A2P 10DLC
Application-to-Person 10-Digit Long Code. A US regulatory framework requiring registration for businesses sending SMS. Required for all US-based Twilio numbers to prevent message blocking.Agent
An AI-driven voice agent built on PolyAI that interacts with customers in real time across voice (telephony), chat (webchat), and SMS. See Agent settings, Quickstart.Agent Studio
The primary interface for designing, configuring, and deploying AI agents. Used for creating agent logic, configuring settings, and monitoring performance. See Getting started, Build essentials.Annotation
A manual tag applied during Conversation Review to highlight issues like incorrect transcriptions or missing topics.ASR (Automatic Speech Recognition)
Converts spoken language into text. PolyAI uses ASR models optimized for conversational accuracy with support for multiple languages, accents, and industry-specific vocabulary. See Speech recognition.ASR biasing
A technique to improve speech recognition by instructing the ASR model to prioritize specific words or phrases. See ASR biasing.Barge-in
When a user interrupts the agent mid-sentence. The system detects this and stops playback to process the user’s input immediately.BYOM (Bring Your Own Model)
Integrate your own LLM by exposing an API endpoint that follows the OpenAIchat/completions schema. See BYOM.
Channel
The medium through which users interact with the agent. In custom functions, the channel is exposed asconv.channel_type with values like "sip.polyai" (voice), "webchat.polyai" (webchat), "chat.polyai" (agent chat), and "sms.twilio" (SMS). In the Conversations API, the channel field uses analytics-level labels such as VOICE-SIP and WEBCHAT. See Telephony, Webchat, SMS.
Chunking
Splitting large bodies of text into smaller pieces for retrieval. Connected knowledge sources are scraped, chunked, and matched against user input via RAG. See Connected knowledge, RAG.CSAT (Customer Satisfaction Score)
A metric collected via post-conversation voice surveys. Callers rate their experience on a 1–5 scale; responses are tracked in dashboards. Triggered usingconv.goto_csat_flow() in the end function. See Surveys (CSAT).
Connected knowledge
External knowledge sources (websites, PDFs, Zendesk) connected to your agent via Build > Knowledge > Connected. Read-only, synced from external sources, cannot trigger actions. See Connected knowledge.Containment rate
Percentage of conversations fully handled by the agent without human handoff. A key metric for agent effectiveness. See Standard dashboards.Conversation metadata
Structured data about a conversation: duration, start time, associated variant. See Conversations API, Call data.Conversation diagnosis
A debugging tool within Conversation Review showing which flows, functions, and topics were activated during a conversation. See Conversation diagnosis.Dialogue context
The conversation state maintained throughout a call, including turn history, state variables, and all context needed for processing. See Conv object, Variables.Draft
The state between the latest published version and ongoing changes. Drafts become versions upon publishing. See Environments and versions.DTMF (Dual-Tone Multi-Frequency)
Touch-tone input from phone keypads. Allows users to enter PINs, account numbers, or menu selections by pressing keys. See DTMF.Embedding
A numerical vector representation of text used for semantic search and retrieval. PolyAI computes embeddings for Knowledge topics to enable RAG-based matching. See RAG.End function
A function that finalizes an interaction: closing conversations, sending confirmations, triggering logging, or performing clean-up. See End function.Environment
The deployment stage of a project:sandbox, pre-release, or live. See Environments and versions.
