Every part an agent is assembled from, one section per term. Each entry gives the definition, the ways people ask about it, and the page that documents it in full.

Agent layers

Global prompt

The always-on layer shaping identity, tone and guardrails on every turn. It runs regardless of which playbook or workflow is active, and is written in four default sections: #Role, #Goal, #Tone and #Guardrails.

Also asked as:

the instructions that apply to every single reply
set the bot’s personality and house rules once
the system prompt for the whole assistant

Related: Instructions · Guardrails · System prompt

Documented in: Global prompt › What is the global prompt? · Global prompt › Best practices · Global prompt › #Role · Global prompt › #Goal · Global prompt › #Tone

Instructions

The routing layer: which playbook or workflow handles what. Where the global prompt defines how the agent behaves, instructions define what it does with a request.

Also asked as:

tell the assistant which skill handles which kind of request
the decision layer that picks a playbook
where the agent’s routing rules live

Related: Global prompt · Trigger · Reference

Documented in: Instructions › What are agent instructions? · Instructions › Writing agent instructions

Start message

The first thing the agent says. An agentic start message is written under a # Starting Message heading in the instructions; a fixed greeting is a Message step in the initialization workflow with Wait for user input turned on.

Also asked as:

the greeting the bot sends before the visitor types anything
open every chat with the same welcome line
make the assistant speak first

Related: Instructions · Initialization workflow

Documented in: Instructions › Adding a start message · Initialization workflow › Static start message

Trigger

What tells the agent when a playbook or workflow applies. Older material calls this the “LLM Description”; the current name is Trigger. System tools carry one too, describing what the tool does and when to use it.

Also asked as:

describe when a skill should be used
how does the model know which playbook to pick
the description the router reads

Related: Instructions · Playbook · System tools

Documented in: Playbooks › Creating playbooks · Querying the knowledge base › Trigger

Behaviour

The model, voice and memory settings an agent runs under: default model, memory, timezone, voice output and input, session timeouts, chat persistence, events and the global no-match and no-reply responses.

Also asked as:

the settings page for the model, the voice and memory
where do I change what the agent runs on
agent-wide settings rather than per-playbook ones

Related: Default model · Memory · Voice output

Documented in: Behaviour

Agent tools

The tools attached to the agent itself rather than to one playbook. Also called global tools: APIs, integrations, MCP servers and functions added at the agent level, so every playbook can use them without adding each one individually. They apply in agentic contexts only: a workflow calls tools through its own steps.

Also asked as:

make one tool available in every playbook
attach a tool at the agent level
share a tool across all skills

Related: Tool · System tools · Playbook

Documented in: Global tools › Adding a global tool · Global tools › When are global tools available

Global tools

The tools added at the agent level — APIs, integrations, MCP servers and functions — so every playbook can use them without adding each one individually. The same thing as Agent tools; the Studio page is titled Global tools.

Also asked as:

add a tool once for every playbook
the agent-level tools list

Related: Agent tools · Tool

Documented in: Global tools › Adding a global tool

Guardrails

Non-negotiable rules the agent must always follow, written under a # Guardrails heading in the global prompt. Models are tuned to pay extra attention to that section. Good guardrails are specific and actionable; “be professional” is a tone instruction, not a guardrail.

Also asked as:

rules the bot must never break
hard limits on what the assistant may say or do
where do I put the things the agent is not allowed to do

Documented in: Global prompt › #Guardrails

Preventing hallucination

A guardrail pattern: the agent says it does not have the information instead of guessing, never combines partial knowledge base results with assumptions, and never invents a plausible reply when a tool call fails.

Also asked as:

the assistant invents facts that are nowhere in our content
keep replies to what the knowledge base actually says
the bot confidently gives wrong information
the assistant invents answers our documents never gave
stop the assistant from guessing
only answer from the documents we uploaded

Related: Guardrails · Hallucination · Grounding

Documented in: Global prompt › #Guardrails

Protecting sensitive data

A guardrail pattern: never reveal internal pricing logic, margins or cost structures, never share one customer’s information with another, never read back full card numbers, SSNs or passwords.

