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

# Review Calls and Test Agents

> Use AI FDE to investigate calls and test agent behavior

AI FDE can help explain call problems and, when available, run simulation tests.

## Analyze a call

Use **Ask** mode for a read-only call investigation.

Provide one of the following:

* A CallKaro call link.
* A call ID.
* A raw call-log URL.

Then ask one focused question.

<CardGroup cols={2}>
  <Card title="Connection problem" icon="phone-xmark">
    Why did this call never connect?
  </Card>

  <Card title="Transfer problem" icon="phone-arrow-right">
    What happened during the transfer?
  </Card>

  <Card title="Slow response" icon="gauge-high">
    Where did the call experience high latency?
  </Card>

  <Card title="Conversation quality" icon="comments">
    Why did the customer disconnect early?
  </Card>
</CardGroup>

**Example:**

```text theme={null}
Analyze this call in Ask mode: [call link]
Why did the appointment-booking function fail?
Use the transcript and call logs, and tell me what evidence is available.
```

AI FDE reviews the available transcript and runtime information, explains what happened, and notes any missing evidence.

<Info>
  Call analysis in Ask mode does not change the agent.
</Info>

## Act on the findings

After reviewing a call:

1. Confirm whether the same issue appears in other calls.
2. Open the relevant agent and version.
3. Use **Plan** mode to propose a safe fix.
4. Review the plan.
5. Build and test a new version.
6. Publish only after verification.

<Warning>
  Do not change a live agent based on one unusual call. Look for a repeated pattern first.
</Warning>

## Run simulation testing

<Info>
  Simulation features appear only for accounts with **Simulations** access.
</Info>

AI FDE can help you:

* Review existing test cases.
* Draft a new test case.
* Run a test batch across selected versions.
* Check passed, failed, and grading-failed runs.
* Compare versions and identify repeated failures.

### Create a test case

1. Ask AI FDE to create a test case.
2. Review the proposed user behavior and success criteria.
3. Select **Approve & save**.
4. AI FDE saves the test case after approval.

Test cases require approval even when using Auto mode.

**Example:**

```text theme={null}
Create a simulation test case where the customer is busy,
asks for a callback tomorrow at 3 PM, and should not be pushed to book a demo.
The test passes when the agent confirms the callback and ends politely.
```

### Start a simulation batch

Tell AI FDE which versions and test cases to use.

```text theme={null}
Run the callback and pricing-objection test cases against versions v3 and v4.
Start the batch and report the batch details.
```

AI FDE reports when the batch starts. Check the results after the simulation runs finish.

<Warning>
  Simulation batches use credits. Confirm the selected versions, test cases, and expected run count before starting.
</Warning>

## Review simulation results

Ask AI FDE to summarize:

* Pass rate.
* Failed cases.
* Grading failures.
* Common transcript problems.
* Differences between versions.
* Results that appear inconsistent or flaky.

**Example:**

```text theme={null}
Review the latest simulation batch.
Group failures by cause, compare v3 with v4,
and recommend which version is safer to publish.
```

## Related documentation

<CardGroup cols={2}>
  <Card title="AI Auditor" icon="clipboard-check" href="/ai-auditor/introduction">
    Automatically review calls and find repeated quality issues.
  </Card>

  <Card title="Review Audit Results" icon="magnifying-glass-chart" href="/ai-auditor/review-audit-results">
    Understand scores, evidence, and recommended fixes.
  </Card>

  <Card title="Create a Simulation" icon="vial" href="/agents/simulation/how-to-create-a-simulation">
    Learn the basic simulation workflow and test-case setup.
  </Card>
</CardGroup>