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

# Observability

> Runtime metrics for your agents — volume, success rate, failures, and the calls behind every number.

**Observability** (`/dashboard/observability`, under Quality & Testing)
is the operational view of your calls: how many, how successful, what's
failing, and which agent is responsible. Where [Reports](/guides/reports)
answers questions in prose, Observability gives you the live numbers.

## Scope the view

Two filters at the top drive the whole page:

* **Time window** — last hour, 6, 24, or 72 hours, 7 days, or 30 days.
  Charts bucket hourly for short windows and daily from 48 hours up.
* **Agent** — all agents or a single one.

For very busy windows, detailed metrics are sampled from the most
recent calls; a notice tells you when that happens and how many calls
the sample covers.

## What you get

**Summary cards** — Calls, Success Rate (based on AI grading, with
infrastructure errors excluded from the denominator), Unsuccessful, No
Conversation (caller disconnected early), Errors (infrastructure
failures), In Progress, and Average Duration.

**Charts** — call volume over time and the success-rate trend, bucket
by bucket.

**Call status breakdown** — successful / unsuccessful / no conversation
/ error as proportions of the window.

**Agent performance** — a per-agent table of calls, failed calls,
issues raised, and average grade. Sort out which agent is dragging the
success rate before you start editing prompts.

**Recent failed calls** — each with its end reason ("Caller hung up",
"No answer", "Call failed", …) and an **Open call** button straight into
the transcript and recording.

**Bucket details** — the raw per-bucket table: calls, successes,
errors, success rate.

## Export

The **Export** button downloads the current window's full observability
payload as JSON — handy for piping into your own dashboards or handing
to an analysis tool.

<Tip>
  Don't watch this page for regressions by hand — create an
  [alert rule](/guides/alerts) on success rate, failure rate, or call
  volume and get emailed/Slacked when a threshold breaks.
</Tip>

## Next steps

<CardGroup cols={2}>
  <Card title="Alerts" icon="bell" href="/guides/alerts">
    Threshold alerts on the same metrics.
  </Card>

  <Card title="Review calls" icon="waveform-lines" href="/guides/review-calls">
    Drill from a number into the actual conversations.
  </Card>
</CardGroup>