Also asked as:

stop the assistant from leaking private details
make sure the bot never reads out a card number
keep internal pricing out of replies

Related: Guardrails · PII redaction

Documented in: Global prompt › #Guardrails

Staying in scope

A guardrail pattern that keeps the agent to its domain: no legal, medical or financial advice, no comments on competitors, and an offer to connect with the right team when the conversation drifts outside it.

Also asked as:

keep the assistant on topic
refuse questions unrelated to our product
the bot answers things it has no business answering
the assistant answers questions that have nothing to do with our business
keep the bot from going off topic
refuse unrelated requests politely

Related: Guardrails · Global no-match

Documented in: Global prompt › #Guardrails

Transaction safety

A guardrail pattern for actions with consequences: no refund, cancellation or account change without explicit customer confirmation, business rules never overridden, identity verified before account data is touched.

Also asked as:

make the bot ask before it does anything irreversible
require an explicit yes before a refund goes through
verify who the customer is before changing their account
the bot confirmed a refund without double-checking
ask the customer to confirm before acting

Related: Guardrails · Initialization workflow

Documented in: Global prompt › #Guardrails

Writing a strong global prompt

The practices the docs recommend: keep it short (100–1000 words; every word runs on every turn), be direct (declarative rules, not hedging), keep it static so the model can reuse its cache, and keep step-by-step procedures out of it — those belong in instructions, a playbook or a workflow.

Also asked as:

how long should the global prompt be
my prompt is huge and replies got slower
what belongs in the global prompt versus a playbook

Related: Global prompt · Instructions · Playbook

Documented in: Global prompt › Keep it short · Global prompt › Be direct · Global prompt › Variable Consistency · Global prompt › Common mistakes · Global prompt › How long should my global prompt be?

Playbooks and workflows

Framework

The choice made when a project is created. Agentic (recommended): the agent uses playbooks, workflows, tools and reasoning to navigate conversations on its own. Conversation flow: you design the path in the visual builder and the agent follows it step by step.

Also asked as:

difference between an agentic project and a conversation flow project
should I let the AI decide or draw every branch myself
which project type to pick when starting

Related: Playbook · Workflow

Documented in: Choosing a framework › Agentic vs. Conversation flow

Playbook

Agentic. Instructions driving toward a goal, calling tools as needed. Use for open-ended work: booking, qualifying, handling a return.

Also asked as:

an AI skill that works toward a goal using tools
let the model handle an open-ended task on its own
the agentic alternative to a drawn flow

Related: Workflow · Crew · Playbook step · Entry conditions

Documented in: Playbooks · Playbooks › Writing good instructions · Build (API)

Workflow

Deterministic. Steps drawn on a canvas from the Start chip. Use where the path must be fixed: authentication, IVR menus, scripted onboarding.

Also asked as:

a fixed step-by-step flow on the canvas
a scripted path the conversation must follow exactly
the deterministic alternative to a playbook

Related: Playbook · Step · Workflow step · Nesting workflows

Documented in: Workflows · Workflows › Steps

Step

One node on a workflow canvas. Named with only the first word capitalised: Message step, API step, Function step. Steps come in four categories: agentic, conversation, logic and tools.

Also asked as:

a block on the canvas
the building pieces of a flow
the individual pieces I drag onto the canvas
what each box in a workflow does

Related: Workflow · Actions

Documented in: Workflows › Steps

Actions

Steps attached to a port without being drawn on the canvas, keeping the canvas clean without losing any functionality.

Also asked as:

attach something to a port without cluttering the canvas
hide small steps behind a connection
tuck a small step behind a connector
keep the canvas tidy

Related: Step

Documented in: Workflows › Actions

Nesting workflows

Calling one workflow from another with the Workflow step, so a process is built once and reused.

Also asked as:

reuse a flow inside another flow
call a sub-flow from the main one
break a big process into smaller flows
a workflow that calls another workflow

Related: Workflow step

Documented in: Workflows › Nesting workflows

Entry conditions

Deterministic checks that must evaluate to true before the agent can route to a playbook or workflow. While they fail, it stays unavailable. They apply only when the agent routes there directly.

Also asked as:

only allow a skill once a condition is met
gate a playbook behind a check
the agent should not route to a flow until something is true

Related: Required variables · Exit conditions

Documented in: Playbooks › Entry conditions · Workflows › Entry conditions

Required variables

Variables that must hold a value before the agent can route into a playbook or workflow. Until every one is filled, the playbook or workflow stays unavailable.

