:

* **Manual** — tick the checkbox when you've done it, or
* **Tool steps** — click **Review** to open the Execute modal, where the AI's draft is ready to edit and send.

The Execute modal adapts to the tool: an email composer (To/Cc/Bcc, subject, rich-text body, contact picker), a Slack message composer, a support ticket form, or a calendar event with attendees pulled from the customer's contacts. You choose the sender identity — the Quivly bot, your own account, or the account owner. Sending as yourself requires your own [Gmail](/integrations/communication/gmail) or Slack connection.

When you're done: **Mark complete**, or **Dismiss** with a quiet window (1 week to never) so the same recommendation doesn't come right back. Thumbs up/down feedback tunes future recommendations.

## Deciding an approval

Approvals show the workflow context, a snapshot of the data the agent saw, and the proposed change — which you can edit inline before approving. **Approve** resumes the workflow (with your edits); **Reject** sends it down the reject branch. Stacked approvals of the same kind can be decided in bulk. Every decision records who decided, when, and any note.

## FAQ

<AccordionGroup>
  <Accordion title="Where do recommendations come from?">
    From [signal rules](/product/signal-rules) you configure (Settings → Signals) and from agents. Nothing appears in the inbox unless a rule or agent you set up generated it.
  </Accordion>

  <Accordion title="Does anything send automatically?">
    No. Tool steps in a recommendation only execute when you review and send them. Agents can act autonomously only if you built them without a Review gate.
  </Accordion>

  <Accordion title="Can I edit the AI's draft before sending?">
    Yes — every field in the Execute modal is editable, and approvals support approve-with-edits.
  </Accordion>
</AccordionGroup>