Large language model (LLM)
A large language model (LLM) is a neural network trained on very large amounts of text to predict likely continuations. That lets it understand instructions, hold conversations and generate fluent language.
Also called: language model, foundation model, generative model
An LLM is the reasoning part of a modern voice agent. Given the conversation so far, a description of the business and its rules, and any information retrieved from other systems, it decides what to say next. It is not a database; it does not "know" a business’s opening hours unless they are provided to it.
That distinction matters for a buyer. The model brings language ability and general knowledge; everything specific to the business has to be supplied, in a system prompt, through retrieval, or through function calls, and kept current. A model working from stale or missing information will still answer confidently, which is where hallucination comes from.
Models differ in size, speed and cost, and a voice product has to balance them: a larger model may reason better but reply more slowly, and on a phone call the delay is audible.