An engine is a conversation, designed as a diagram. You build it on a canvas out of a handful of parts, test it in the simulator, then publish it. This page covers every part of the builder. If you are brand new, skim Personas first: every engine runs as a persona.

Core concepts

A few words show up everywhere in the builder. Learn these and the rest falls into place.

Phase: a stage of the conversation. A phase is an opening message (how the AI kicks off that stage) plus a deck of cards it can play while the conversation sits in that phase. A typical engine has a few phases: a cold open, a warm-up, an invitation.
Deck: the set of cards attached to a phase. On each turn the engine plays one card from the deck, chosen by a weighted random draw, so the same phase feels different every time.
Card: a single move: chat, ask a question, send a photo, drop a link, close out. Each card has its own settings (see the card reference below).
Gate / transition: a rule that moves the conversation from one phase to another, or fires a reaction. Some gates sit inline between two phases; others float and watch the whole conversation.
Fact: a detail the engine learns about the lead (their name, city, interests). Facts are extracted from the chat as it happens and can be dropped back into messages with a tag.
Start points: the nodes that say where a conversation begins: the AI messaging first, the user messaging first, or an external event.
Floating observers: gates that are not wired between two specific phases. They watch the entire conversation and fire whenever their condition is met (for example, a fact appears, or the lead accepts a link).

The canvas

The builder is a zoomable canvas of nodes connected by edges. You drag pieces in from the side, drop cards onto phases, and draw lines to wire the flow.

Node palette

Start: the conversation-start nodes (AI Opener, Inbound Chat, Event Trigger).
Phases: add a new phase node to the canvas.
Rules: the logic gates and observers you wire between or around phases.

Card library

The card library lists every card type. Drag a card onto a phase to add it to that phase's deck. The same card type can appear in multiple phases with different settings.

Wiring edges

Drag from a node's output handle to another node to create an edge. Edges are how a start point reaches its first phase and how gates move the chat forward.

Auto-arrange

Hand-wired graphs get messy. The Auto-arrange button (the wand, top-right of the flow canvas) re-lays the whole graph in a tidy left-to-right layout with no overlapping nodes, then fits it to view. The new positions are saved when you save the engine. Run it before you finish any engine.

Card types

Cards are the moves an engine can make inside a phase. These are the built-in types, by their library names:

Casual Chat (CHAT): free-form message. This is the one card that fully honours your custom Instructions, so it is your workhorse for steering the wording.
Fact Finder (PROBE): asks for the next missing fact, automatically targeting whatever detail the engine still needs.
Send Photo (PHOTO): sends the best-fit photo from the persona's vault, auto-picked from each photo's description.
Proof Photo (VERIFICATION): asks the lead to send a proof / verification photo of themselves.
Drop the Link (LINK_DROP): drops your funnel link. Requires a funnel URL set in the global settings.
End / Close Out (CLOSE): wraps up and ends the conversation.
React (REACT), Tease / Flirt (TEASE), Open Up (SHARE), Relate (RELATE): quick built-in beats that keep a conversation feeling human. See the gotcha below: these four run a fixed line.

ध्यान दिनुहोस्

React, Tease / Flirt, Open Up and Relate run a fixed, built-in line and ignore the Instructions field. Anything you type into those four cards (and any tags) is not used. If you need custom wording, use a Casual Chat card instead.

Gates & transitions

Gates decide when the conversation moves on or reacts. There are two shapes: inline counters wired between two phases, and floating observers that watch the whole chat.

Inline counter

After N Messages (MESSAGE_COUNT): wired phase → gate → phase. It counts turns and advances the conversation once a threshold is hit. The threshold is 1–20. This is the standard way to step a chat forward through its phases.

Floating observers

Fact Check (FACT_EXISTS): fires when a specific fact has been extracted from the conversation.
Conversion Checker (CONVERSION_CHECK): fires based on whether the lead accepted or rejected the last link you dropped.
Signal Detector (DETECTOR): an internal observer that backs bot-accusation defense. It is not offered in the palette; you do not place it yourself.

Start points

Every engine needs to know how a chat begins. Add one or more start nodes and wire each to the phase where that path starts.

AI Opener / Outreach: the AI sends the first message (cold outreach). Use this for proactive funnels.
Inbound Chat / Default: the user messages first (the normal inbound DM). An engine must have exactly one Inbound Chat / Default start.
Event Trigger: starts a chat when a specific external event fires. It needs an event id (for example, story_reaction).
Follow-up: re-engages a lead who has gone silent, pulling them back into the conversation.

