How it works
What actually happens between a phone ringing and an appointment appearing in your calendar — in plain language.
On this page
You don't need to know any of this to use Vocallie. But knowing roughly what happens makes the settings make sense, and makes it much easier to work out why a call went the way it did.
The one-paragraph version
You describe your business in the panel. When a call happens, a live AI voice model has a real conversation with your caller, using your instructions and your knowledge base. While it talks it can look things up, check your calendar, and write to your CRM. When the call ends, the system charges you for the minutes used, writes a summary, and runs whatever automations you've set up.
The four ways a call starts
Every call in Vocallie begins in one of four ways. They all end up in the same place — a real conversation, logged in your call log.
- 1
Someone rings your number (inbound)
Your phone number is mapped to an agent. The call arrives, the system works out which agent should take it — optionally routing by who the caller is, using smart routing — and that agent answers.
- 2
You click call (outbound, right now)
From the Dialer or from a contact record. This dials immediately — it skips the queue entirely, because you asked for it now.
- 3
The system decides it's time (outbound, queued)
Follow-ups, auto-callbacks and campaign leads don't dial straight away. They join a call queue, and one dispatcher works through it — respecting your calling hours, your line limits, retry rules and your balance.
- 4
Every line was busy
If a call arrives when all your lines are in use, the caller isn't dumped into silence. The system captures their number, plays a short "we'll ring you straight back" message, and queues an auto-callback for when a line frees up.
What happens during a call
Once the conversation starts, the agent isn't just talking — it has a set of tools it can reach for mid-sentence, without you doing anything:
| The agent can… | Which means |
|---|---|
| Search your knowledge | It looks up your prices, policies and FAQs before answering. |
| Check availability | It reads your real calendar or resource timetable. |
| Book an appointment | It writes a real booking, linked to the caller and the call. |
| Update your CRM | Add a note, change a pipeline stage, fill in custom fields, schedule a follow-up. |
| Call your own systems | Any Action you've configured — an order lookup, a stock check. |
| Transfer to a human | Rings your team, in the order you set, and hands the call over. |
| End the call | When the conversation is genuinely finished, or a limit is hit. |
Two limits are always watching in the background: a maximum call length, and silence handling — if the caller goes quiet, the agent prompts them a couple of times, then politely hangs up rather than burning your minutes on an empty line.
What happens after a call
The moment the call ends, several things happen without you touching anything:
- The call is finalised — duration, whether it was answered, and why it ended.
- Your wallet is charged for the minutes used, and a line appears in your billing ledger.
- An AI summary is written (on plans that include it): a one-line summary, a sentiment read, and any extracted variables you asked the agent to pull out.
- Automations run — your post-call rules fire: create a follow-up, move a stage, add a note.
- Webhooks fire — if you've subscribed, the event is pushed to your own systems.
- Queued calls report back — a campaign lead or follow-up is marked done, or scheduled for a retry.
The rules that are always enforced
A few things are checked on the server every single time, and cannot be bypassed from the panel:
- You only see your own data. Every record belongs to your workspace and is filtered by it.
- Permissions, then plan features, then money. A teammate can't reach a page they don't have permission for; your plan decides which modules exist; and no call is placed if your wallet can't cover it.
- Concurrency limits. Your plan sets how many calls can run at once. Automated calls are deliberately capped below that, so an outbound campaign can never eat the lines your real customers are trying to ring in on.
- Calling hours. Automated outbound only dials inside the window you define, in your timezone.
Note
This is why an action sometimes "doesn't do anything": it was blocked by a rule, not by a bug. The Troubleshooting page maps the common blocks to their cause.
The voice itself
The live conversation runs on a real-time speech model — audio in, audio out, no transcribe-then-reply round trip. That's why it can be interrupted mid-sentence and why it responds in well under a second.
Two settings you'll meet because of this:
- Response speed — how long a pause has to be before the agent decides you've finished talking. Too fast and it interrupts; too slow and it feels sluggish.
- Filler phrases — short natural sounds ("let me check that for you") played while a slower lookup runs, so there's never dead air.
Both live in Agents → Voice & behaviour.
Where things are stored
Everything you see in the panel is stored in one place and shared across every screen. A caller who books an appointment on a Tuesday call shows up as: a call with a transcript, a contact in your CRM, an appointment in your calendar, an activity entry on their timeline, and a line item in your billing ledger — all pointing at each other, without you filing anything.