---
title: "Review, listen to, and coach your calls"
description: "Call History end-to-end: transcripts, recordings, live listen-in, and whispering instructions to your agent mid-call."
---

Every call — inbound, outbound, simulation, or widget session — lands
in **Call History** (`/dashboard/call-history`), newest first. This is
where you find out what your agent actually said, and where you can
step into a live call and steer it.

## Find the call

The list filters by **Date**, **Type**, **From**, **To**, **Agent**,
**Outcome**, **Needs attention** (calls flagged for a follow-up) and
**Business hours**, and the **Columns** button controls what each row
shows. The
**Outcome** column carries the label the call report chose (the agent's
own list, or Success / Failure), **Summary** the one-line summary, and
**Needs attention** the follow-up reason. Phone numbers take labels (e.g.
"Acme Corp") so recurring callers are recognizable at a glance. Click any
row to open the call.

## Set up call reports

Open the agent builder and expand **Reporting**. Next to **Outcome
labels**, choose **Edit**. Use **Suggest from prompt** or write your own
labels, and give each one a description under **When to use it**. Keep
2–12 labels, or choose **Replace all → Success / Failure** to restore the
defaults.

Under **Summary rules**, choose **Edit** to add instructions for the
report's wording, then **Done**. Test the draft with a browser call that
includes a conversation. After grading finishes, open the call in Call
History and check its outcome, summary and follow-up flag. Choose
**Deploy** to apply the changes to production calls. For API setup, see
[agent reporting fields](/api-reference/agents#deployed-fields).

## Label a phone number

Any number in the **From** or **To** column takes a label ("Aetna", "Main
line", "Jack Morrison"): hover the number and choose the pencil. Labels are
shared by everyone in the organization. They appear in Call History, on
Home, in the Client Portal Calls table and in portal CSV exports. The
**From** and **To** filters search labels too: type a label to pick every
call from that party.

## Read the conversation

The transcript shows every turn, word for word — including the
agent's tool calls and their results. It's the fastest way to spot
where the prompt needs tightening. Alongside it:

- **View** the exact prompt snapshot the call ran with (and download
  it) — what the agent was told, not what the agent is told today.
- **Create test case** — turn this conversation into a reusable
  [simulation scenario](/guides/simulate-a-call).

## Hear it back

Ended calls have a recording — scrub the waveform to replay any
moment, or download the audio.

## Listen in — live

While a call is **in progress**, the recording spot offers **Listen
in** instead. Click it and you join the call silently: you hear
everything, and nobody on the call hears you. Use it to watch a new
prompt take its first real traffic, or to supervise a
[campaign](/guides/outbound-campaigns) as it runs. The
[**Live** page](/guides/monitor-live-calls) shows every in-progress
call across the org, one click from the same experience.

## Whisper — coach your agent mid-call

Once you're listening, a whisper bar appears. Whatever you type goes
**straight to your agent**: the caller never hears it, and the agent
follows the instruction live, mid-conversation.

Try one on your next simulation: `wrap up the call politely`. Or in
production: `the caller is a VIP — offer the expedited option`,
`stop offering refunds`, `ask for their order number`.

Whispers are first-class steering — private operator guidance of up
to 500 characters per message. They show up in the call's internal
history but never in the customer-facing transcript, and the agent
treats them as instructions, not conversation.

<Tip>
  Listen-in and whisper are also available over the API — see the
  [calls reference](/api-reference/calls) — so you can build your own
  supervision console.
</Tip>

## Close the loop

Spotted a problem? File an issue on the call — issues carry severity
and status and queue up for triage on the
[Issues](/guides/issues) page. Recurring themes across many
calls are what [Reports](/guides/reports) and
[Observability](/guides/observability) are for.

## Delete a call

Owners and admins can permanently delete calls. Open a call and choose
**Delete call**, or tick several calls in the list and choose **Delete** in
the selection bar. Deleting removes the recording, transcript, summary,
extracted data, grades and issue reports, and cannot be undone. It also
deletes any [validation examples](/guides/validation-sets) made from the
call; the confirmation says how many. A call that
is still in progress, or whose billing is still being settled right after
it ends, can't be deleted yet; try again in a few minutes.

Your billing history keeps the charge for each deleted call, without any of
its content. To delete calls from your own systems, use the
[calls API](/api-reference/calls#delete-a-call).

---

## Next steps

<CardGroup cols={2}>
  <Card title="Simulate a call" icon="flask" href="/guides/simulate-a-call">
    Practice listening in and whispering on an AI caller first.
  </Card>
  <Card title="Calls API reference" icon="code" href="/api-reference/calls">
    Transcripts, history, audio, grading, and export over REST.
  </Card>
  <Card title="Issue reports API" icon="triangle-exclamation" href="/api-reference/issue-reports">
    File and query issues programmatically.
  </Card>
  <Card title="Fix the prompt, redeploy" icon="robot" href="/guides/build-an-agent">
    Every deploy is versioned — iterate fearlessly.
  </Card>
</CardGroup>