ध्यान दिनुहोस्

Prompting tags are not filled in on the AI Opener. A {name} typed into the opener would show literally to the lead. Keep the AI Opener written out in plain words.

Facts & the Knowledge tab

The Knowledge tab is where you manage the facts the engine collects. There are two kinds:

System facts: built-in details the engine always tries to learn (such as the lead's name and location).
Custom facts: your own fields, specific to your funnel.

Each fact is assigned to a phase: the stage during which the engine should try to learn it. The Fact Finder card then chases whatever fact is still missing for the current phase.

ध्यान दिनुहोस्

Each fact can belong to exactly one phase. If you want the engine to learn something at a different stage, move the fact. You cannot assign it to two phases at once.

The node inspector

Click any node to open the inspector on the right. What you can edit depends on the node, but the common fields are:

Instructions / directive: plain-language guidance for what the AI should do. On a card this steers the message; on a phase this is the phase's purpose. (Remember the four quick cards ignore it.)
Phase opener: the opening message for a phase: how the AI starts the conversation when it enters that stage.
Use limit (0–20): the maximum number of times a card may play in a conversation. 0 means unlimited.
Wait between uses: how many turns to wait before a card can be played again, so it does not repeat back-to-back.

Prompting tags

Inside a card's Instructions you can type a tag and the engine swaps in the real value before the message sends. The available tags:

Persona details

{name}: the persona's name
{age}: the persona's age
{origin}: where the persona is from
{city}: the persona's city
{job}: the persona's job
{hobbies}: the persona's hobbies and interests
{bio}: the persona's bio
{style_tags}: the persona's texting-style traits
{time}: the current time of day, so the persona can sound morning- or evening-aware

Your funnel

{platform}: the app you are messaging on
{funnel_platform}: your funnel or destination platform

Actions

{photo}: sends the best-fit photo from the vault, auto-picked by description
{link}: drops your conversion link, with filter-dodging applied

Card-specific

{fact}: on a Fact Finder card: the next detail to find out about the lead
{vibe}: on a Send Photo card: the current photo mood, casual or spicy
{context}: on an entry / opener card: the inbound first message the opener reacts to

Global & Identity settings

The engine-wide settings live in the Identity / global settings tab:

Messaging platform: the app the conversation runs on (drives the {platform} tag).
Turn limit (1–100, default 60): the maximum number of turns before the conversation stops.
Max message bubbles (1–5): how many separate message bubbles a single reply may be split into.
Funnel platform: your destination platform (drives {funnel_platform}).
Funnel URL: the link the Drop the Link card sends.
Link masking: applied automatically to dodge link-filters when the link is dropped.
Lead value: what a converted lead is worth, used by analytics to estimate revenue.

Rejections & bot-defense

You can script how the engine handles awkward moments. A lead pushing back, or accusing the AI of being a bot. Each scenario pairs a trigger with a response, which can be a piece of text or a picture, so the conversation recovers gracefully instead of stalling.

Validation & publishing

The builder checks your engine as you work and reports issues in the Problems panel. Two levels matter:

Errors block publishing. You cannot publish an engine that fails validation.
Only hard-limit errors block Save. You can save a work-in-progress that still has softer problems and come back to it.
The Problems panel jumps to the node when you click an issue, so you can fix it in place.

ध्यान दिनुहोस्

A phase must have at least one card and an opening message. An empty phase will not validate.

ध्यान दिनुहोस्

Gates and start nodes cannot be renamed. Only phases and event triggers take a custom name: let the builder auto-name everything else.

ध्यान दिनुहोस्

Publishing to the marketplace needs a fully clean validation: it is stricter than Save. Clear every error in the Problems panel before you publish.

Engine CRUD & the marketplace

New: start a fresh engine from a blank canvas.
Copy (clone): duplicate an existing engine to iterate without touching the original.
Rename: rename one of your own engines.
Delete: remove an engine you own.
Publish / unpublish: share a clean engine to the marketplace, or pull it back.

Next steps

With an engine built, you are ready to put it to work. Read up on the persona it runs as in Personas, start from a proven funnel in Marketplace templates, dry-run it in Testing & the simulator, and connect it live with Going live.