Overview
This page defines key terms used throughout Sulus documentation, grouped into core AI/speech concepts, call types, integration/infrastructure terms, pricing and compliance terms, and performance terms.
Glossary
Assistant SupportGlossary: Key Platform Terminology
Core AI & Speech Terms
| Term | Definition |
|---|---|
| Large Language Model (LLM) | A machine learning model trained on large amounts of text, used to generate text in a probabilistic manner, token-by-token. |
| Inference | The process of running a prompt against an LLM to receive text output; often called "running inference." |
| STT | Speech-to-text — the process of converting physical sound waves into raw transcript text (transcription). |
| TTS | Text-to-speech — the process of converting raw text into playable audio data. |
| Speech Endpointing | The process of detecting the start and end of a line of speech in an audio signal, a core function of conversation turn detection. A basic heuristic is silence detection; a more robust approach also weighs conversation state and intent to judge whether a caller is pausing or finished speaking. See Voice Pipeline: Endpointing & Turn Detection for the full mechanics. |
| Backchanneling | Verbal or non-verbal feedback a listener gives a speaker during a conversation — expressions like "yeah," "OK," "uh-huh," "hmm," or "right." This feedback usually isn't semantically significant; it signals attention, understanding, or agreement rather than new information. |
Call Types
| Term | Definition |
|---|---|
| Inbound Call | A call received by an assistant from an external caller — the assistant is the one answering, and the call comes inward to a number. |
| Outbound Call | A call made by an assistant to a target phone number — the assistant is the one dialing, and the call goes outward to another number. |
Integration & Infrastructure Terms
| Term | Definition |
|---|---|
| Server URL | An endpoint you expose to the platform to receive conversation data in real time. Unlike a traditional webhook, a server URL can reply with a meaningful response that changes what happens next in the call — the transfer-destination-request event is one example, where your response supplies the actual transfer destination. |
| Webhook | A server endpoint you expose to receive external data in real time. Traditionally, webhooks are unidirectional and stateless — the endpoint only replies with a status code to acknowledge receipt, without its response changing the caller's behavior. |
| SDK | Software Development Kit — a pre-packaged library and set of platform-specific tools that speeds up integration for developers. See Web SDK Integration, Flutter SDK Integration, React Native SDK Integration, iOS SDK Integration, and Python Server SDK for the platform-specific guides. |
Pricing & Compliance Terms
| Term | Definition |
|---|---|
| At-Cost | The platform's practice of passing through STT, LLM, and TTS provider charges without markup — you're billed at-cost for what those underlying providers charge, meaning "without profit to the seller" on that specific line item. |
| Telemarketing Sales Rule (TSR) | A US Federal Trade Commission (FTC) regulation governing telemarketing calls, separate from the FCC's TCPA. It's mentioned here as related context; for the full outbound-calling compliance picture the platform documents in detail, see TCPA Consent & Outbound Calling Compliance and Recording Consent Plan. |
Performance Terms
| Term | Definition |
|---|---|
| Voice-to-Voice Latency | The time from a caller finishing speech to the assistant's first audio chunk being played back. The target is under 1 second, with 500–700ms considered a strong result. Responding too fast can feel unnatural as well, so this metric is watched in both directions rather than simply minimized. |
In summary: these terms recur throughout the rest of the documentation, and this page exists as a single place to look one up rather than re-deriving it from context on whichever page happens to mention it.