Core concepts

Turn repeatable work into skills and workflows

Skills help an AI teammate repeat a known procedure. Workflows help the organization run structured processes across people, AI teammates, tools, and systems.

What skills are

A skill is a structured procedure connected to an AI teammate. It can define how that teammate handles a certain request, performs a standard operation, or completes a known task.

In the product, the skills library is labeled `Workflows and Skills`, and the sidebar/menu label is `Flows`.

What workflows are

An organization workflow is a repeatable business process owned by the organization, not by one teammate.

Workflows can coordinate steps across AI teammates, tools, user input, code, messages, conditions, and finish steps. Use them when a process needs more structure than a chat.

How teams build them

Skills and workflows use a shared editor. The editor opens read-only first, then lets users edit, autosave drafts, reset a draft, publish, and view runs.

Publishing saves the workflow graph, starting step, payload format, state definition, execution mode, debounce seconds, and debounce key.

How workflows start

A workflow can start manually, through a public trigger, through a private trigger, or from another workflow path.

Runnable workflows need a published flow map and a starting step. Manual runs validate payload fields, and optional state can be used later through state values.

How teams monitor work

Workflow runs are persisted as run snapshots and move through statuses such as created, in progress, suspended, completed, failed, or aborted.

Teams can open logs and run details to understand what happened during execution.

Frequently asked questions

When does a team use a skill?+

Use a skill when one AI teammate needs a repeatable procedure for a known task.

When does a team use a workflow?+

Use a workflow when a business process spans multiple steps, systems, people, or AI teammates.

Are all builder node types runnable today?+

No. Current runnable structured nodes include LLM, tool, run workflow, run employee, run code, condition, user input, send message, and finish. Mapping and wait-callback types exist in backend definitions but are not runnable today.

Start building your AI team