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

# Review and approve what Hive proposes

> How to read a Hive proposal in the Review decision panel, choose Approve & run now, Decide later or Do not run, and who can approve which actions.

Hive prepares actions — a reply to a customer, an invoice reminder, a ticket update — but by default it does not
carry them out until a person approves. The **Review decision** panel is where that happens. Its promise is
printed at the top: "Nothing runs until you approve it."

## Where approvals appear

There is no separate approvals inbox. Proposals show up where the work started:

* **Signals** — a signal with a prepared action shows a review button: **Review reply**, **Review chaser**,
  **Review nudge** or **Review message** when Hive has drafted something, **Review action** otherwise, and
  **Review actions** when there are several. Approvals raised by workflow runs also appear here.
* **Ask Hive** — actions you request in a chat are reviewed inline in the conversation, and pass the same checks.

Old links to `/approvals` open the Signals page.

## Read the proposal

The panel is organised so you can check the exact action before you decide:

<AccordionGroup>
  <Accordion title="Chips: Needs your approval and the risk level">
    The risk word comes from the action's blast radius: **Low risk** (B0), **Medium risk** (B1–B2) or **High risk**
    (B3–B4). See [Autonomy and blast radius](/approvals/autonomy-and-blast-radius).
  </Accordion>

  <Accordion title="What happens next">
    Every step of the plan, each marked "runs when you approve" or "runs automatically". Hive attempts only the steps
    shown; completed actions are recorded in the [Audit log](/approvals/audit-log) and the signal is marked handled.
  </Accordion>

  <Accordion title="Proposed message">
    The exact text that will be sent, when the action sends something. If Hive could not read the conversation
    behind it, the panel says **Conversation context unavailable — review the source before sending.**
  </Accordion>

  <Accordion title="Risk and evidence">
    **Why this asks first**, **Predicted blast radius**, **Rollback confidence** and **If you do nothing** — the
    cost of not acting.
  </Accordion>

  <Accordion title="Account picker">
    When more than one of your connected accounts could send the action (for example two mailboxes), choose which
    one.
  </Accordion>
</AccordionGroup>

Before approving, check:

1. the intended outcome;
2. the exact people or records affected;
3. any money, customer, access or reputation impact; and
4. whether the action can be reversed.

<Warning>
  Approving means approving the exact plan shown — target, amount, scope and wording. Check those, not just the
  headline.
</Warning>

## Decide

| Button | What happens |
| - | - |
| **Approve & run now** | Hive carries out the plan, subject to the workspace's safety checks. |
| **Decide later** | The proposal is set aside and comes back, like a snoozed signal. |
| **Do not run** | The proposal is rejected. Nothing is sent or changed. |

Drafts cannot be edited in the panel. If you want different wording or a different action, choose **Do not run**
and ask Hive for a replacement. A denial also teaches Hive: it proposes a rule that this shape of action should not run automatically, which takes effect once it is promoted (see [Learned behaviours](/brain/learned-behaviours)).

Approval is bound to the exact parameters you saw. If anything in the plan changes, it needs a fresh approval.

### Proposals expire

A proposal waits **7 days**. The panel counts down ("This request expires in …"). After that it can no longer be
approved, so a stale action cannot run on out-of-date information.

## Who can approve what

Only **Operators** and above can decide on proposals. The panel also routes high-impact actions to more senior
roles: when an action's blast radius is above your role's level, it says the action "is above your access" and
names the role it routes to.

| Role | Highest blast radius the panel routes to it |
| - | - |
| Owner | B4 |
| Admin | B4 |
| Department lead | B3 |
| Operator | B2 |
| Analyst | Read-only — cannot decide |

Analysts see the full reasoning ("Read-only — Analysts can see exactly why Hive proposed this, but only Operators
and above can act on it.").

Role limits are one check among several. Confidence, workspace policy, spend caps and the kill switch can still
require review or block an action. See [Roles and permissions](/admin/roles-and-permissions).

## When the kill switch is on

While an Owner or Admin has engaged the kill switch, the panel says "Kill switch is on — approvals are frozen
company-wide." Nothing can be approved until someone resumes autonomy. See
[Safety and control](/admin/safety-and-control).

## After you approve

The action runs through Hive's executor, which holds the connector credentials — the AI model never does. The
result is written to the [Audit log](/approvals/audit-log) with the full chain from signal to plan to decision to
result. Reversible actions can be undone from there.

## Related

<CardGroup cols={2}>
  <Card title="Autonomy and blast radius" icon="stairs" href="/approvals/autonomy-and-blast-radius">
    When Hive may act without asking.
  </Card>

  <Card title="Audit log" icon="receipt" href="/approvals/audit-log">
    Receipts and undo.
  </Card>

  <Card title="Triage signals" icon="list-check" href="/signals/triage-signals">
    Act on, snooze or dismiss signals.
  </Card>

  <Card title="Safety and control" icon="shield-halved" href="/admin/safety-and-control">
    Kill switch, policy and spend caps.
  </Card>
</CardGroup>