This glossary defines the terms you meet when you add an AI voice agent to a website, from speech recognition and WebRTC to grounding, MCP and data processing. Each entry gives the standard meaning first, then a separate note on how Kawl Now, a browser voice AI agent for SaaS websites, handles it, including what it does not do.
Speech-to-speech AI is artificial intelligence that turns spoken input into spoken output, usually with one model that works on the audio itself instead of converting speech to text and back.
Voice AI is the use of artificial intelligence to recognise, understand and produce human speech, so that people can talk to software and hear it answer.
A voice assistant is software that understands spoken requests in natural language and responds by answering, carrying out a task or passing the request to another service.
A voice user interface (VUI) is a way of using a computer, app or device by speaking to it and hearing it respond, instead of or alongside a screen, keyboard or touch.
A voice widget is a small interface, usually a button that opens a panel, that a website adds with a few lines of code so visitors can start a voice conversation from the page, with an AI voice agent or a person.
WebRTC (Web Real-Time Communication) is a set of web standards that lets browsers and apps send audio, video and data to each other in real time, without plug-ins.
A website voice agent is an AI voice agent that visitors talk to in their web browser, through a widget on a website, using their microphone rather than a phone line.
An AI agent is software that uses an AI model, usually a large language model, to decide its own next steps, such as which tool to call, while it works towards a goal set by a person.
An AI hallucination is output from a generative AI model that sounds plausible and confident but is false, or isn’t supported by the information the model was given.
A context window is the maximum number of tokens a language model can take into account at once, covering its instructions, the conversation so far and any documents added to the prompt.
Conversational AI is the set of technologies that let software hold a conversation with people in natural language, by text or by voice, working out what they mean and replying in kind.
An embedding is a vector, a list of numbers, that a machine learning model produces to represent a piece of data such as a word, a sentence or a document, so that items with similar meanings get similar vectors.
Function calling, also called tool calling, is a way for a language model to ask an application to run a named function with structured arguments, so the model can use the result in its reply.
Grounding in AI is the practice of basing a model’s response on information from specific, trusted sources supplied when the request is made, rather than only on what the model learned in training.
A large language model (LLM) is a neural network, trained on very large amounts of text, that generates language by repeatedly predicting the next token in a sequence.
Natural language understanding (NLU) is the branch of natural language processing that works out what a person means from what they say or type, such as their intent and the details they mention.
Prompt injection is an attack in which instructions are planted in text a language model processes, such as a user’s message or a retrieved document, so that the model follows them instead of the instructions its developers gave it.
Retrieval-augmented generation (RAG) is a technique in which a system retrieves relevant passages from a knowledge base and adds them to a language model’s prompt, so the model can answer from information it wasn’t trained on.
Semantic search is a search technique that ranks results by how closely their meaning matches the query, usually by comparing embeddings, rather than by whether they contain the query’s exact words.
A system prompt is the set of instructions an application gives a language model ahead of the user’s input, setting its role, tone, rules and limits for the conversation.
A vector database is a database that stores vector embeddings and finds the stored vectors closest to a query vector, which is how semantic search and many retrieval-augmented generation systems look up related content.
A virtual agent is AI software that holds customer service conversations, by chat or by voice, handling requests that a human agent would otherwise take.
An AI receptionist is software that answers a business’s phone calls with an AI voice, greeting callers, answering common questions, taking messages, booking appointments and transferring calls to staff.
Average handle time (AHT) is the average time a customer interaction takes from start to finish, including the time an agent spends talking, any time the customer is on hold and the follow-up work after the interaction ends.
Call deflection is the practice of meeting a customer’s need through self-service or another channel, such as a help centre, a chatbot or a text message, so that they don’t need to phone the support team.
A chatbot is software that holds a conversation with people, usually by text, answering their messages automatically with scripted responses or with replies generated by an AI model.
CSAT, short for customer satisfaction score, is a measure of how satisfied customers are with a product, service or single interaction, taken from a short survey that asks them to rate their satisfaction on a scale.
Customer self-service is any way for customers to get help or complete a task on their own, without contacting a member of staff, such as a help centre, an FAQ page, a community forum or an AI agent.
First contact resolution (FCR) is the percentage of customer issues that are fully resolved in the customer’s first interaction with support, with no follow-up, transfer or repeat contact needed.
A help centre is the customer-facing support website where a company publishes its self-service content, such as knowledge base articles, FAQs and guides, alongside ways to contact its support team.
Interactive voice response (IVR) is an automated telephone system that plays recorded or synthesised prompts to callers and responds to their key presses or speech, so it can route the call or let the caller complete a task without a person.
A knowledge base is an organised, searchable collection of articles that explain how a product or service works and how to solve common problems, written for customers, staff or both.
Live chat is real-time text messaging between a website or app visitor and a person on the company’s team, usually through a chat window in the corner of the page.
An API key is a unique string that a program sends with its requests to an API, so the service can tell which account or project is calling and apply that account’s access and limits.
Cross-origin resource sharing (CORS) is a browser mechanism that lets a server say, in HTTP headers, which other origins may read its responses to requests made by web page scripts.
A data processor is an organisation that handles personal data on behalf of another organisation, the controller, which decides why and how the data is used.
Data residency is the geographic location where an organisation’s data is stored and processed, usually chosen to meet legal, contractual or customer requirements.
Embed code is a short snippet of HTML, usually a script tag or an iframe, that you paste into a web page to add content or a feature served by another service.
The General Data Protection Regulation (GDPR) is the European Union law that sets the rules for collecting, using, storing and sharing personal data, and it has applied since 25 May 2018.
An MCP server is a program that offers tools, data or prompts to AI applications over the Model Context Protocol, so any compatible client can discover and use them.
The Model Context Protocol (MCP) is an open standard that lets AI applications connect to outside tools and data through servers that all speak the same JSON-RPC protocol.
The OpenAPI Specification (OAS) is an open standard for describing HTTP APIs in a JSON or YAML document, so people and tools can see every endpoint, parameter and response without reading the source code.
Personally identifiable information (PII) is information that can identify a specific person, such as a name, email address or phone number, either on its own or combined with other details.
A REST API is a web API built in the REST architectural style, where each resource has a URL and clients read or change it with standard HTTP methods such as GET, POST, PATCH and DELETE.
Row-level security (RLS) is a database feature that controls which rows of a table each user or role can read or change, using policies the database applies to every query.
A software development kit (SDK) is a package of libraries, tools, documentation and code samples that helps developers build software for a particular platform, service or programming language.
A webhook is an HTTP request that a service sends automatically to a URL you choose when a particular event happens, so your system hears about the event without asking.