Also asked as:

don’t start the booking skill until we know the customer’s email
a playbook that needs certain details before it can run
force the agent to collect information first

Related: Entry conditions · Variable

Documented in: Playbooks › Required variables · Workflows › Required variables

Exit conditions

Situations in which a playbook hands control back to the workflow through a named path. They apply when the playbook runs through a Playbook step (or globally on a Crew step); when the agent routes to the playbook directly they are ignored.

Also asked as:

leave a skill and carry on with the flow when something happens
get out of the AI part back into the fixed steps
a playbook inside a workflow that should stop under a condition

Related: Playbook step · Crew · Entry conditions

Documented in: Playbooks › Exit conditions · Playbook step › Exit conditions · Crew › Exit conditions

Initialization workflow

A workflow that runs before the agent takes control of a conversation: load context from your systems into variables, send a fixed greeting, verify identity, or collect required details up front.

Also asked as:

run something before the first message
check who the visitor is before the assistant speaks to them
fetch account data at the start of every session

Related: Workflow · Start message · Transaction safety

Documented in: Initialization workflow › Setting an initialization workflow · Initialization workflow › Authentication · Initialization workflow › Loading context · Initialization workflow › Onboarding

Crew

Groups several playbooks so they hand off to each other inside one step, rather than routing through a separate agent. Lower latency than routing out and back. A root playbook is triggered first; exit conditions are global to the crew.

Also asked as:

several skills working together in one step
pass control between playbooks without going back to the router
a team of sub-agents inside a workflow

Related: Playbook · Exit conditions

Documented in: Crew › Using the Crew step · Crew › Configuration

Legacy steps

The steps and concepts from the V3 canvas that no longer exist in V4, and what replaced each: the AI Response step (now the Message step’s prompt mode), the AI Set step, NLU intent classification, flows/topics/components, explicit intent triggers, and the Choice step (now Buttons or a Playbook step).

Also asked as:

which old steps are gone in the new version
what happened to the AI Response step
migrating a V3 agent to V4

Related: Message step · Buttons step · Operator step

Documented in: V4 migration guide · V4 migration guide › AI Response Step (Deprecated) · V4 migration guide › Choice Step (Legacy)

Steps

API step

Makes an HTTP request from inside a workflow, at a fixed point in the path rather than when a model decides to. It runs an API tool with its input variables mapped explicitly.

Also asked as:

call our backend at a set point in the flow
run a saved request without letting the model choose when
fetch data from our API in the middle of a flow
make an HTTP request from the canvas

Related: API tool · Workflow

Documented in: API step › Using the API step · API step › Configuration

Buttons step

Presents a fixed set of choices instead of open input, so the next turn is one of a known few. Each button can carry a URL, and a no-match path handles a reply that matches no label.

Also asked as:

give the visitor a fixed set of choices to click
present options as clickable buttons in a flow
a menu of choices instead of free typing
show a disclaimer with buttons before continuing

Related: Buttons tool · Card step

Documented in: Buttons › Using the Buttons step · Buttons › Configuration

Call forward step

Transfers a live voice call to another number, in international format, with optional extensions.

Also asked as:

send the caller on to a different number
transfer a phone call from the flow
forward the call to our reception line

Documented in: Call forward › Configuration

Card step

One rich message: an image, a title, a description and optional buttons.

Also asked as:

one image with a heading, some text and buttons
show a single rich message with a photo in it
a product tile with a call to action

Related: Carousel step · Cards tool

Documented in: Card › Using the Card step · Card › Configuration

Carousel step

Several cards the user can browse through in one turn.

Also asked as:

a row of items to swipe through
show many products side by side
scrollable set of cards

Related: Card step · Carousels tool

Documented in: Carousel › Using the Carousel step · Carousel › Configuration

Code step

Runs JavaScript inline at one point in a workflow. A function is the reusable form of the same idea. It cannot call external servers; use the API step or a function for that.

