Glossary

This glossary defines Flowtative terms as they appear in the interface. It is the reference for the whole documentation. The matching French term is given in parentheses.

Reference terms

These terms are used consistently across the documentation.

Term	French	Definition
Agent	Agent	AI assistant configured in Flowtative: a language model, a system prompt, and optionally a knowledge base, tools, a linked flow and a voice. An agent replies in writing (chat) or by phone (callbot). Not to be confused with the Agent role on the Team page, which refers to a human member.
Flow	Flow	Conversation journey drawn on a canvas, made of blocks connected by links. A flow structures what the agent does step by step: greeting, identification, routing, transfer, end of call.
Block	Bloc	Basic step of a flow, for example Trigger, LLM agent, Message, Datastore, SMS, Human validation or Hang up. Each block has inputs, outputs and settings.
Tool	Outil	Action the agent can trigger during a conversation: search a knowledge base, call an external service, transfer a call. Tools are managed in Settings > Tools and linked to agents.
Knowledge base	Base de connaissances	Set of indexed documents (text, PDF, CSV, web pages) where the agent looks up information to answer. Managed in Knowledge.
Datastore	Datastore	Table of structured data (typed fields and records) that agents and flows can read and write: customers, contracts, appointments. Managed in Datastores.
Campaign	Campagne	Series of outbound calls: an agent calls the contacts of a contact base within a calling window, with retries. The outcome of each call is written back to the contact's record.
Contact base	Base de contacts	Datastore created with the Contact base type, used as the call list of a campaign. It has a phone field in international format and a unique deduplication field. Its columns can be used as variables by the agent.
Retry	Relance	New call attempt to a contact who was not reached during a campaign, according to the campaign's retry rules. Depending on the setting, a voicemail can trigger a retry or close the contact.
Caller ID	Numéro présenté	Phone number shown to the contact on an outbound call.
Automation	Automatisation	Running an agent or a flow without inbound interaction: on a schedule, or on a call from an external system.
Approval	Validation	Request sent to a human by a flow (Human validation block) before continuing: approve, edit or reject a proposed reply or action. Approvals are handled in Approvals.
Transcript	Transcription	Written history of a past conversation (chat or call), with the audio recording when enabled. Viewed in Transcripts.
Evaluation	Évaluation	Scoring of a conversation's quality against an evaluation grid, performed by a judge agent. Managed in Evaluation.
Team	Équipe	The members of your workspace, their role and their status (active, invited, suspended). Managed in Team.
Settings	Paramètres	Settings for your profile and the workspace: webhooks, connectors, notifications, tools, MCP servers, skills, telephony.

Agents and models

LLM (language model) (LLM) — AI model that generates the agent's replies. It is chosen with the agent's Provider and Model fields.
System prompt (Prompt système) — The agent's background instructions: its role, behavior and guardrails. It can contain variables.
Variable (Variable) — Value inserted into a prompt at conversation time, written between double curly braces, for example {{company}}. In the tester, Test variables let you give them a value. See Variables and dynamic data.
Template (Template) — Ready-to-customize agent model (for example Accueil / standard, Support N1 (RAG), Callbot vocal), offered when you create an agent.
Creation assistant (Assistant de création) — Conversational assistant that creates an agent from a natural-language description: it asks its questions, proposes a plan, then creates the base, the tools, the flow and the bot.
RAG (RAG) — Searching a knowledge base before replying. Turned on in the agent's Knowledge (RAG) section with Enable RAG.
Tool calls (Appels d'outils) — The agent's ability to trigger tools during a conversation (function calling).
ReAct agent (Agent ReAct) — Reasoning mode in which the agent alternates thinking and acting in a loop.
Memory (Mémoire) — Keeping the conversation context from one message to the next. The Memory mode is Rolling (forgets the oldest messages) or Strict (keeps everything and refuses when the window is full).
Structured output (JSON) (Sortie structurée (JSON)) — The agent replies according to a defined schema, for example to qualify a lead (budget, need, timeline).
Temperature (Température) — Setting for how much replies vary: the lower it is, the more consistent the replies.
Token (Token) — Unit of text processed by the model. Max tokens limits reply length; the Conversation window sets how much of the conversation the model sees.
Skill (Skill) — Instruction loaded on demand by the agent. The prompt only holds the skill catalog (name and description). Managed in Settings > Skills and attached in the agent's Skills section.
Linked flow (Flow lié) — Flow associated with an agent, with the version to run.
Debug mode (Mode debug) — Detailed conversation log (variables, nodes, writes), useful to diagnose a flow. It may contain personal data.
Audit log (Journal d'audit) — Capture of conversation transcripts (tokens, latency, tools) that feeds observability.

