Workflows

Workflows in Runwork are durable, long-running processes designed for business operations that span minutes, hours, or even days. Unlike simple automations that execute immediately and complete, workflows maintain state across time, handle failures gracefully, and can pause for human decisions before continuing.

A workflow is a series of steps that execute in sequence or parallel. Each step can query your entities, call integrations, or invoke AI agents. The key difference from automations is durability: if the infrastructure restarts mid-execution, your workflow picks up exactly where it left off. No lost state, no duplicate operations.

Human-in-the-loop approval is built in. When a workflow reaches an approval step, it pauses and notifies the designated approver via email, Slack, or in-app notification. The approver reviews the context, clicks approve or reject, and the workflow continues down the appropriate path. This enables real business processes like expense approvals, contract reviews, or hiring decisions, the kind of work that needs a human judgment call, not just automated code.

You create workflows in your own AI tool, like everything on Runwork. Tell your agent: "When a contract is uploaded, extract the key terms, send it to legal for review, and if approved, create the customer record and notify sales." Your agent builds the workflow with all the necessary steps, approval gates, and error handling, and deploys it to your team cloud.

Every workflow instance is tracked from start to finish. You can see which instances are queued, running, paused for approval, errored, or complete. The full execution history shows every step taken, every decision made, and every retry attempted. When something goes wrong, you know exactly what happened and can replay the workflow from any point. Your AI agent can trigger workflows, check instance status, and view execution history directly. From the terminal, runwork workflows trigger starts any workflow with parameters.

You do not have to start from a blank page. Describe the process you already run and your agent builds it against your real data and the tools you have connected, so the first version is the thing rather than a sketch of it.