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

# Use Kubit with MLflow Tracing

> Read MLflow traces in place from Databricks with an MLflow Tracing workspace, with nothing to install in your app.

If your agents write MLflow traces to Databricks, Kubit reads them in place. Create an **MLflow Tracing** workspace, connect Databricks and apply the MLflow preset. Nothing changes in your app.

## Before you start

* [ ] MLflow traces in a Databricks table.
* [ ] A Databricks connection for your organization. See [Connect Databricks](https://learn.kubit.ai/warehouses/connect/databricks.md).

## Steps

### 1. Create the workspace

Create a workspace and pick the **MLflow Tracing** origin (`MLflow traces in Databricks`). Pick an **Organization connection**, or click **Create connection**. See [Set up your organization](https://learn.kubit.ai/start-here/admin.md).

![Create Workspace dialog with Workspace Name Support Agent, Workspace Timezone Europe/Athens (UTC+2\*) and Origin MLflow Tracing, with the hint MLflow traces in Databricks, above the Continue button.](https://mintcdn.com/kubit/iy-nIyF9Xtsr91WV/images/agent-observability/integrate/create-workspace-mlflow.png?fit=max\&auto=format\&n=iy-nIyF9Xtsr91WV\&q=85\&s=68d495eeeec477b62d87cb057b8588a8)

### 2. Apply the MLflow preset

The workspace opens on the **MLflow** preset tab in the Modeler. Under **Span table**, click **Choose table** and pick your MLflow `<prefix>_otel_spans` table. Under **Tables to create**, check **Price table** and **Evaluator result tables**. The Evaluator result tables let the Evaluator run and save its labels. **Created in** shows where they go, and you can pick another location. Then click **Set up MLflow**.

For now, Kubit counts every run in the table as finished, with no 15-minute wait.

> [!WARNING]
>
> The **Schema** tab stays grayed out until you apply the preset, with the tooltip `Apply the MLflow preset first. Mapping a table by hand would turn it off for good.` Once a table is mapped by hand, the preset is off for that workspace and can't be turned back on.

### 3. Pick where Polaris results go

On the **Evaluator** page, pick a schema under **Evaluator results**. Polaris writes its labels and scores there, next to the traces. See [Where Polaris results are stored](https://learn.kubit.ai/warehouses/overview.md#where-polaris-results-are-stored).

| **Constraint** | **Value** |
| - | - |
| **Warehouse** | Databricks only. Otherwise the preset fails with `MLflow sources require a Databricks connection.` |
| **Span table columns** | `trace_id`, `span_id`, `parent_span_id`, `name` (string), `time` (timestamp), `start_time_unix_nano`, `end_time_unix_nano` (bigint), `attributes` (variant), `status` (struct), `events` (array of struct) |
| **MLflow assessments** | Not read by the preset |

To try Kubit before you connect a warehouse, export the traces to a file and upload it to a **Sandbox** workspace instead. See [Upload traces to a Sandbox workspace](https://learn.kubit.ai/agent-observability/integrate/sandbox-upload.md#mlflow).

## Verify

The workspace activates on the apply that completes it: a warehouse connection, a fact table with an **Event** and a **Timestamp** column, and at least one enabled subject. Before that, it shows `No data yet. Finish setting up your model.`

## If it fails

| **Symptom** | **Cause** | **Fix** |
| - | - | - |
| The **MLflow** preset tab is grayed out with `The schema was already set up by hand, so the MLflow preset no longer applies.` | A table was mapped by hand on the **Schema** tab | The preset can't be turned back on for that workspace. Create a new **MLflow Tracing** workspace and apply the preset first. |
| `No data yet. Finish setting up your model.` | The model is missing a connection, an event and timestamp column, or a subject | Finish the Modeler steps |

## Next Steps

* [Modeler](https://learn.kubit.ai/warehouses/modeler.md#agent-observability-presets): how presets work.
* [Polaris](https://learn.kubit.ai/agent-observability/use/polaris.md): what Polaris adds to your traces.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.