Steps

Every workflow has one Start and at least one End. Between them, each step does one thing and then takes a path to the next. In the dashboard you pick a step from the palette and fill in its panel. Through the API, a step is one entry in the definition's nodes, with a type and a config.

Start

The fields every contact brings. mobile_number, name, email, reference_id, user_id (the person's ID in your app) and user_field (which of your app's fields that ID is in: user_id, email, phone or your own) are always accepted; add your own, each with a type and whether it is required: string, number, boolean, timestamp, phone, email, url or json. A value is converted to its type as the contact enters ("50,000" becomes 50000). A contact missing a required field is refused there and then, before anything is called.

reference_id is how a contact is told apart: a contact can only have one run of a workflow under way at a time. Leave it out and the phone number is used, or the user ID for someone with no number.

{ "id": "start", "type": "start", "config": { "fields": [
  { "name": "plan", "type": "string", "required": true },
  { "name": "amount_due", "type": "number" }
] }, "cases": [ { "when": { "always": true }, "then": { "to": "call" } } ] }

Call

Calls the contact with one of your agents, then waits for the call to end. Under How to reach them, choose a Phone call (the default) or In their app: the person's phone rings in your app, as a call does (In their app below).

How the call went is in call.status: completed, no_answer, busy, voicemail, failed, cancelled, timeout (it did not end in time) or error (it could not be placed: no number, or the agent was deleted). With it come call.duration_s and call.execution_id, the call's id in Call History.

In their app

With In their app, the step rings the person your app has signed in (Identity and users), on their phones that use your app, with hosted ringing. It can ring them now, or book the call for a time, as a scheduled call.

{ "id": "call", "type": "agent",
  "config": { "agent_id": "agt_0123", "channel": "app", "identity": "entry.user_id",
              "at": "entry.preferred_time", "reason": "About your order", "timeout_s": 1800 },
  "cases": [ { "when": { "cmp": "==", "left": { "var": "call.status" }, "right": { "const": "completed" } },
               "then": { "to": "results" } } ],
  "on_no_match": "not_reached" }

A call in the app follows every rule of a call your app or your agent books: it rings only within the agent's calling hours, outside quiet hours, in the person's own time zone, and a booked time outside them moves to the next time that works. A booked call shows in the person's upcoming calls in your app, and in the agent's Scheduled calls. The timeout counts from when it rings, so a call booked for tomorrow is waited for from tomorrow. If you cancel the run, a call it booked that has not rung yet is cancelled too; a call that is ringing or live is never cut.

How it ended is in call.status, the same values as a phone call:

When the call can never be placed as written, call.status is error and call.error says why: no_identity (no ID for this person), invalid_identity (an ID that names nobody), no_app_user (none of their phones has used your app in 30 days), app_calls_not_set_up (your app can't be rung yet: see Hosted ringing), too_soon (a booked time sooner than the agent lets calls be booked) or invalid_time. An empty balance, a switched-off agent, quiet hours, too many calls already booked for the person, or a time further ahead than the agent books calls don't fail the step: the call is asked for again later, and that is never an attempt. A booked time is fixed when the step first runs, so asking again never moves it later; a time that only became too soon while it waited moves to the soonest the agent allows. Calls booked into quiet hours ring when they end, spread over two minutes so a list of them doesn't ring at the same second.

Read call results

Reads the fields your agent's post-call analysis (Agent → Custom Analysis) found in the call its Call step placed. Put it right after that Call step, and pick the fields you want.

{ "id": "facts", "type": "extraction",
  "config": { "agent_node_id": "call", "selected": ["interested", "callback_time"] },
  "cases": [ { "when": { "cmp": "==", "left": { "var": "extraction.interested" }, "right": { "const": true } },
               "then": { "to": "crm" } } ],
  "on_no_match": "not_interested" }

The analysis is ready a minute or so after an answered call, and the step waits for it, at most 10 minutes. For a call nobody answered there is nothing to read, so the step moves on at once with no fields. Set to The call or chat that started the run, it reads the call or the chat whose end started the run (Start when a call ends).

Check call history

Counts how many times the person was already called in a window before now, by any of your agents: lists, other workflows, scheduled calls and callbacks, and the API all count, and so do this run's own earlier calls. Put it before a Call, or make your Try again go back to it, and its paths decide whether to call: stop at three calls a week, stop once someone has answered, or go on. Nothing else changes: only a workflow with this step looks at the history.

