> ## 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.

# Clustering

> See similar agent traces grouped together in a 3D semantic map.

Clustering shows your agent traces as dots in a 3D **Semantic map** so you can quickly see repeated behaviors and failure patterns.

## How to use it

1. Open **Clustering**.
2. Choose an **Agent**.
3. Choose a **Job** if more than one clustering job is available.
4. Look at the **Semantic map**.
5. Drag to orbit and scroll to zoom.
6. Click a dot to inspect the trace behind it.
7. Review **Clusters** to see the behavior groups Kindred found.
8. Click **Open trace** to view the trace in **Behavior Search**.

## How to interpret the results

Each dot is a trace or turn.

Nearby dots are similar behaviors. A tight group usually means your agent is repeating the same workflow, tool path, failure mode, or response pattern.

A cluster is a named group of similar traces. Cluster cards show:

* Number of traces
* Error percentage
* Average duration
* Average tokens, when available

Use Clustering when you want to know:

* What behaviors are happening most often?
* Which repeated patterns are failing?
* Which trace should I replay?
* Which behavior should become an expectation?

The page shows completed clustering runs for the selected agent. The current data window is **Past 7 days**.

## Next steps

* Open an example in [Behavior Search](/features/behavior-search).
* Replay a representative turn with [Replay](/features/replay).
* Turn recurring requirements into [Expectations](/features/expectations).