Sequencing

A single AI call is one touch. Sequencing is the whole follow-up campaign around it — call, and if they don't pick up, wait an hour and try again; if they answer, text them the link. You draw that plan once on a canvas, and the engine runs it for every contact, picking up where each step leaves off.

The short version: a Finn is one conversation. A sequence is many steps over time — call → wait → branch on what happened → retry or send an SMS → repeat. It's the layer above your agents.

When you actually need it

You don't need a sequence to make calls — a deployment already does that. Reach for sequencing when the follow-up logic matters:

• "Call them. If no answer, retry after an hour. After 3 tries, give up."
• "If they pick up, text the payment link. If it goes to voicemail, try again tomorrow."
• "Call → wait a day → call again with a different message."

If your outreach is "call this list once," skip sequencing. If it's "keep working each lead until something happens," this is the tool.

The pieces (nodes)

You build a sequence by dragging nodes onto the canvas and wiring them together:

After every Voice step, the sequence remembers what happened so a Condition can branch on it:

• call_answered → true / false
• call_outcome → answered, no_answer, voicemail, failed, …
• call_voicemail → true / false

Build your first cadence

Let's build the classic: call the borrower; if they don't answer, wait and retry; if they do, text the payment link.

1. Go to Sequences in the sidebar → New sequence.
2. Drag on: Start → Voice → Condition, then two paths off the Condition.
3. Voice node → pick your Finn, set it to outbound.
4. Condition node → add two branches:
   • Answered → rule: call_answered is true.
   • No answer → mark this one Default (the fallback).
5. Wire Answered → SMS:
   • To: {contact_phone}
   • Message: Hi {contact_name}, here's your payment link: …
6. Wire No answer → Wait (1 hour) → Voice (retry) → End.
7. Wire the SMS → End.

When the graph is sound, flip the sequence to Live. The editor won't let you go live with missing pieces — it tells you exactly what's wrong.

Run it

Open the sequence → Run → Run live. Drop in contacts, one per line:

9876543210
Jane Doe,9876543210
Jane Doe,9876543210,91
+14155551234

(name,phone,countryCode — the part with digits is the phone.) Each contact starts its own run. The History tab shows every run live:

• running — moving through steps
• waiting — paused on a call outcome or a wait timer
• completed — reached the end
• failed — a step errored with no fallback wired

Heads up: Run live places real calls and sends real texts. Each call is billable and checks your wallet first. Use the Preview path button for a dry walk that sends nothing.

How the timing works

• A Voice step parks the run as waiting until the call finishes, then fills in the call_* values for the next Condition.
• A Wait step resumes shortly after the timer is up (checked every ~30 seconds — so cadences are minute- and hour-grained, not to-the-second).
• Runs are durable: they survive restarts and never double-fire, so a "wait 2 days" step is safe.

Tips & limits

• One starting point per sequence for now.
• Sending SMS needs a number on your account; otherwise the SMS step is skipped (and follows an error path if you wired one).
• A failed text won't kill the cadence by default — the call flow keeps going.
• One run = one contact. 500 contacts = 500 independent runs, each on its own timeline.

What's not here yet

You may see extra node types on the canvas — WhatsApp, API call, custom Function, Loops, Parallel branches, Calendar and CRM steps. Those are coming. For now, stick to Voice, Wait, Condition, SMS — if a sequence uses an unsupported node, going live will tell you which one to remove.

Start simple: one call, one retry, one text. That alone turns a cold list into a worked one.