# Workflows

A workflow automates the work between the systems a team already has: reading, drafting, deciding, documenting and notifying. An agent runs it, asks people when it needs a decision, waits for replies or timers, and can start coding work along the way. People make the decisions; the agent does the reading, drafting and follow-up between them.

> **Note:** Workflows are not a CI/CD pipeline and do not replace one. They handle the human process around it.

![The Workflows page on its Shared tab: Weekly release readiness, Software release review, Campaign weekly readout, Content localization, Incident response and Studio readiness check, each with its description, labels, agent, status, next scheduled run and a Run now button.](https://volobox.io/docs/images/workflows-list.light.webp)

*The Workflows page on its Shared tab: each workflow with its description, labels, agent, status, next scheduled run and Run now.*

## Two kinds of body

A workflow has one of two bodies, chosen when you create it:

- A **playbook** is markdown an agent reads and follows. It suits a process that has to adapt to things you cannot describe in advance.
- **Code** is JavaScript that Volobox runs on the server, for a process that must take the same path every time. The model is still available as a step you call on purpose (`ai.generate`) for the parts that need judgment or prose.

A Code body can wait for a timer, an answer, an approval or a coding run, and continue on the next line when the wait is over. See [Write a code workflow](/docs/write-code-workflow) and the [code workflow API](/docs/code-workflow-api).

![The Weekly release readiness workflow on its Code tab, checked with no problems: a JavaScript body that names three phases, lists open tasks and pull requests, asks the model to assess readiness and creates the release notes task.](https://volobox.io/docs/images/code-workflow-editor.light.webp)

*A Code workflow on its Code tab: the JavaScript body with its phases, checked with no problems.*

A playbook can be converted to Code once, in place. A Code workflow never becomes a playbook. Everything around the body (triggers, inputs, skills, auto-approved tools, access, runs) is the same either way.

## No diagram builder

There is no node-and-arrow builder, by design. Instead, Volobox derives read-only views from the playbook or the Code body, an **At a glance** card and a **Storyboard**, that say what starts the workflow and what it reads, writes and stops on.

**Fill with AI** drafts or changes a workflow from a description; you review the changes and save them yourself. A gallery of templates gives a starting point.

## Triggers

| Trigger | Starts a run |
| --- | --- |
| **Manual** | When someone selects **Start run** |
| **Manual on entity** | From an item's page: a project, task, request, meeting, repository, pull request, Docs page or space, file, or ticket, with that item as context |
| **Schedule** | Once, every few minutes or hours, daily, weekly, or by a cron expression |
| **Event** | When a workspace event matches its conditions |

Events cover tasks and requests, pull requests and AI reviews, meetings, chat messages and polls, support tickets, alerts, app events, Docs and Files, coding, Azure DevOps pipeline runs, and more. Conditions narrow them: one project, a status change, who made the change.

**Preview matches** shows how often a trigger would have fired over the last week of real activity. A workflow's own changes never start it again.

A workflow with a message event on one channel is a **channel agent**: it acts on new messages without being `$mentioned`. See [Agents](/docs/agents).

## Inputs and access

**Inputs** are typed: Volobox items (a project, task or repository), people, groups and agents, and values such as text, numbers, dates and choices. Manual starts ask for them; schedules and events store fixed values.

- A **Personal** workflow runs within the permissions of the person who started it.
- A **Shared** workflow runs as its agent, and has its own roles: Viewer, Runner, Editor and Admin.

## Approvals and questions

A run follows the same approval model as an agent. A write that needs approval waits on a card in the right person's Inbox, unless the workflow lists that tool as auto-approved.

When the run needs a decision, it asks a specific person a structured question, which lands in their Inbox, and waits for the answer. The person who started a playbook run can send it an **intervention** to steer it. A Code run follows its code and takes no messages.

> **Note:** A reply to a customer is always a card a person decides on.

## Rehearse and Dry Run

**Rehearse** runs the workflow against live data but records every write as a **Planned effect** instead of performing it. You see exactly what it would have changed before you schedule it.

> **Note:** During a rehearsal, questions still reach people. No approval cards are raised, and nothing recorded leaves Volobox.

**Dry Run** removes the write tools altogether, to check that the reads work.

## Runs and run logs

Each run has a **Journey** (phases, waits, timers, coding events, interventions) and an **Artifacts** tab for the files it produced. **View run logs** combines the journey with the agent's tool calls and errors, and can be exported.

![A run of the Weekly release readiness workflow, waiting: an Approval required card with the Release Manager agent’s proposed task, Publish GA release notes for Fieldwork, and below it the run’s journey through the Hand off, Assess readiness and Collect evidence phases.](https://volobox.io/docs/images/workflow-run.light.webp)

*A waiting workflow run: the approval card for the task its agent proposes, and below it the run's Journey through its phases.*

The run list shows each run's status and duration, and its AI cost to people allowed to see cost. A **Stats** tab shows runs, runtime and cost over time. Editors can pin a representative run as a worked example.

## AI delivery templates

The **AI delivery templates** are Code workflows that carry work from a task to a reviewed pull request. They cover task triage and start, request refinement, AI review, fixing AI findings, CI repair, reviewers and reminders, handing a review to the agent, risk rating, and what happens after a pull request merges or closes.

Nothing is installed in a project by itself. A team creates the workflows a project needs from the templates, one at a time, and can read and change each one. The templates are offered for shared workflows only, and not where Code workflows are off.

> **Note: The rules the templates cannot change**
>
> An agent's approval never counts as a person's, the branch protection on the code host is the floor, and fix and CI repair rounds are capped.

The risk workflow rates a change and writes the rating on the pull request. Approvals and branch protection decide the merge. AI review is advice: it never approves a pull request or blocks a merge. See [Coding](/docs/coding).

## Related

- [Agents](/docs/agents)
- [Coding](/docs/coding)
- [Home and Inbox](/docs/home-and-inbox)
- [Apps](/docs/apps)
- [Write a code workflow](/docs/write-code-workflow)
- [Workflows](/product/workflows)
