Floatbot builds AI agents that handle phone calls, chat, text and email, and unusually for the category it exposes the underlying stack: RAG, cognitive search, a speech-to-text API and its own LLM layer are all named products rather than implementation details.
That transparency is useful when evaluating. Most contact centre AI vendors present a finished agent and leave you guessing how retrieval, grounding and fallback actually work – which is exactly what determines whether the agent stays accurate as your knowledge base drifts. A vendor naming its RAG and search layers is inviting the technical questions rather than deflecting them.
There is both a self-service builder and agent assist for live humans, spanning voice and digital in one platform. As always with voice agents, the deciding test is failure behaviour: what happens on an accent it mishandles, a question outside the knowledge base, or a caller who simply wants a person. Evaluate the escalation path as carefully as the happy one.
