> ## Documentation Index > Fetch the complete documentation index at: https://docs.platform.toloka.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Resume or cancel a pipeline run > Resume a run after an automatic funding pause, or cancel it and understand the effects on results and charges. A deployed run does not always run straight through to completion. If your team balance stops covering the work, the run pauses itself and holds its progress. Resume it to continue each item from the node where it stopped, or cancel it to end the run for good. **Resume** and **Cancel** are in the UI, on the deployed pipeline's **Pipeline** tab. They sit on the bottom toolbar, with **Clone** and **Export** inside its **More pipeline actions** menu. **Resume** appears on the toolbar only while the run is paused, and also on the run's top-up banner once your balance covers the remainder. You cannot pause a run yourself — pausing is something the platform does when funds run out. The agent cannot resume or cancel a run. You do that here. ## Before you start Open the run's **Pipeline** tab after deployment succeeds. The run controls disappear when cancellation starts or all items have finished processing. Resume is available only in the UI; canceling is also available through the [public API](https://docs.platform.toloka.ai/docs/api-reference.md). ## When a run pauses itself for funds When your team balance can no longer cover the next node's cost, the run pauses on its own and a banner shows how much to top up. [Billing](https://docs.platform.toloka.ai/docs/core/billing.md#pipeline-runs) covers funding and this automatic pause. Work can begin before deployment finishes. If funds run out during deployment, some items may show **Paused**. Wait for deployment to finish before resuming. The top-up banner and run controls remain hidden until then, and remain hidden if deployment fails. Once deployment is finished, top up until the balance covers the run's worst case, then resume the run normally. The run stops handing out new work, but work already underway continues. An item partway through a node finishes that node, then holds before the next one. Because mid-node work always completes, exactly one more node per item can finish around the moment the run pauses. Items that have already finished are unaffected — no work in progress is lost. You can tell the run is paused: **Resume** appears on the toolbar, each held item shows a **Paused** badge, and a `paused: N` chip appears among the item filters. The progress bar above them counts only finished items, and shows the paused share as a percentage on hover. A pause does not refund work already done. The work that finishes right as the run pauses is billed pay-as-you-go like any other work — see [Billing](https://docs.platform.toloka.ai/docs/core/billing.md#pipeline-runs). ## Resume a run **Resume** appears only while the run is paused. On the run's banner it appears once your balance covers the run's worst case, not just the expected remainder — so top up, then resume. Click **Resume** and confirm the prompt. Every paused item continues through the pipeline from the node where it stopped, and finished items are untouched. The unpaid node is retried, so no item is skipped. While the resume is starting, the toolbar shows **Resuming…**, and **Resume** is not available again until it settles. ## Cancel a run > [!WARNING] > > Canceling a run is permanent. You cannot resume a canceled run, and the action cannot be undone. **Cancel** works on a run that is either in progress or paused. Click it, then click **Cancel run** in the prompt. To back out, click **Keep Running**. The prompt also asks **Why are you canceling?** The answer is optional and does not change what happens to the run. Unlike a pause, canceling actively stops labeling at the node. An expert who already holds a task finishes it. Labeling tasks that no expert holds are canceled. If review rejects a task, its expert can still fix it in a redo. A Code or Match node partway through finishes and the item is then canceled. Items not yet started on their next node are canceled right away. Canceling a paused run cancels its paused items without resuming them. It does not restart the run. The run shows as canceled when cancellation finishes. Over the API, the run then reports `failed` with the error `Run canceled`. Some items can end later: an item whose task an expert still finishes, and an item that waits for review. A canceled item pays for the work recorded until it ends on its node, and the rest of its reserved cost returns to your balance. See [What a canceled or failed item costs](https://docs.platform.toloka.ai/docs/data/concepts/pricing.md#what-a-canceled-or-failed-item-costs). ### Items that wait for Human QA On a Generation node with human review, items can wait for Human QA when you cancel. For each such node, the prompt lists the items in Human QA and in labeling, and about how much they can still cost. An item counts as in Human QA from the moment its expert submits it, also while LLM QA checks it. Then select what happens to the items that wait for Human QA: * **Accept without review** is selected when the prompt opens. The items are accepted without review. You pay for the accepted items as for completed items, and you get their data. Their Human QA verdicts read `skipped`. An item that LLM QA rejected is not accepted: like any rejected task, it can go back to its expert for a redo. An item that a reviewer already works on ends with that reviewer's decision. * **Wait for review** keeps the items in Human QA, and each item ends with the review result. These items can end after the run shows as canceled and still be charged. If an item still waits 14 days after it started on its node, **Accept without review** applies to it. The choice also applies to items that reach Human QA after you cancel, such as an item whose task an expert still finishes. If the prompt cannot show the choice, because it still counts the items or the count failed, the cancel uses **Wait for review**. An item accepted without review ends as canceled. The run's **Items** tab shows it as canceled with no item output. In the item's **Logs**, its node shows the answer under **Partial output (captured at cancellation)**. The [run export](https://docs.platform.toloka.ai/docs/data/guides/export-results.md) and the API also show the item as canceled, and their `partialOutput` carries its answer when the End node returns it. All its outputs are in the dataset. A later run that includes the item labels it again and charges for it again. Over the [API](https://docs.platform.toloka.ai/docs/api-reference.md), `POST /runs/:runId/cancel` with the body `{}` selects **Wait for review**. The body `{"pendingHumanQa": "accept_without_review"}` selects **Accept without review**. In the [CLI](https://docs.platform.toloka.ai/docs/cli.md), `toloka pipeline cancel --accept-without-review` selects **Accept without review**. Without the flag, the command selects **Wait for review**. With **Wait for review**, **Accept without review** applies to an item that still waits 14 days after it started on its node, and you pay for the item if it is accepted. ## Related * [Label a large dataset in batches](https://docs.platform.toloka.ai/docs/data/guides/label-at-volume.md) — the batch cycle these run controls belong to * [Billing](https://docs.platform.toloka.ai/docs/core/billing.md#pipeline-runs) — funding a run, and the automatic pause when the balance runs out This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.