Learn Callingtap by doing

A receptionist is the simplest useful agent: it answers, understands why the person called, answers the common questions, and books or routes the rest. Here's the path from a blank workspace to that agent answering real calls.

1. 1Start with the widget — it's free and instant. Create an agent (Dashboard → Agents → Create), name it after the job — "Front desk — Sharma Dental" — and attach it to a website widget. The free self-hosted stack (Whisper, Ollama, Kokoro) powers it with nothing to configure.
2. 2Write a prompt like a job description. Who the agent is, the business hours, the services and prices it may quote, what it must never promise, and exactly what to do when it doesn't know: take a name and number and promise a callback. The agent follows the prompt literally, so be concrete.
3. 3Feed it your facts. Upload your FAQ, price list and policies to the agent's knowledge base. Grounded answers come from retrieval over your documents — an agent without knowledge will improvise, and improvisation is where AI receptionists go wrong.
4. 4Set the greeting and test brutally. The greeting is pre-synthesised, so keep it short and warm. Then test in the playground: ask awkward questions, interrupt it mid-sentence (barge-in should feel natural), mumble, go silent. Fix the prompt, not your expectations.
5. 5Add the phone when you're happy. Connect your carrier (Twilio, Vonage or a SIP trunk) under Settings → Telephony, attach a number, and point the agent at it. Your receptionist is now answering the real line.
6. 6Review like a manager. Every call lands in your dashboard with a transcript, a QA scorecard and extracted action items. Read the first twenty transcripts yourself — that hour teaches you more about prompt-writing than any tutorial.

Stuck on a term? The glossary is written for exactly this guide.

Credits are the one currency inside Callingtap. You never need to do exchange-rate maths mid-call — but you should understand the model cold, because it decides what your agents cost to run.

The unit. One credit is worth $0.01 of plan value. Every plan includes a monthly allowance: Free 200 credits, Starter 2,900 ($29), Growth 9,900 ($99), Scale 29,900 ($299). When the allowance runs out, calling pauses rather than surprising you — wallet top-ups and usage live under Dashboard → Billing.

The rate. A call draws credits per minute, at the rate of the stack the agent runs on. The economical stack draws about 7 credits a minute; the standard stack about 16. So a five-minute standard-stack call is roughly 80 credits — about $0.80 of plan value — and you can see the exact figure per call in your call history, and per service on the Credit Rates page.

The free rule. Open-source, self-hosted services are never charged. Whisper transcription on your server, an Ollama model, a Kokoro voice — they show on the rate card as "Free — 0 credits", permanently. That's why the Free plan's website widget works at all: it runs entirely on the free stack. Telephony is separate again: your carrier bills your own account directly, with no Callingtap markup — credits only ever pay for the AI side.

The visibility rule. Tenants see credits — how many a service costs, how many a call used, how many are left. That's deliberate: your team should be able to predict a bill without knowing what any underlying provider charges us.

AI calling is regulated calling. The rules differ by country, they change, and getting them wrong costs money and reputation. This guide is the working basics for Callingtap's launch markets — general information, not legal advice. Have your own counsel review your campaigns before you dial at scale.

Consent comes first, everywhere. Only call people who have agreed to hear from you, and keep the evidence: who consented, when, and for what. Callingtap captures consent when leads are imported and stores it in a consent ledger against each contact; a withdrawal flips that contact to Do Not Call automatically. In the GCC, consent expectations are strict and enforced — treat "we bought a list" as a red flag, not a source.

India: respect the DND registry and the clock. Promotional calling into India sits under TRAI's framework: numbers registered as Do Not Disturb must not receive promotional calls, and calling windows are restricted (promotional calls are generally confined to daytime hours). Callingtap re-checks every number against your tenant DNC list and the platform's global DNC list before every dial attempt — but your list hygiene is your responsibility: scrub imports, honour opt-outs immediately, and keep templates and caller identity honest.

Opt-outs are cross-channel. When someone says stop — on a call, on WhatsApp, anywhere — Callingtap suppresses them across campaign dialling, WhatsApp follow-ups and booking messages. Build your scripts so the agent recognises an opt-out in any language your callers speak, and never argue with one.

Recordings and disclosure. Tell callers the call may be recorded and that they're speaking with an AI — disclosure expectations exist in several GCC states and are simply good practice everywhere. Retention should be deliberate too: set how long recordings and transcripts live under your retention settings instead of keeping everything forever.

The platform's own commitments are written down in the DNC Policy and Privacy Policy. For campaign-specific legal questions — sector rules, cross-border data, sector regulators — a qualified lawyer in the market you're dialling is the only safe answer.