# Run history

**Run history** shows the runs a flow has produced and lets you open any one to see how it executed, step by step. Use it to confirm a live flow is succeeding and to investigate runs that failed. You open run history from the builder with **View Runs**.

**Permissions.** Run history is part of the builder, so viewing it requires the **Admin** or **Partner** role. Reviewers work in [Human review](./human-review.md) rather than run history. See [Users and access](../onboarding/users-access.md).

## Open run history

In the builder, click **View Runs**. A count badge on the control shows how many runs the flow has. The **Runs** panel opens with the flow's runs listed.

If the flow hasn't run yet, the panel reports that there are no runs to show. Runs appear here once the flow has processed input — whether from a live trigger in production or from [Run Draft](../build-ai-agents/test-debug-draft.md) during testing.

## Browse the runs list

Each run is one row in the **Runs** panel, with three columns:

- **Started At** — when the run began.
- **Source** — which version of the flow produced the run: a published version (for example, **v1.0.0**), a draft test run (**Draft**), or an earlier run (**Legacy**). This tells you at a glance whether a run came from your live, published flow or from testing a draft.
- **Status** — the run's outcome, shown as a colored badge: **Completed** (green) for a run that finished its path, **Failed** (red) for one that stopped on an error, and an in-progress state (for example **Running**) for a run still under way.

Click **Refresh** to load the latest runs while the panel is open.

## Open a run

Click a run to open its detail view. A breadcrumb at the top (**Runs › [start time]**) takes you back to the list, and a summary line shows the run's overall **Status**.

Below the summary, the run is laid out as the sequence of nodes that executed, in a table with **Name**, **Status**, **Started**, and **Action** columns:

- Each **node** that ran appears as a row with its own status and start time. A node row shows how many steps it contains; expand it with the arrow to see those steps.
- Internal preparation steps are grouped under a **Pre-Processing** section you can expand the same way.
- A node that failed carries a **Failed** status, and its error message is shown inline so you can spot where the run stopped.

## Inspect a step

On any step, click **Details** to open the step detail view. It shows:

- The step's **name** and **Status**, with timing — **Started**, **Completed**, and **Duration**.
- **Error** — shown first and highlighted when the step failed, so you can read exactly what went wrong.
- **Input** — the data the step received.
- **Output** — the data the step produced.

Input, output, and error are shown as formatted data, each with a **Copy** button; a section reads "No data available" when there's nothing to show. This is how you confirm a node received the values you expected and produced the right result — for example, that extraction returned the fields you defined, or that a write-back node sent the correct payload. As in draft testing, sensitive values such as variable values are masked in these views.

## Investigate a failed run

When a run shows **Failed**:

1. Open the run from the list. Its summary **Status** is **Failed**.
2. Find the node with the **Failed** status — its error message appears inline on the row.
3. Expand the node and click **Details** on the failing step to read the full **Error**, along with the **Input** that step received. Together these usually tell you whether the cause was the incoming data or the node's configuration.
4. Fix the cause in the draft and re-test before publishing a new version. Correcting configuration and re-running is done in the draft — see [Test and debug a draft](../build-ai-agents/test-debug-draft.md) — not from run history.

## Confirm a successful run

For a run that shows **Completed**, open it and check that each node completed and that the outputs are what you expect. Use **Details** on the final nodes to confirm the flow produced the right results — the extracted fields, the matched record, or the payload written back to your system.

## Current limitations

- **Run history is viewed per flow, from the builder.** You open it with **View Runs** inside the AI Agent you're working on.
- **The runs list is browse-and-refresh.** There's no run filtering or search in the panel today; scan the list and use **Refresh** to pull the latest runs.
- **Step input, output, and error are shown as raw data.** They're presented as formatted data records rather than a summarized view.

## Where to go next

- [Test and debug a draft](../build-ai-agents/test-debug-draft.md) — reproduce and fix an issue in the draft.
- [Drafts and publishing](../build-ai-agents/drafts-publishing.md) — publish a corrected version and activate it.
- [Item status reporting](./item-status-reporting.md) — see work summarized by status.
- [Human review](./human-review.md) — act on items the flow sends for review.
- [Monitor and Review](./overview.md) — how runs fit the production operating model.