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

# Review traces

> Inspect model, speech, and tool activity to understand a conversation.

Open a completed conversation or agent run and select its trace view when available. Trace details help you understand delays, unexpected responses, or failed actions.

## What to inspect

| Trace information          | Question it helps answer                                    |
| -------------------------- | ----------------------------------------------------------- |
| Speech recognition         | Did the agent hear the caller correctly?                    |
| Model inputs and responses | Did it receive the intended prompt and context?             |
| Tool requests and results  | Were the right parameters sent, and did the action succeed? |
| Timing and errors          | Which part of the turn was slow or unsuccessful?            |

The information available depends on the services used and the tracing configuration. A missing trace is not evidence that a call or tool failed.

## Connect an observability service

Open **Resources → External integrations** to configure a service for your selected workspace:

* [Datadog](/integrations/datadog) exports agent metrics, individual call results, and recorded conversation traces. It excludes prompts, transcripts, and tool content.
* [Langfuse](/integrations/langfuse) sends conversation traces to your project. These may contain prompts, conversation content, and tool data.

Admins and Owners can manage both connections. Run a new browser test or inbound call after setup and verify delivery in the destination service.

For the overall outcome, check the [run record](/core-concepts/calls-and-runs).