Also asked as:

run a bit of JavaScript in the middle of a flow
inline script without creating a whole function
do some math or string formatting on a variable

Related: Function · Set step

Documented in: Code › Using the Code step · Code › Limitations

Condition step

Branches a workflow on a rule you write: a variable, a comparison operator and a value per path. The Operator step is the model-driven alternative.

Also asked as:

branch on a variable’s value
if this then that inside a flow
send people down different paths based on their data
go down a different path depending on what the user typed
split the flow based on the reply
route people to different branches

Related: Operator step · Variable

Documented in: Condition › Using the Condition step · Condition › Configuration

End step

Ends the conversation, optionally with a final message, static or generated from a prompt.

Also asked as:

finish the chat with a closing message
close the session at a certain point in the flow
hang up or end the chat when the flow is done
the last step of a workflow

Related: End tool

Documented in: End › Using the End step · End › Configuration

Function step

Runs a function tool at a fixed point in a workflow: inputs are mapped from variables, outputs are written back to variables, and the returned path decides where the flow goes next.

Also asked as:

run my custom code at one point in the flow
map a function’s result into variables
use my custom function at a point in the workflow
branch on what my code returns

Related: Function · Function path

Documented in: Function › Using the Function step · Function › Configuration

Integration step

Calls a third-party service such as Salesforce, Shopify or Zendesk from a fixed point in a workflow, with the tool’s required inputs filled from variables.

Also asked as:

create a ticket in our helpdesk from the flow
run a connector action at a set point
log a Salesforce case from a workflow
send a Gmail message at a fixed point in the flow

Related: Integration

Documented in: Integration › Using the Integration step

Listen step

Waits for the user to say or type something before continuing, and saves the reply to a variable.

Also asked as:

wait for an answer and store what they said
capture the visitor’s reply into a variable
pause the flow until the user responds

Related: Variable · Buttons step

Documented in: Listen › Using the Listen step · Listen › Configuration

Live agent handoff step

Hands the conversation to a human in a contact center such as Zendesk, Kustomer, Genesys or Ujet, with file exchange in both directions and provider-specific settings in a JSON metadata field.

Also asked as:

hand the chat to a person
escalate to a support rep
transfer the conversation into our helpdesk
let a person take over from the bot
route angry or frustrated customers to a human
hand off to Zendesk or another contact center

Related: Call forward step · Transaction safety

Documented in: Live agent handoff › Core functionality · Live agent handoff › Configuration · Live agent handoff › Handing off to Zendesk

Handoff metadata

The optional JSON settings on the Live agent handoff step — noteBodyText (an internal note for the human agent), disableFileUpload (hide the file control), waitTimeMessage (how the expected wait is communicated) and other provider-specific fields — with variables allowed in any value.

Also asked as:

pass an escalation reason to the human agent
stop customers sending files during a handoff
tell the user how long they will wait for an agent

Related: Live agent handoff step · File uploads

Documented in: Live agent handoff › Configuration · Live agent handoff › disableFileUpload · Live agent handoff › waitTimeMessage · Live agent handoff › noteBodyText · Live agent handoff › Provider-specific capabilities

MCP step

Runs one tool from a connected MCP server from inside a workflow, where the flow reaches it, with inputs mapped explicitly rather than chosen by the agent.

Also asked as:

use a tool from an MCP server inside a flow instead of letting the model decide
always run one MCP tool at the same place in the process
run an MCP tool as part of a fixed process
call a specific MCP tool from the canvas

Related: MCP tool · Workflow

Documented in: MCP › Using the MCP step · MCP tool › In a workflow

Message step

Says something to the user, either scripted (with optional variants) or generated from a prompt.

Also asked as:

send text to the visitor
have the bot say something at this point
a generated reply inside a fixed flow

Related: Start message

Documented in: Message › Using the Message step · Message › Configuration

Operator step

A silent, model-driven branch. Evaluates conditions in natural language and routes accordingly, producing no output the user sees.

Also asked as:

let the model decide which branch to take
AI routing without writing comparison rules
classify the situation and pick a path

Related: Condition step · Playbook step

Documented in: Operator › Using the Operator step · Operator › Configuration

Playbook step

Runs a playbook inside a workflow. Exit conditions decide when control returns to the flow, and Listen for other triggers lets events, keypad input and a switch to the agent interrupt it.

