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

# Introduction

> Kindred helps teams debug flaky AI agents by replaying real agent turns and showing the first meaningful difference.

Kindred reads the traces your agent sends to Langfuse. You instrument your agent with the normal Langfuse SDK or integration, then Kindred syncs those traces and compares the original behavior against replay behavior.

## How it works

1. Capture traces with Langfuse.
2. Sync traces into Kindred.
3. Choose a turn and replay it.
4. Kindred compares original vs replay.
5. Kindred highlights the first important difference it finds.

## Choose your path

* **New to Langfuse:** start with [Use a Coding Agent](/get-started/use-a-coding-agent) to patch your agent with Langfuse instrumentation and Kindred metadata.
* **Already using Langfuse:** go straight to [Quickstart](/get-started/quickstart), connect your Langfuse project in Kindred, and click **Sync now**.
* **Ready to replay:** configure the endpoint Kindred should call with [Configure Replay URL](/replay/configure-replay-url).

<Note>
  For onboarding, use Langfuse SDK instrumentation and add the Kindred metadata fields documented here.
</Note>