Entity
Typed extracted data from user input (phone numbers, dates, names, addresses). Helps structure and validate information collected during conversations. See Flows, Variables.Event-sourced
A data architecture pattern where all changes are stored as a sequence of events. The Conversations API v3 uses this for reliable, scalable data ingestion. See Conversations API v3.Flows
A conversation logic system that guides users through structured interactions with decision points, branching logic, automated actions, and transition functions. See Flows, Examples.Function
A reusable backend operation executed during a conversation. Can retrieve external data, modify state, perform calculations, or log structured data. See Functions, Function classes.Global rules
Behavior constraints applied across all agent interactions for consistency, compliance, and tone control. Enforced by the LLM during response generation. See Rules.Handoff
Transferring a user from the AI agent to a human agent. Can be rule-based, user-requested, or AI-driven. See Call handoff.IPA (International Phonetic Alphabet)
A standardized pronunciation system used in the Pronunciations feature to define how the agent pronounces specific terms. See Voice configuration.Keyphrase boosting
Biases the ASR model toward recognizing specific words and phrases for better domain-specific transcription accuracy. See Keyphrase boosting.LLM (Large Language Model)
The AI model powering the agent’s understanding and response generation. PolyAI’s proprietary Raven model family is recommended for most deployments — purpose-built for conversational AI with superior grounding and natural speech. Third-party models (GPT, Claude) are also supported. See Model.Live
The production environment where the agent handles real customer traffic. See Environments and versions.Managed Topics
Version-controlled, editable topics in Build > Knowledge > Managed Topics. Support sample questions, content, actions (function calls, handoffs, SMS). See Managed Topics, RAG.Multi-site configuration
Customizing agent responses based on location using variant management. See Variant management, CSV imports.NLU (Natural Language Understanding)
The component responsible for interpreting user input, detecting intents, and extracting entities. See Flows, Variables.Node
A building block within a conversation flow defining what the agent says, listens for, and where the conversation goes next. See Flows, No-code flows.Out-of-domain (OOD)
User input that falls outside the agent’s configured knowledge. Tracked in analytics; may trigger fallback behaviors. See Conversation review.Policy engine
The decision-making component that controls conversation flow by executing nodes in priority order: global nodes, current node, then fallback nodes. See Flows.PolyScore
An AI-generated quality score (0–10) assigned to each eligible voice conversation. Evaluates conversation quality, task success, and customer experience. Displayed as a color-coded badge in Conversation Review and accessible via the Conversations API. See PolyScore.Pre-release
A staging environment for UAT before promoting to production. See Environments and versions.Processing
The sequence of operations on each conversation turn: ASR, language understanding, knowledge retrieval, decision making, response generation, and delivery. See Processing order, Architecture.Project
A complete agent configuration including knowledge, flows, functions, settings, and voice. Can have multiple variants. See Variant management.Pronunciations
Custom rules for adjusting how the agent pronounces specific terms using IPA and regex-based substitution. See Voice configuration.Performance metrics
Key indicators for agent effectiveness and quality. See Standard dashboards, Safety dashboard.RAG (Retrieval-Augmented Generation)
A technique where the system retrieves relevant knowledge before generating a response, so the model can ground its answer in specific content. See RAG.Ranking and retrieval
The system that computes and manages embeddings for Knowledge topics, performing vector operations and runtime matching against user input. See RAG.Raven
PolyAI’s proprietary model family for conversational AI, specialized for customer service across voice and chat. Raven 3.5 is the recommended model for all deployments, supporting both voice and chat. See Raven.Response control
Modifies how agents respond: rate-limiting, interruption handling, compliance filtering. See Response control.Safety dashboard
Monitors flagged content, risky user utterances, and safety filter performance. See Safety dashboard.Sandbox
The development environment for creating, modifying, and testing agent versions before promotion. See Environments and versions.Smart Analyst
An AI-powered analysis tool that answers natural-language questions about recent conversation data. Samples up to 500 conversations per query and surfaces insights on containment, escalation reasons, and caller sentiment. Found under Analytics > Smart Analyst. See Smart Analyst.SBC (Session Border Controller)
A network element managing and securing SIP calls, handling routing, security, and interoperability between VoIP networks. See Telephony.Secrets
Sensitive credentials or tokens stored securely and accessed by the agent for integrations. See Secrets.SIP (Session Initiation Protocol)
A signaling protocol for initiating, maintaining, and terminating voice calls over IP networks. See Telephony, Call handoff.SIP header
Metadata fields in SIP messages for routing instructions or custom information. Custom headers typically start withX-. See Call handoff.
SSML (Speech Synthesis Markup Language)
XML-based markup for controlling TTS output: pauses, pitch, emphasis. See Voice configuration.Step
A self-contained conversation state within a flow, consisting of text prompts and optional functions. See Flows, No-code flows.Stop keyword
A regex pattern that halts or logs agent responses containing specific phrases. See Stop keywords.Start function
A function triggered at the beginning of an interaction for state initialization, authentication, or conditional logic. See Start function.Transcript correction
Editing AI-generated conversation transcripts to improve ASR accuracy and training data quality. See Speech recognition.TTS (Text-to-Speech)
Converts written text into spoken language. PolyAI supports multiple providers and custom voice configurations. See Voice configuration, Choosing a voice.Turn
A single exchange in a conversation, defined as the execution between two consecutive user inputs. See Conversation review.UAT (User Acceptance Testing)
Testing in pre-release to validate agent behavior before production. See Environments and versions.Utterance
A unit of meaning that can be expressed in multiple ways (e.g. “Hello”, “Hi”, “Bonjour”). See Multilingual agents.Validation
Checking whether user-provided information is complete, accurate, or matches expected criteria within a flow. See Flows.Variable
Values stored and retrieved during a conversation viaconv.state. Persist for the duration of a single conversation. See Variables.