Also asked as:

put an AI skill in the middle of a fixed flow
embed a playbook between two scripted steps
go from a drawn flow into the agentic part and back
a scripted flow that then hands over to an AI playbook
show a disclaimer and buttons first, then let the agent take over

Related: Playbook · Exit conditions · Crew

Documented in: Playbook step › Using the Playbook step · Playbook step › Configuration · Playbook step › Exit conditions

Set step

Assigns a value to a variable: a literal, a prompt-generated value, or an expression.

Also asked as:

store something in a variable at this point
update a value during the flow
set a variable to a value in a workflow
save a computed value for later steps

Related: Variable · Code step

Documented in: Set › Using the Set step · Set › Configuration

Workflow step

Calls another workflow from the current one.

Also asked as:

jump into another flow and come back
reuse a common sequence
call one flow from inside another
chain workflows together

Related: Nesting workflows

Documented in: Workflow › Using the Workflow step

Tools

Tool

Something a playbook can call mid-conversation: functions, API tools, MCP tools, integrations, and the built-in system tools. Each also has a step form for use at a fixed point in a workflow.

Also asked as:

things the assistant can do besides talking
let the agent take an action in another system
what can the agent actually do besides chat
actions the assistant can take in other systems

Related: Agent tools · System tools

Documented in: Agent tools

Function

Custom JavaScript that runs in Voiceflow’s sandbox and returns through a named path. Code runs remotely in an isolated sandbox; it communicates back only through its return value (outputVars, next, trace) and can make HTTP requests with fetch.

Also asked as:

write custom code the agent can run
a JavaScript tool that calls an external API
return different outcomes from my own code

Related: Function step · Function path · Function variable · Code step

Documented in: Function tool › Creating a Function tool · Function tool › How function code works · Function tool › Making API requests with fetch

Function tool

A function exposed to a playbook as something it can call. The agent calls it on its own when the conversation requires it; in a workflow the same function runs through a Function step.

Also asked as:

let the assistant run my custom code when it needs to
expose a function to a skill
give a playbook access to my JavaScript
custom logic the agent can invoke

Related: Function · Function step

Documented in: Function tool › Using the Function tool · Function tool › In a playbook

Function path

One of the named exits a function can return through. Paths are what let a function branch rather than just return a value. They apply in Function steps only, not inside a playbook.

Also asked as:

make my function send the flow one way or another
success and failure exits from custom code
multiple outcomes from one function
route the workflow based on my code’s result

Related: Function · Function step

Documented in: Function tool › next: choosing a path

Function variable

An input or output a function declares. Inputs are passed in when it runs (args.inputVars); outputs are written back for later steps to read (outputVars).

Also asked as:

pass a value into my code and get one back
read the function’s result in a later step
declare inputs and outputs for my function
what a function receives and returns

Related: Function · Variable

Documented in: Function tool › Accessing input variables · Function tool › outputVars: setting variables

API tool

A saved HTTP request an agent can make, tested from the editor and used in a playbook (the agent decides when) or a workflow (the API step, at a fixed point).

Also asked as:

call a REST endpoint from the assistant
make an HTTP request to our own backend
connect the bot to an API that has no prebuilt connector

Related: API step · API tool variable · Asynchronous tool call · Secret

Documented in: API tool › Creating an API tool · API tool › Testing the API tool · API tool › Using the API tool

API tool variable

An input an API tool substitutes into its request, so one saved request can serve many calls.

Also asked as:

put a variable into the request URL or body
parameterise a saved request
a placeholder in the request that changes per call
dynamic values in an API request

Related: API tool

Documented in: API tool › Creating an API tool

Asynchronous tool call

Running an API or Function tool in the background so the conversation continues immediately: fire and forget when the response is not needed, deferred response when it still matters. The async toggle is set per instance, not per tool.

Also asked as:

don’t make the visitor wait for a slow request
run a call in the background and carry on talking
log something without holding up the reply

Related: API tool · Function

Documented in: API tool › Running API tools asynchronously · Function tool › Running Function tools asynchronously

Integration

Calls a third-party service such as Salesforce, Shopify or Zendesk through a prebuilt connector: Airtable, Gmail, Google Sheets, HubSpot, Make, Salesforce, Shopify, Twilio and Zendesk each expose a set of actions.

