Resources · Glossary
Glossary
The words voice AI runs on — and the ones Callingtap adds — defined in plain English, the way we'd explain them on a call.
Action items
Tasks Callingtap extracts automatically after a call — 'send the quote', 'call back Tuesday' — each with an owner and a due date, so follow-through doesn't depend on someone re-listening to the recording.
AI agent
A configured voice assistant in Callingtap: a prompt that defines who it is and what it must do, an AI stack that hears and speaks for it, a knowledge base it answers from, and the numbers, widgets or campaigns it works on.
Barge-in
The caller interrupting the agent mid-sentence — and the agent stopping, listening and adapting instead of talking over them. Natural conversation depends on it; Callingtap agents support barge-in on live calls.
BYOK (bring your own key)
Callingtap's connection model: telephony, AI and integration accounts are yours. You paste your own provider keys, they're encrypted at rest, and providers bill you directly with no platform markup.
Campaign
An outbound calling run: a contact list, an agent, a caller number and a schedule. Callingtap dials the list, agents hold the conversations, and every attempt is compliance-checked before it goes out.
Concurrency
How many calls can be live at the same moment. Plans cap concurrency (for example, 1 simultaneous call on Free, 10 on Scale) alongside the number of campaigns and contacts.
Consent ledger
Callingtap's record of who agreed to be contacted, when, through which channel and on what basis — captured when leads are imported and updated when consent is withdrawn. Withdrawing consent also places the number on the Do Not Call list automatically.
Credits
Callingtap's billing unit. One credit is worth $0.01 of plan value; plans include a monthly allowance, and calls draw credits per minute at the rate of the agent's stack. Open-source, self-hosted services are never charged — they read 'Free — 0 credits'.
DNC (Do Not Call)
Lists of numbers that must not be dialled — a tenant's own list plus a platform-wide list. Callingtap re-checks every number against both lists before every single dial attempt, not just when a campaign is created.
Endpointing (turn detection)
Deciding when the caller has finished speaking so the agent can reply. Good endpointing waits through natural pauses without making the caller feel rushed — or leaving dead air.
Flow builder
Callingtap's visual editor for conversations that need structure: steps, branches, questions and tool calls arranged as a diagram instead of one long prompt. Anyone on the team can design a flow — no code.
Follow-up automation
Rules that act after a call ends: fire a webhook to your systems, or send a WhatsApp template message, when the outcome matches — for example, every 'interested' call triggers a brochure and a CRM update.
Greeting
The first thing an agent says when a call connects. In Callingtap the greeting is pre-synthesised before the call, so the caller hears it instantly instead of waiting for speech to be generated.
Human handoff
Passing a live conversation to a person — with the context of what was already said, so the caller never has to start over.
Knowledge base
The documents and notes an agent answers from — your pricing, policies, product details. Callingtap retrieves the most relevant entries for each question so answers are grounded in your material.
Latency
The delay between the caller finishing a sentence and the agent answering. It's the sum of transcription, model thinking and voice synthesis; Callingtap traces it per turn so slow setups can be found and fixed.
LiveKit
The open-source real-time media server Callingtap uses to carry live voice — for the website widget and browser calls — including the self-hosted instance at lk.callingtap.com.
LLM (large language model)
The model that understands the conversation and decides what the agent says — from self-hosted open models like Llama (via Ollama) to cloud models you connect with your own keys.
Opt-out
A person asking not to be contacted again. In Callingtap an opt-out suppresses future calls and follow-ups across channels — campaign dialling, WhatsApp follow-ups and booking messages — not just the channel it arrived on.
PSTN
The public switched telephone network — the ordinary phone network. Reaching real phone numbers requires a telephony provider (Twilio, Vonage, a SIP trunk) connected to it; browser voice uses WebRTC instead.
Prompt (system prompt)
The standing instructions that define an agent: its role, goals, tone and hard rules. Along with the knowledge base, it's the main lever for call quality — the agent follows it literally.
QA scorecard
An automatic quality review of a call — greeting, compliance, objection handling, outcome — scored from the transcript and conversation metrics, with coaching views for the agent's owner.
SIP (Session Initiation Protocol)
The standard protocol for setting up voice calls over IP networks. A SIP trunk connects Callingtap to a carrier — including the Indian providers many GCC and India teams already use.
Streaming
Generating and speaking the reply in pieces as the model produces it, instead of waiting for the full answer — the difference between a conversation and a walkie-talkie. Callingtap streams both model tokens and synthesised sentences.
STT (speech-to-text)
Turning the caller's speech into text the model can read — also called transcription or ASR. Callingtap uses Whisper (self-hosted) by default, with cloud transcribers available via your own keys.
Tenant / workspace
One customer organisation inside Callingtap: its users, agents, contacts, numbers and credits, fully separated from every other tenant. 'Workspace' is the word the product uses day to day.
Transcript
The written record of a call, produced live from speech-to-text. Transcripts power summaries, QA scoring, action items and search across your call history.
TTS (text-to-speech)
Turning the agent's reply text into spoken audio — the agent's voice. Callingtap's free default is Kokoro; multilingual and cloud voices connect through your own accounts or the platform's voice providers.
VAD (voice activity detection)
Detecting whether audio contains speech at all — used to notice when a caller starts and stops talking, filter silence, and drive barge-in and endpointing.
VoIP (Voice over IP)
Voice carried over internet connections rather than phone lines. Callingtap calls are VoIP end to end until a telephony provider bridges them to the PSTN.
Webhook
An automatic HTTP message from Callingtap to your systems when something happens — a call starts, completes or fails, an appointment is booked. Callingtap signs every webhook (HMAC-SHA256) and logs every delivery attempt.
WebRTC
The browser technology that carries live audio and video without plugins. The website widget's voice runs on WebRTC through LiveKit — which is also why browser audio needs a user tap on some phones.
White-label
Rebrandable as your own product: your logo, your colours, your domain. Callingtap is built to be resold and re-skinned by agencies and platforms, not just used as-is.
Missing a term? The documentation shows most of these in action, and the Learn guides use them the way a working team would.