Skip to main content
The pending actions queue is where the Sales AI Co-pilot stages every change before it touches the world — every email it drafts, every CRM update it proposes, every task it suggests. You review, then decide. Nothing sends or updates without your explicit approval. Find the queue at /dashboard/crm/memory/ under the Pending Actions tab.

​
Why approval-first

The Co-pilot is good. It’s not perfect. A draft email might miss context you have but the model didn’t. A proposed deal-stage change might be premature. A suggested task might duplicate one you already created. The queue gives you a moment to review before any of it goes out. It also gives the Co-pilot something to learn from. Approvals, edits, and rejections train its sense of your preferences over time.

​
Action categories

Email drafts

Outbound emails — replies, follow-ups, scheduling. The most common category.

CRM updates

Property changes, deal-stage moves, contact tags, lead-score adjustments.

Tasks

New tasks pulled from meeting transcripts or proposed proactively.

Notes

Summary notes appended to a contact or deal’s timeline.

Sequence enrollments

Adding a contact to a multi-step nurture sequence.

Calendar bookings

The Co-pilot proposes a meeting time when a prospect asks for one.

​
Reviewing an action

1

Open the queue

/dashboard/crm/memory/ > Pending Actions tab.
2

Read the action

Each item shows what would happen, the rationale (“Why the Co-pilot suggested this”), and the affected records.
3

Decide

Approve, edit, or reject.

​
The three decisions

The action runs immediately. For emails, the message sends. For CRM updates, the change writes. For tasks, the task is created on the right record. The action moves to the History tab with timestamp.

​
Bulk actions

For high-volume queues, you can bulk-approve or bulk-reject. Common patterns:
  • Bulk approve all “task creation” — Tasks pulled from a meeting transcript are usually safe; you don’t need to approve each one individually.
  • Bulk approve “lead-score updates” — Score adjustments are non-destructive (you can override) and rarely need per-item review.
  • Bulk reject “sequence enrollments” — If you decided to pause a nurture campaign, reject all queued enrollments at once.
Don’t bulk-approve email drafts unless you’ve thoroughly reviewed them. A bad draft sent at scale is much worse than a bad draft sent once.

​
Auto-approve rules

You can configure the Co-pilot to auto-approve certain low-risk action categories so they don’t pile up in the queue. Common safe categories:
CategoryAuto-approve risk
Adding tags to contactsVery low — easily reversible
Updating last-activity timestampsVery low — informational
Creating tasks (no due date past 7 days)Low — visible in your task list
Appending notes to recordsLow — additive, not destructive
Lead-score adjustmentsLow — non-binding signal
Sending emailsNever auto-approve by default
Deal-stage movesDon’t auto-approve (impacts forecast)
Sequence enrollmentsDon’t auto-approve (impacts contact experience)
Configure auto-approve from the Co-pilot settings. Auto-approved actions are still logged, so you can review the audit trail and revert if needed.

​
Audit log

Every action — approved, edited, rejected, or auto-approved — is logged in the History tab with:
  • Action category and content
  • Decision (approve / edit / reject / auto-approve)
  • User who decided (or “system” for auto-approve)
  • Timestamp
  • For edits: a diff of what changed
  • For rejections: the reason
Filter by date range, decision, and category. Exportable as CSV for review.

​
Notifications

You don’t need to live in the queue. Configure notifications to surface high-priority actions:
  • In-dashboard badge — count of pending actions next to the Co-pilot icon
  • Email digest — daily or weekly summary of items waiting for review
  • Slack DM — for high-priority items (e.g., replies to inbound prospect emails) within minutes
  • Mobile push — via the Hiveku iOS/Android app

​
Time-sensitive actions

Some actions have implicit deadlines — a reply to a prospect’s question, a meeting booking, a confirmation. The Co-pilot tags these as time-sensitive and can:
  • Push them to the top of the queue
  • Send a stronger notification (Slack DM + push)
  • Optionally auto-send if you don’t review within a window (e.g., “auto-send replies to scheduling questions if I don’t review within 30 minutes”)
The auto-send behavior is opt-in per action category.

​
When to reject vs edit

  • The action shouldn’t happen at all
  • The Co-pilot misread the situation entirely
  • You’d prefer no message over an imperfect one
  • Privacy or compliance issue
Rejections give the strongest training signal — they tell the Co-pilot “don’t do this kind of thing.” Edits give a softer signal — “this kind of thing, but tweaked.”

​
API

# List pending actions
curl https://api.hiveku.com/api/sales/ai/pending \
  -H "Authorization: Bearer hk_live_xxx"

# Approve an action
curl -X POST https://api.hiveku.com/api/sales/ai/pending/act_01HQ/approve \
  -H "Authorization: Bearer hk_live_xxx"

# Reject with a reason
curl -X POST https://api.hiveku.com/api/sales/ai/pending/act_01HQ/reject \
  -H "Authorization: Bearer hk_live_xxx" \
  -d '{"reason": "Too pushy for this prospect"}'

# Edit and approve
curl -X PATCH https://api.hiveku.com/api/sales/ai/pending/act_01HQ \
  -H "Authorization: Bearer hk_live_xxx" \
  -d '{"body": "Updated email body...", "approve": true}'

​
Troubleshooting

Auto-approve safe categories (tags, timestamps, notes), set up notifications for only high-priority items, and set the Co-pilot’s “drafting verbosity” lower in settings so it stages fewer actions.
Add a rule: “Never draft replies to no-reply addresses” or “Don’t propose discounts above 10%”. The rule overrides the underlying behavior. See Skills and rules.
Open the History tab, find the action, and click “Revert” if it’s reversible (most CRM updates are). Then either move that category off auto-approve or add a rule narrowing what’s allowed.

​
Next steps

Using the Co-pilot

Day-to-day workflow.

Enable the Co-pilot

First-time setup.