Also asked as:

connect our CRM or helpdesk to the assistant
a prebuilt connector to a third-party app
send an email or create a ticket from the bot

Related: Integration step · Tool

Documented in: Agent tools › Available integrations · Agent tools › Setting up an integration · Airtable tool · Gmail tool · Google Sheets tool · HubSpot tool · Make tool · Salesforce tool · Twilio tool · Zendesk tool

System tools

Native capabilities built into Voiceflow that need no setup: knowledge base search, web search, buttons, cards, carousels, skip turn, end and call forward. Enabled by a toggle at the agent level (then available in every playbook) or per playbook, where a local override can change a tool’s settings and trigger.

Also asked as:

the built-in tools the assistant comes with
switch on web search or buttons for the agent
change a built-in tool’s behaviour in one playbook only

Related: Knowledge base tool · Buttons tool · Web search · Skip turn · End tool

Documented in: System tools › Enabling system tools · System tools › Local overrides

Knowledge base tool

The system tool that searches the knowledge base to answer a question. Configurable: the trigger, a custom query, query re-writing, the chunk limit (1–10, default 3), metadata filtering, tool messages, and source URLs in the reply.

Also asked as:

make the assistant answer from our uploaded documents
control how many passages the bot pulls per question
show where an answer came from

Related: Knowledge base · Chunk · Metadata

Documented in: System tools › Knowledge base · Querying the knowledge base › Configuring the knowledge base tool

Buttons tool

The system tool that lets the agent generate clickable options in its replies, so users choose rather than type.

Also asked as:

let the AI offer quick-reply choices on its own
generated buttons in an agentic reply
the assistant suggests options to tap
AI-generated choice buttons

Related: Buttons step · System tools

Documented in: System tools › Buttons

Cards tool

The system tool that lets the agent build rich cards — title, description, image, action buttons — from the conversation context.

Also asked as:

the assistant shows a product tile it composed itself
dynamic cards rather than a fixed Card step
let the agent show a rich card with an image and buttons
cards generated during the conversation

Related: Card step · Carousels tool

Documented in: System tools › Cards

Carousels tool

The system tool that lets the agent show several cards in a horizontally scrollable carousel when it needs to present multiple options at once.

Also asked as:

the bot lists several items to swipe through
generated carousel of results
the agent shows several products to scroll through
a swipeable set of options in an agentic reply

Related: Carousel step · Cards tool

Documented in: System tools › Carousels

Web search

A system tool that lets the agent search the web for current information during a conversation.

Also asked as:

search the internet when the answer isn’t in our content
give the bot live web results
up-to-date information from outside the knowledge base
let the assistant look something up on the web
answer with current information from the internet
browse for facts the knowledge base doesn’t have

Related: Knowledge base tool · System tools

Documented in: System tools › Web search

Skip turn

A system tool that lets the agent stay quiet and wait for the user instead of replying — when someone says “hold on” or “give me a second”.

Also asked as:

the bot should wait silently when told to hold on
don’t reply until the person is ready
pause without sending a message

Related: System tools · Global no-reply

Documented in: System tools › Skip turn

Call forward tool

The system tool that forwards the active call to a phone number or SIP address when the agent decides to. Voice projects only.

Also asked as:

let the assistant transfer the call when it judges it necessary
agentic transfer to a human line
transfer the call to a person when the agent decides
agentic call transfer to a SIP address

Related: Call forward step

Documented in: System tools › Call forward

End tool

A system tool that lets the agent end the conversation when it judges it complete.

Also asked as:

let the assistant close the chat itself
the bot decides the conversation is over
the agent ends the call on its own
let the bot wrap up the conversation when done

Related: End step

Documented in: System tools › End

Behaviour settings

Default model

The LLM that powers the agent: the model, its temperature and its max token limit. Overridable per playbook.

Also asked as:

change which AI model the assistant uses
switch to a different provider’s model
set the temperature for the whole agent
does a bigger model cost more per message
which model burns the most credits

Related: Model · Temperature · Outage protection

Documented in: Behaviour › Default model · Playbooks › Model settings

Memory

How many conversation turns the agent keeps in context (10–100, default 50), stored in the {vf_memory} variable. Higher values give the agent more history to work with but increase latency and token usage.

