Workflows

A workflow is a repeatable process that Gondo can run for you.

It might check for new items, read information, make a judgement, update another system, or ask a person to approve something before continuing.

The graph model

A workflow is a graph.

• Nodes are steps.
• Edges connect steps.
• Conditions can decide whether a path should run.
• Data flows from earlier nodes to later nodes through ctx.data.

This lets a workflow be simple or branching. A small workflow may be a straight line. A larger one may branch based on approval, category, risk, or missing information.

Node types

Gondo workflows use four main node types.

Trigger

Every published workflow has one trigger that describes how it normally starts.

A manual trigger starts when a person or chat asks the workflow to run.

An auto trigger runs on a schedule. It can either always start a run or check a connected app first and only start runs when there is new work.

A webhook trigger starts when an external service sends an authenticated request. Its code reads the event and calls triggerWorkflow(data) for each run. Use store for handled-event markers, just as Auto uses it for polling progress.

An employee can also explicitly run any of its own published jobs by supplying the trigger's expected inputs, regardless of trigger kind.

Task

A task node is an AI reasoning step.

Use task nodes for work that involves judgement, messy inputs, documents, images, file reading, classification, drafting, or browser UI operation.

If the step is "look at this and figure out what it means", it is probably a task node.

Code

A code node runs deterministic JavaScript.

Use code nodes for structured API calls, data transformation, or reliable writes when the input is already clear.

If the step is "given these fields, call this API", it is probably a code node.

Human

A human node pauses the workflow and asks a person for input.

Use human nodes for approval, review, editing, escalation, or missing information. The run waits until someone completes the human step.

Drafts, publishing, and runs

Workflows can have draft changes before they are published.

A published workflow is the version used for normal Manual, Auto, or Webhook runs.

A draft is an editable version used while building or testing.

A run is one execution of a workflow. Runs keep a record of what happened, which nodes completed, which data was produced, and whether the workflow finished or paused.

Workflow settings

Some workflow information is metadata, not part of the normalized graph definition. The workflow title shown in lists is metadata.

The graph definition focuses on:

• version
• runDisplay
• variables
• nodes
• edges

This distinction matters mostly when reading the YAML reference.