Voice and telephony

Callbot (Callbot) — Voice agent that answers the phone. In the editor, the Voice · callbot part groups its settings; it is turned on with Enable voice (callbot).
STT (transcription) (STT — transcription) — Converting the caller's speech into text.
TTS (voice synthesis) (TTS — synthèse vocale) — Converting the agent's text into speech. The TTS engine and the Voice are chosen in the TTS — synthesis section.
Text substitutions (Substitutions de texte) — Rules that correct how voice synthesis pronounces certain words.
Turn-taking (Tour de parole) — Detecting when the caller has finished speaking, which triggers the agent's reply.
Barge-in (Interruption) — The caller's ability to interrupt the agent while it speaks.
Realtime (Temps réel) — End-to-end audio session with a voice model, without separate STT and TTS steps.
End of call (Fin d'appel) — Conditions under which the agent ends the conversation and hangs up.
DTMF (DTMF) — Phone keypad keys pressed by the caller.
DID (SDA) — Direct phone number assigned to your workspace and connected to an agent. Managed in Settings > Telephony & Trunk.
Inbound number (Numéro entrant) — Number that hands calls over to an agent, shown in the agent's Identity & prompt section.
SIP trunk (Trunk SIP) — Phone link between your infrastructure and Flowtative, through which your numbers' calls travel.
Calling window (Créneau d'appel) — Days and time ranges during which a campaign may call. A preset restores the French legal window.

Flows

Trigger (Déclencheur) — First block of a flow: the AI's first speaking turn, when the call is answered or the chat opens.
Hang up (Raccrocher) — Block that ends the conversation, with or without a farewell message, and can transfer the call to an advisor.
Autonomous flow (Flow autonome) — Flow that runs with no one on the other end and can be targeted by an automation. It is marked with Make autonomous in the editor.
Version (Version) — Saved state of a flow. Version history lets you preview an older version and restore it; restoring creates a new version, without deleting anything.
MCP server (Serveur MCP) — External service that makes tools available to your agents. Declared in Settings > MCP servers.

Monitoring and quality

Analytics (Analytics) — Your agents' activity indicators over a period.
Observability (Observabilité) — Operational health of agents: volume, cost, latency, tool reliability and error rate.
Trace (Trace) — Chronological, step-by-step detail of a conversation's execution.
Supervision (Supervision) — Live monitoring of ongoing calls.
Evaluation grid (Grille d'évaluation) — List of criteria used to score a conversation.
Judge agent (Agent juge) — Agent that applies an evaluation grid to conversations.

Workspace and access

Workspace (Espace) — Your organization's Flowtative environment, which you access with your organization identifier.
Role (Rôle) — Profile assigned to a team member, with a description shown in the application: Administrator, Service admin, Supervisor, Agent or Reader.
Webhook (Webhook) — Automatic delivery of a session's full transcript to a web address in your system, when the session ends. Managed in Settings > Webhooks.
Connector (Connecteur) — Link between Flowtative and a third-party application used by your company.
Command palette (Palette de commandes) — Search and quick-action window, opened with ⌘K / Ctrl+K.