Platform · Automation

Workflow automation with humans in the loop

A VegaDūta workflow is a graph of typed steps that runs on its own: it starts from a message, a phone call, or a webhook, hands steps to AI agents, branches on the result, and pauses for a person where you put an approval or a question. You compose workflows from 35 node types and trigger them from 11 channels, from WhatsApp messages to plain webhooks.

While building, you write expressions to pass data between steps, pin a step's output so later steps can be tested without re-running it, and run the draft before promoting a version to production. More than 240 ready-made recipes across about 90 industries give you a starting point, and any workflow can be shared with a public link.

By the VegaDūta team · Last updated 3 October 2026

35 node types to compose from

A VegaDūta workflow is a graph of typed nodes, each doing one job. Agent nodes hand a step to an AI agent, condition nodes branch, loop and parallel nodes control flow, and code nodes run custom logic. Workflow, sub-flow and supervisor nodes let larger automations delegate to smaller ones.

Trigger, agent, knowledge, condition, approval, output
Tool, HTTP request, code, media, device control
Loop, parallel, fan-out, supervisor, workflow, sub-flow
Human input, voice menu (keypad), bulk email, CRM lead

Expressions, pinned data, and draft runs

Fields on a node accept expressions that read the trigger payload and the outputs of earlier nodes, with a preview as you type and a lint panel that flags problems before a run. Pinning saves one node's output on the draft so the steps after it can be built and tested without calling that node again. A draft run uses pinned data and says so in the run summary; live runs from webhooks, schedules, channels, and share links always execute every node for real.

An expression language for passing data between steps, documented at /help/workflow-expressions
Pin an output from a step test, a past run, a received event, or pasted data
Draft runs for editors; production runs follow the version labelled production
Background runs return a run id immediately and are polled for status, for workflows that take minutes

Sales and outreach nodes

Two nodes cover outbound sales work. The bulk email node sends a personalised message to each recipient in a list, with a per-run recipient cap, a send-rate limit, an unsubscribe link, and a suppression list. The CRM lead node creates or updates a lead in the built-in leads list, which tracks each lead through stages from new to won or lost. Replies and bounces to sent email are detected and recorded.

Triggered from 11 channels

Workflows start from real-world events, not just cron. A customer message on WhatsApp, Telegram, SMS, Messenger, Instagram, Slack, Discord, or X can kick off a run, as can a phone call, an email, or an authenticated webhook from your own systems. The same workflow can serve every channel you connect.

Retries you configure, not a fixed policy

Every workflow carries its own execution limits rather than inheriting one platform-wide policy: how many retry attempts a failed step gets (anywhere from 0 to 5), the maximum number of steps a single run may take, and a per-step timeout. Retry counts and the step ceiling apply on both execution engines. One honest difference: the per-step timeout is enforced by the durable Temporal engine — on the default in-process engine it is recorded on the workflow but a step runs until it finishes, and a run whose process dies is marked failed by the recovery reaper instead of resuming.

0–5 retry attempts per workflow — set 0 for a step you never want repeated
Irreversible steps (email sends, tool calls, sub-workflows) are never retried automatically — a repeat would duplicate the side effect
A quota rejection is never retried — it would fail identically and burn the attempt

Humans stay in the loop

Approval nodes pause a run until a person approves or rejects the step, and human-input nodes pause it to ask a person a question through a form. A running workflow can also be paused and resumed by hand. The paused state is stored on the run record, and the run continues from the node where it stopped. This is how VegaDūta keeps risky actions behind a human decision.

Recipes and sharing

More than 240 workflow recipes ship ready to adapt, organized into about 90 industries, so a clinic, kirana store, or agency starts from something that already works. Finished workflows get public share links. If you are comparing tools, we keep an honest comparison against n8n for India-focused teams.

Frequently asked questions

What workflow engine does VegaDūta use?

Workflows run on VegaDūta's own execution engine by default. Failed steps are retried up to the limit set on the workflow, and a run waiting on an approval or a human answer is stored on its run record and resumed from that node. A Temporal-based engine exists in the codebase as an opt-in and is off by default.

Can a workflow wait for human approval before acting?

Yes. An approval node pauses the run until someone approves or rejects it, and the paused state is stored on the run record until a decision arrives. This is the recommended pattern for payments, deletions, and anything else you want a human to sign off on.

What can trigger a VegaDūta workflow?

Eleven channels: webhook, WhatsApp, Telegram, SMS, phone call, Facebook Messenger, Instagram, Slack, Discord, X, and email. A single workflow can be wired to several triggers at once, so the same automation answers customers wherever they message you.

Can I control how many times a failed workflow step is retried?

Yes — retries are configured per workflow, from 0 to 5 attempts, alongside a maximum step count and a per-step timeout. Both engines honour the retry count and the step ceiling; the per-step timeout is enforced by the durable Temporal engine, which is the engine to choose for long, time-bound runs. A quota-exceeded rejection is deliberately excluded from retry, since it would fail identically on the next attempt — as are irreversible steps, whose repetition would duplicate the side effect.

Do I have to build workflows from scratch?

No. More than 240 workflow recipes ship with the platform, organized into about 90 industries, and you adapt them rather than starting blank. You can also try the builder without signing up in the sandbox at /try.

Can I test a workflow without re-running every step?

Yes. Pin a node's output and draft runs and step tests use the saved data instead of running that node. The run summary states how many nodes were pinned, and live runs never use pins.

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.

Try the sandbox — no signup Browse workflow recipes