# Workflows

> Step-by-step automations that run on demand, on a schedule or on an event.

## Repeat a process

A workflow follows a saved recipe: declared inputs, tool calls, AI prompts, agent tasks and conditions. Use it for repeatable work you want to inspect run by run. Use an [agent](/en/docs/agents) when the task needs its own approach.

You declare inputs once and pass their values per run. Manual runs collect them in a form; each automatic trigger assigns its own values. Steps use those inputs and earlier results.

## From trigger to result

Manual runs, schedules, webhooks and events enter the same pipeline through its declared inputs. Steps produce results for later steps, connected services or Workspace files. Independent reads and prompts can run together; writes and agent steps preserve their declared order.

Every workflow has a persistent Workspace folder and temporary folders for individual runs. See [paths and retention](/en/docs/workspace/paths).

## Monitor work and cost

The workflow list shows status, triggers, connected tools, average Spark cost and the last outcome. The average uses up to ten finished runs. Daily cost projections include enabled schedules on active workflows; manual runs, webhooks and events contribute nothing to that projection.

Runs continue when you leave the page. Open **Runs** to inspect step outputs, errors and Sparks; agent steps link to their agent audit. Viewers can inspect workflows and traces. Owners and editors can build, run and configure them.

> **Watch out** — Agent steps execute without asking for tool approval. Review [workflow guardrails](/en/docs/workflows/guardrails) before enabling automatic runs.

## Related

- [Build a workflow](/en/docs/workflows/build)
- [Run, schedule and trigger](/en/docs/workflows/triggers)
- [Understand Sparks](/en/docs/concepts/sparks)