Skip to content
Reponse.ai LogoDocs
DocsAPISDKsGuides

Search Documentation

Search for an article...

Getting Started

  • Overview
  • Architecture
  • Installation
  • Authentication
  • First Request

Products & Collections

  • Products — List
  • Products — Get
  • Products — Metafields
  • Collections — List
  • Collections — Get
  • Collections — Products

Cart & Checkout

  • Cart — Create
  • Cart — Get
  • Cart — Add Item
  • Cart — Update Item
  • Cart — Remove Item
  • Cart — Apply Promotion
  • Checkout — Stripe
  • Checkout — Payment Intent
  • Checkout — ACP

Orders & Fulfillment

  • Orders — Create
  • Orders — Get
  • Orders — Confirm
  • Orders — Cancel
  • Orders — Fulfill
  • Orders — Refund

Commerce

  • Inventory — Get
  • Inventory — Update
  • Shipping — Rates
  • Discounts — List
  • Discounts — Create
  • Discounts — Validate
  • Subscriptions — Manage

Loyalty & Gift Cards

  • Loyalty — Balance
  • Loyalty — Redeem
  • Loyalty — Referral
  • Gift Cards — List
  • Gift Cards — Redeem

Tickets & Support

  • Tickets — List
  • Tickets — Create
  • Tickets — Reply

Platform

  • Product Feed — JSON
  • Product Feed — CSV
  • Theme — Get
  • Approvals — Execute
  • Approvals — Reject
  • Geocode

SDKs

  • SDK Overview
  • TypeScript SDK
  • React Hooks

Guides

  • Chat Widget
  • Shopify Sync
  • Storefront Starter
  • Loyalty Program
  • Discounts & Promotions
  • Subscriptions
  • Klaviyo Integration
  • Custom Domains
  • Agentic Commerce (ACP)
  • A2A Protocol
  • AI Engines
  • MCP Server

Webhooks

  • Webhooks Overview
  • Events Reference
  • Shopify Webhooks
  • Stripe Webhooks
  • Reviews (Stamped / Trustpilot)
  • Logistics
  • Email Inbound

Resources

  • Environment Variables
  • Rate Limits
  • Changelog
DocsGuidesAI Engines

AI Engines

The AI engines that power Reponse activation.

2 min read/Last updated Jul 9, 2026
On this page

Overview

Reponse AI engines define the personality, sales funnel strategy, and response style of your chat assistant. Each engine is optimized for a specific traffic type and buyer mindset. The dispatcher selects the right engine automatically based on UTM parameters, or you can force one explicitly.

Prerequisites

  • A Reponse workspace with products synced
  • At least one published campaign

Engine Profiles

Reponse ships with three engine families, available in both French (`fr-`) and English (`en-`) locales:

EngineAudienceToneBest For
core-std-v1Organic / direct visitorsWarm, consultativeDefault engagement, brand storytelling
ecom-expert-v1Search / high-intent usersData-driven, ROI-focusedTechnical buyers, comparison shoppers
ecom-turbo-v1Social / paid trafficEnergetic, urgency-drivenImpulse buys, flash sales, social ads

Each engine folder contains two files:

  • sales-funnel.ts — Defines the conversion funnel stages and transition rules.
  • style.ts — Sets the tone, vocabulary, and formatting constraints.

Shared behaviors (identity, safety, RAG protocol) live in `engines/shared/`.

Naming Convention

Engines follow the pattern: `[LANG]-[FAMILY]-[PROFILE]-[VERSION]`

fr-ecom-turbo-v1
│   │     │     │
│   │     │     └─ Version
│   │     └─────── Profile (std, expert, turbo)
│   └───────────── Family (core, ecom)
└───────────────── Language (fr, en)

UTM-Based Dispatch

The dispatcher (`engines/dispatcher.ts`) automatically routes visitors to the best engine based on their UTM parameters:

UTM SignalEngine Selected
utm_medium=searchecom-expert-v1
utm_campaign contains roi, expert, techecom-expert-v1
utm_content contains technique, dataecom-expert-v1
utm_source is facebook, instagram, tiktokecom-turbo-v1
utm_medium=social or paidsocialecom-turbo-v1
utm_campaign contains promo, flash, turboecom-turbo-v1
No UTM or unmatchedcore-std-v1 (default)

The intent classifier can further override the engine mid-conversation if the user's messages signal a different buying intent.

What Engines Are Grounded On

All engines share the same knowledge base per workspace:

  • Product catalog — titles, descriptions, prices, variants
  • Product facts — curated selling points and specifications
  • Vision analyses — AI-generated insights from product images
  • RAG documents — vector-matched knowledge sources
  • Workspace policies — commercial policy, marketing policy, brand voice

Forcing an Engine

You can bypass the dispatcher by passing an `engine` field in the chat API request body:

bash
curl -X POST "https://api.reponse.ai/v1/chat" \
  -H "Authorization: Bearer rp_live_abc123..." \
  -H "Content-Type: application/json" \
  -d '{
    "workspaceId": "ws_01H...",
    "message": "Compare these two products",
    "engine": "en-ecom-expert-v1"
  }'

Troubleshooting

IssueSolution
Wrong engine selectedCheck the UTM params in the page URL. The dispatcher logs the selected engine.
Engine responses feel genericEnsure product facts and knowledge sources are populated for your workspace.
English visitors get French engineVerify the locale is set to en in your campaign or widget configuration.
Intent override unexpectedThe intent classifier may switch engines mid-conversation. Review the conversation logs.
PreviousA2A Protocol
NextMCP Server