{ "id": "check_history", "type": "history",
  "config": { "window": { "days": 7 } },
  "cases": [
    { "when": { "cmp": ">=", "left": { "var": "history.calls" }, "right": { "const": 3 } }, "then": { "to": "called_enough" } },
    { "when": { "cmp": "==", "left": { "var": "history.status" }, "right": { "const": "unavailable" } }, "then": { "to": "not_checked" } }
  ],
  "on_no_match": "call" }

The person is found by their number (or, in the app, their ID in your app), never by the contact: the same person is a different contact in each list, and every one of those calls counts. Only calls that really went out count: not a call that was booked and never rang, or one refused before it was placed (an empty wallet, an agent switched off).

HTTP request

One HTTP request to your system: push a qualified lead to your CRM, look a customer up, book a slot. Setup is in Connect your systems. The answer is in response.status and in the values you map from its body. When your system could not be reached, there is response.error. A 4xx or 5xx is an answer like any other, so a path can act on it.

Wait for your system

Pauses the run until your system calls back, for example when a deal changes stage or a payment arrives, or until the timeout. How your system calls back is in Connect your systems.

callback.status is received or timeout, with your values beside it. A path must say what happens on timeout.

Send a WhatsApp message

Sends a message from one of your connected WhatsApp numbers (Integrations → WhatsApp). Choose an approved template, with its language and the values for its placeholders, or a text. A text only reaches someone who wrote to your number in the last 24 hours; to start a conversation, use a template.

{ "id": "nudge", "type": "whatsapp", "config": {
    "channel_id": "your-whatsapp-channel",
    "template": { "name": "payment_reminder", "language": "en",
                  "parameters": [ { "var": "entry.name" }, "{amount}" ] },
    "variables": { "amount": { "value": { "var": "entry.amount_due" } } } },
  "cases": [ { "when": { "cmp": "==", "left": { "var": "message.status" }, "right": { "const": "sent" } },
               "then": { "to": "done" } } ],
  "on_no_match": "not_sent" }

message.status is sent or failed, with message.id, or message.error saying why: no number to send to, the channel needs reconnecting, or the code WhatsApp gave.

Send an SMS and Send an email

Send with one of your messaging connections (see Connect your systems): an SMS with Twilio, an email with SendGrid or Resend. An SMS is up to 1,600 characters, sent to entry.mobile_number unless you choose another value. An email has a subject and a text, sent to entry.email unless you choose another. Both can use placeholders, and both set message.* as WhatsApp does.

Run an automation

Starts one of your automations (Workflows → Automations) with the values you choose from the run. That brings everything an automation can do into a workflow: Slack, HubSpot, Google Sheets, Notion, Salesforce, Stripe and the rest. automation.status is started or failed; the automation then runs on its own.

Wait

Waits for a time (delay: days, hours and minutes, up to a year) or until a moment (at). The moment can be a fixed time with its offset, or a value of type timestamp with a time zone, such as a follow-up time your system sends with the contact. It has one way out, its to.

{ "id": "until_follow_up", "type": "time",
  "config": { "at": { "var": "entry.follow_up_at" }, "timezone": "Asia/Kolkata", "to": "call_again" } }

Try again

Goes back to an earlier step, such as the Call, after a wait, for a limited number of attempts: 1 to 10, each with its own wait. It is the only way a workflow can go round in a loop. Once every attempt is used, its paths decide what happens next, usually an End.

When it goes back to a Call, the next call is booked for its time straight away. Until it rings, it is under Upcoming in Call History, marked Booked by a workflow, and it meets the same checks as any call when its time comes: your balance, the agent being on, its calling hours and your lines. It carries the same values the step would send at that time. If you cancel it there, the run moves on with call.status cancelled. A call in your app that has no time of its own is booked the same way, in the person's upcoming calls.

{ "id": "again", "type": "retry",
  "config": { "target_node_id": "call", "attempts": [ { "delay": { "minutes": 30 } }, { "delay": { "hours": 4 } } ] },
  "cases": [ { "when": { "always": true }, "then": { "to": "unreached" } } ] }

End

Ends the run with a label, the name you will count it by (qualified, paid, unreached), and an outcome: success, failure or neutral. Each published version counts its runs by outcome, and every list shows its totals.