Also asked as:

how far back does the assistant remember
the bot loses track of earlier parts of the chat
previous turns getting dropped from context
the bot forgets things said earlier in the conversation
how much of the chat history does the assistant keep
it stops remembering after a while

Related: Context window · Chat persistence · Behaviour

Documented in: Behaviour › Memory

Faster processing

Priority processing for supported models. Uses more credits; off by default.

Also asked as:

make replies come back quicker
reduce model latency at extra cost
responses take too long to come back
speed up slow replies

Related: Default model · Credits

Documented in: Behaviour › Faster processing

Outage protection

Automatic detection of a model provider outage, with a fallback LLM provider you configure per provider.

Also asked as:

what happens if the model provider goes down
fallback model during an outage
switch to another model provider automatically when one is down
keep the agent running through a provider incident

Related: Default model

Documented in: Behaviour › Outage protection

Timezone

The timezone for {vf_now} and the other built-in time variables, and for any time-aware behaviour: the agent’s internal clock.

Also asked as:

the bot gets the current time or date wrong
set which clock the assistant uses
the agent thinks it’s a different time of day
make date and time values match our region

Related: Built-in variables

Documented in: Behaviour › Timezone

Voice output

The text-to-speech side of a voice agent: the provider (ElevenLabs, Cartesia, Rime, Google Chirp, Amazon, Microsoft), the voice, the synthesis model, real-time sync of audio and text, and voice tuning. Also available in chat projects through the widget’s voice mode.

Also asked as:

change the voice the assistant speaks with
pick a different text-to-speech provider
make the voice sound more natural or slower

Documented in: Behaviour › Voice output · Behaviour › Provider, voice, and model · Behaviour › Voice tuning · Behaviour › Sync audio and text output

Pronunciation dictionary

Rewrites words or phrases in the agent’s responses before they are sent to text-to-speech, so names, places and product terms are pronounced correctly.

Also asked as:

the voice mispronounces our brand
teach the assistant how to say a name
fix how a word is spoken on calls
the voice says a customer’s name wrong
our product name is pronounced incorrectly on calls
correct how a word sounds in the spoken reply

Related: Voice output · Keywords

Documented in: Behaviour › Pronunciation dictionary

Background audio

Optional ambient audio and audio cues played on a call, for example while the agent is processing. Also available in chat projects through the widget’s voice mode.

Also asked as:

play hold music while the bot is thinking
sound effects during the pauses on a call
callers hear dead air while the assistant works

Related: Voice output

Documented in: Behaviour › Background audio and audio cues

Voice input

The speech-to-text side of a voice agent: the provider (Deepgram, Gladia, Cartesia, AssemblyAI, Google), the transcription model and language, plus keywords, end-of-turn detection and keypad input.

Also asked as:

the assistant mishears what callers say
change the transcription language
pick a different speech recognition provider

Related: Keywords · End-of-turn detection · Keypad input

Documented in: Behaviour › Voice input · Behaviour › Provider, model, and language

Keywords

A comma-separated list of words to boost in transcription: product and company names, proper nouns, industry terms that standard speech recognition misses.

Also asked as:

transcription keeps getting our product name wrong
make speech recognition pick up specific words
boost recognition of unusual terms

Related: Voice input · Pronunciation dictionary

Documented in: Behaviour › Keywords

End-of-turn detection

Controls when the agent decides the caller has finished speaking: a confidence threshold (0.5–0.9, default 0.60) and the timing around it.

Also asked as:

the agent interrupts callers before they finish
the bot waits too long before answering on calls
tune when the assistant starts talking

Related: Voice input

Documented in: Behaviour › End-of-turn detection

Keypad input

Lets callers enter digits on the phone keypad (DTMF) during a call, with a timeout and other settings you configure. Voice projects only.

Also asked as:

the caller presses a number to choose a department
read the digits someone types on their phone
touch-tone menu for a phone agent
press 1 for sales and 2 for support
let callers choose with the keypad
DTMF tones during a call

Related: Voice input · Playbook step

Documented in: Behaviour › Keypad input

Session timeout

How long an inactive session lasts before it ends: minutes for chat projects (1–2,880, default 15, or off to keep sessions alive indefinitely) and seconds for voice projects (10–300, default 60).

