> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.ovalbee.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.ovalbee.com/_mcp/server.

# Runs and automation

Every time a node does its work, OvalBee records a **run**: its inputs, outputs, logs, artifacts, who or what started it, and how long it took. There's no run of a whole flow — only runs of nodes, chained together by [edges](/concepts/edges).

## What starts a run

| Trigger         | When it happens                                                                      |
| --------------- | ------------------------------------------------------------------------------------ |
| **Manual**      | You click **Run node** in the sidebar, or an agent runs the node with your approval. |
| **Parent Node** | A node upstream succeeded and its flow edge points here.                             |
| **Schedule**    | The node's schedule came due.                                                        |
| **Setup**       | **Start** in a template's setup ran the node.                                        |

The trigger is shown on every run in the node's **History** tab.

## Downstream runs

When a run **succeeds**, OvalBee starts every node that the node's flow edges point to. When one of those succeeds, the same happens again, and so on. A failed or cancelled run starts nothing.

A node that was started by an upstream node or a schedule is skipped if it's still busy with an earlier run, so a burst of upstream activity doesn't pile up copies of the same work. Manual runs are always started.

### Isolated runs

Sometimes you want to rerun one node without setting off everything below it: to refresh a preview, to check a setting, to point a node at another dataset. For a single run, open the menu next to the run button and choose **Run only this node**. To make it the node's default, turn on **Run Only This Node** in the node's automation settings: its manual and scheduled runs then run it alone, the button reads **Run only this node**, and the same menu offers **Run with next nodes** for the occasional full run. Some nodes are isolated out of the box. The [Dataset](/node-library/nodes/dataset-ops-dataset) node is one of them, so picking another dataset doesn't start the whole flow.

### Stopping

While a node runs, the run button turns into **Stop task**. The flow toolbar's tasks popover shows everything running in the flow right now.

## Automation settings

Click the cog next to the run button to open a node's automation settings.

#### Run Schedule

Run the node automatically. **Simple** mode takes *every N minutes, hours or days*, and **Advanced** mode takes a cron expression. A scheduled node shows a clock badge on the canvas.

Some nodes come with a schedule. Labeling and Review nodes, for example, refresh immediately when work arrives and also run every 15 minutes to return expired claims and recover from interrupted refreshes.

#### Require Configuration

Block the node until its required inputs are filled. Until then it can't be run by hand, it's skipped when an upstream node would start it, its schedule doesn't fire, and it shows a **Needs configuration** badge.

#### Discard History

Keep routine automated runs out of the History tab: **Never**, **On success** or **Always**. It applies to scheduled and setup runs only — manual runs are always kept. Labeling queue maintenance uses **On success**, so its history shows failures and nothing else.

## Pausing a flow

Pausing stops a flow's **scheduled** runs and nothing else. Manual runs and the chains they start work as usual, and people can keep working in the annotation tool. Skipped schedules don't leave empty runs behind.

Pause and resume from the tasks popover in the flow toolbar, or from **Pause runs** on the flow's card on the Home page. Paused flows are marked *Paused* on Home and on the Labeling page.

## Running again

A node can run as often as you like. Each run is a new entry in its History tab, and after a successful run its outputs point at the fresh results. What happens to data depends on where the node writes:

* **A dataset the node created for its output** gets a **new version** on every rerun. The previous results stay in the dataset's history and nothing is overwritten.
* **A destination dataset you picked** is written as you picked it. Each node's page in the Node Library says exactly what a rerun does — whether it adds, replaces or skips what's already there.

## Run history and logs

Click a node's status icon, or open the **History** tab, to see its runs. Each run has:

* **Info** — status, trigger, who started it, when, and how long it took.
* **Params** — the input values the run received and the outputs it produced.
* **Logs** — live while the run is going, filterable by level. Log lines older than 90 days are removed, even from a run that is still going; an administrator can change the period. The run and the rest of its record stay.
* **Artifacts** — files and results the run produced, ready to open or download.

## The Activity page

**Activity** in the main sidebar lists runs across every flow in the workspace, with totals for the period, the failure rate, the longest run and how many runs are scheduled for today. Switch between **All Runs**, **Running** and **Errors**, pick a period, and click **Details** on any run to open the same record as in the History tab. The node name links straight to the node on its canvas.

## Related pages

* [Edges](/concepts/edges) — what "downstream" means.
* [Flows](/concepts/flows) — pausing and versioning whole flows.
* [Nodes](/concepts/nodes) — statuses and outputs.