VegaDūta

Tutorial · Automation

How to create an automated workflow

This tutorial walks you through building an automated workflow in VegaDūta: a multi-step process that runs on its own, triggered by a webhook or an incoming message, with AI agent steps and human approvals where you want them. You can start from one of roughly 90 ready-made recipes or from a blank canvas, and every run executes on durable Temporal-backed infrastructure — so a crash or restart resumes the run instead of losing it.

Step by step

  1. Start from a recipe or a blank canvas

    Browse the recipe library at /help/workflow-recipes — around 90 sector-organized starting points — and pick one close to your process, or open a blank canvas in the workflow builder. Recipes are editable, not fixed.

  2. Choose a trigger

    Decide what starts a run: a webhook call from another system, or an incoming message on any of 11 channels, including WhatsApp, Telegram, and email. One workflow, one trigger; you can build sibling workflows for other entry points.

  3. Add the nodes

    Drag in the steps: agent nodes that think and act, condition nodes that branch, knowledge nodes that retrieve from your documents, plus code, loop, parallel, and sub-workflow nodes for more involved logic. Connect them in the order the work should flow.

  4. Add an approval node where a human should sign off

    Anywhere the workflow is about to do something consequential — send an external message, change a record — insert an approval node. The run pauses there until a person approves or rejects, then continues down the matching branch.

  5. Run it and watch the execution

    Trigger a test run and watch it execute step by step. Execution is Temporal-backed and durable: long waits, retries, and restarts do not lose state, so a workflow that pauses a day for an approval simply picks up where it left off.

  6. Share it if you want to

    A workflow can be shared with a one-tap public link, letting someone outside your console kick off a run. Keep approval nodes in place on shared workflows so public triggers still pass through your sign-off.

Troubleshooting

When a workflow does not behave, look at the run history before editing the graph.

  • A run stuck 'waiting' is usually parked at an approval node — check the approvals queue.
  • A trigger that never fires is usually a channel not yet connected, or a webhook URL not yet called by the other system.
  • Test condition branches with runs that deliberately take each path; an untested branch is where surprises live.

Frequently asked questions

What can trigger a workflow in VegaDūta?

A webhook call from any external system, or an incoming message on any of 11 supported channels — including WhatsApp, Telegram, and email. That means a customer message can start a full multi-step process, not just a chat reply.

Can a human approve a step before the workflow continues?

Yes. Insert an approval node at any point and the run pauses there until a person approves or rejects it. Because execution is durable, the run can wait as long as it needs to without losing state.

What happens to a running workflow if something crashes mid-run?

Runs execute on Temporal-backed durable infrastructure, so state is persisted step by step. After a crash or restart the run resumes where it left off rather than starting over or disappearing.

Do I have to build workflows from scratch?

No. There are roughly 90 ready-made recipes, organized by sector, at /help/workflow-recipes. Pick the closest one, then edit its triggers and nodes to fit your process — a recipe is a starting point, not a locked template.

See it working in two minutes

The sandbox provisions a real tenant — describe an agent in one sentence and test it, no account, no card. Or browse ~90 industry workflow recipes to see what teams build.

Related

Explore VegaDūta