Also asked as:

the chat ends after a few minutes of inactivity
calls hang up when the caller is quiet for a while
change how long a session stays open
how long before an idle conversation is closed
the call ends if nobody speaks for a minute

Related: Chat persistence · Conversation

Documented in: Behaviour › Session timeout (Chat projects) · Behaviour › Session timeout (Voice projects)

Chat persistence

Whether chat history is kept between sessions in chat projects — never forget, forget after a period, or forget on every reload — set in Behaviour and, per embed, with the widget’s persistence option.

Also asked as:

the widget forgets the conversation when the page is refreshed
keep the chat going after a reload
start every visit with a fresh conversation
the conversation resets whenever the page reloads
keep the same chat when the visitor comes back
a new session every time the page refreshes

Related: Memory · Session timeout · Chat widget API

Documented in: Behaviour › Chat persistence · Chat widget API › Configuring chat persistence

Events

Triggers for agent behaviour from outside the conversation — a webhook, a CRM update, a system alert, any external signal your application sends. Created in Behaviour and fired through the API or the widget.

Also asked as:

make the agent react to something that happened in another system
trigger the bot from a webhook
push an external signal into a live conversation

Related: Behaviour · Chat widget API

Documented in: Behaviour › Events

Global no-match

How the agent responds when it cannot match any skill, user intention or knowledge base result: a generative prompt or a fixed reply.

Also asked as:

the reply used when the input matches nothing
fallback message for gibberish
default answer for questions the bot cannot place
what the bot says when it doesn’t understand
the reply for input that makes no sense
handle questions the agent can’t place

Related: Global no-reply · Staying in scope

Documented in: Behaviour › Global no-match · Behaviour › Fallback

Global no-reply

How the agent re-engages when the user says nothing. Off by default.

Also asked as:

the caller goes silent, what should the bot do
prompt the visitor again when they don’t answer
nudge after a period of no input

Related: Global no-match · Skip turn

Documented in: Behaviour › Global no-reply

The editor

Reference

A pointer in instructions to a tool, system tool, playbook or workflow by name, inserted with /, so the prompt names the resource itself rather than a string that might drift.

Also asked as:

mention a tool inside my prompt so the agent uses it
link a playbook from the instructions
point the assistant at a specific workflow by name

Related: Missing reference · Instructions

Documented in: Referencing tools in prompts › Adding a reference · Referencing tools in prompts › What you can reference

Missing reference

The warning shown when a referenced resource is no longer available to the agent — usually because it was deleted, or is not present in this environment.

Also asked as:

a tool in my prompt shows a warning that it’s gone
instructions point at something that no longer exists
the reference chip turned red
the tool I referenced in my instructions shows as missing

Related: Reference

Documented in: Referencing tools in prompts › When a reference goes missing · Referencing tools in prompts › Editing and removing a reference

Steps block

An ordered block in instructions (type >> or pick Steps from the insert menu) that makes a playbook’s sequence explicit.

Also asked as:

make the assistant follow a numbered order in its instructions
an ordered procedure inside a prompt
numbered instructions the playbook must follow in sequence
enforce an order of operations in a prompt

Related: Instructions · Markdown shortcuts

Documented in: The agent instructions editor › Breaking instructions into steps

Markdown shortcuts

Type markdown at the start of a line (#, -, 1., >) or wrap text (**bold**) and the editor converts it as you type; pasted markdown keeps its headings, lists, quotes and links.

Also asked as:

format my instructions with headings and lists
paste a prompt written in markdown into the editor
bold and italic in the instructions

Related: Keyboard shortcuts · Steps block

Documented in: Markdown shortcuts for instructions › Formatting blocks · Markdown shortcuts for instructions › Pasting markdown

Keyboard shortcuts

The editor’s shortcuts for menus (/ for blocks and references, { for variables and secrets), formatting, blocks and undo.

Also asked as:

hotkeys in the instructions editor
shortcut to insert a variable
what does the slash key do in the editor
quick keys for formatting instructions

Related: Markdown shortcuts · Reference

Documented in: Keyboard shortcuts for the editor › Opening menus · Keyboard shortcuts for the editor › Editing