Solutions / Technical owners
For technical owners

One memory, models that decide, and an API that returns the answer.

What goes in, what the models do, what comes out, and how long it takes to connect.

See it on your storeAPIs and the MCP server
  • Google Cloud, event-based ingestion
  • Live the same day on Shopify and plug-ins
  • ISO 27001, GDPR, EU or Asia hosting
Data in · decisions outtechnical
Events from web, app and backendin
Catalog: full sync plus changes within a minutein
One profile per person, anonymous toomemory
Feeds, search, chat, messages, audiencesout
Decisions
on the fly
per page view and per message
Proof
holdout group
read in your own orders
The whole system on one page

Data in, one memory, decisions out

Every surface reads the same profile and the same catalog. Every decision is scored against the goal you set.

IN What goes in Events from the web tag, plug-ins and four mobile SDKs Orders from the backend, online and offline The catalog: a full sync daily, changes within a minute Custom fields on products, events and profiles MEMORY AND MODELS One profile per person From the first anonymous click, joined by ID, email or phone Predictions: next basket, next purchase, 12-month value, churn 16 recommendation algorithms, a bandit per person, your rules One goal: conversion, basket or lifetime value, with a holdout OUT What comes out Product feeds per person, with live price and stock, on the fly Search ranking, chat answers, banners on the site and in the app Email, push, SMS, Viber, WhatsApp, with products per person Ad audiences and server-side conversions, webhooks, MCP Every order and click comes back in, and the models learn from it
IN What goes in Events from the web tag, plug-ins and four mobile SDKs Orders from the backend, online and offline The catalog: a full sync daily, changes within a minute Custom fields on products, events and profiles MEMORY AND MODELS One profile per person From the first anonymous click, joined by ID, email or phone Predictions: next basket, next purchase, 12-month value, churn 16 recommendation algorithms, a bandit per person, your rules One goal: conversion, basket or lifetime value, with a holdout OUT What comes out Product feeds per person, with live price and stock, on the fly Search ranking, chat answers, banners on the site and in the app Email, push, SMS, Viber, WhatsApp, with products per person Ad audiences and server-side conversions, webhooks, MCP Every order and click comes back in, and the models learn from it
What goes in

Three sources, and the integration is complete

Events, orders and the catalog. The catalog is an operational copy, never the system of record.

EVENTS Everything a visitor does Web tag, plug-ins, four mobile SDKs Server-side, first-party Anonymous visitors included Pages, products, searches, scroll, time on page, chat, carts ORDERS Where they happen Online through the plug-in or tag Offline, from a shop or a call center, through the events API CATALOG An operational copy A full sync daily, or as agreed Every price and stock change applied within a minute Custom fields, first-class INTO ONE MEMORY Releva, per brand Shopify: the app, two years of orders, live in 1 to 2 hours WooCommerce, Magento, PrestaShop, OpenCart, CloudCart, CS-Cart, Gomag, MerchantPro: a plug-in, live the same day A custom store: one tag or Google Tag Manager, 3 to 5 business days for all tools Apps: SDKs for iOS, Android, Flutter and React Native Segment connected in both directions, if you run it
EVENTS Everything a visitor does Web tag, plug-ins, four mobile SDKs Server-side, first-party Anonymous visitors included Pages, products, searches, scroll, time on page, chat, carts ORDERS Where they happen Online through the plug-in or tag Offline, from a shop or a call center, through the events API CATALOG An operational copy A full sync daily, or as agreed Every price and stock change applied within a minute Custom fields, first-class INTO ONE MEMORY Releva, per brand Shopify: the app, two years of orders, live in 1 to 2 hours WooCommerce, Magento, PrestaShop, OpenCart, CloudCart, CS-Cart, Gomag, MerchantPro: a plug-in, live the same day A custom store: one tag or Google Tag Manager, 3 to 5 business days for all tools Apps: SDKs for iOS, Android, Flutter and React Native Segment connected in both directions, if you run it
The memory

One profile per person, from the first click

Devices and channels join by profile ID, email or phone. The data stays yours, in the region you choose.

Reads an email on her phone Browses on a laptop Buys in the app Clicked a Meta ad first Asks the chat a question ONE PROFILE One record per person From the first anonymous click Joined by profile ID, email or phone, across devices With the Google and Meta clicks before each order Segments refresh every 10 to 30 minutes KEPT WHERE YOU NEED IT Your data, isolated Each brand separate Your property, first-party Google Cloud, event-based EU-only hosting on request In-region cloud in Asia ISO 27001, GDPR
Reads an email on her phone Browses on a laptop Buys in the app Clicked a Meta ad first Asks the chat a question ONE PROFILE One record per person From the first anonymous click Joined by profile ID, email or phone, across devices With the Google and Meta clicks before each order Segments refresh every 10 to 30 minutes KEPT WHERE YOU NEED IT Your data, isolated Each brand separate Your property, first-party Google Cloud, event-based EU-only hosting on request In-region cloud in Asia ISO 27001, GDPR
The models

Predictive models decide. Language models only talk.

Trained on your data, served on our own GPUs, scored against your goal.

INPUTS Profile and catalog The person: every event, order and click The product: live price, stock, custom fields PREDICTIVE MODELS Trained on your data Next basket, next purchase 12-month value, churn, RFM 16 recommendation algorithms Per store, plus a transformer trained across stores Served on our own GPUs Checked against orders monthly DECISION A decision per person Your goal: conversion, basket or 12-month value A bandit tests strategies and moves traffic to winners Your rules and caps apply Email before paid channels LANGUAGE MODELS Two jobs only The chat with the shopper, and the long-running processes a marketer starts by asking. Never a decision, never a recommendation.
INPUTS Profile and catalog The person: every event, order and click The product: live price, stock, custom fields PREDICTIVE MODELS Trained on your data Next basket, next purchase 12-month value, churn, RFM 16 recommendation algorithms Per store, plus a transformer trained across stores Served on our own GPUs Checked against orders monthly DECISION A decision per person Your goal: conversion, basket or 12-month value A bandit tests strategies and moves traffic to winners Your rules and caps apply Email before paid channels LANGUAGE MODELS Two jobs only The chat with the shopper, and the long-running processes a marketer starts by asking. Never a decision, never a recommendation.
What comes out

One feed call, every surface

The person and the context go in. Ranked products with live price and stock come back, on the fly.

ONE CALL Person plus context The page, the screen or the email Ranked products back, with live price and stock, on the fly The same feed everywhere On the site and in the app Search ranking, product blocks, banners, the chat Messages Email on your domain, push, SMS, Viber, WhatsApp, app inbox Catalog triggers: price drop, back in stock Ads Audiences by predicted value to Meta, Google, TikTok Server-side conversions, one event ID, counted once Your systems Webhooks on journey steps and low NPS scores Feeds API, events API, the MCP server
ONE CALL Person plus context The page, the screen or the email Ranked products back, with live price and stock, on the fly The same feed everywhere On the site and in the app Search ranking, product blocks, banners, the chat Messages Email on your domain, push, SMS, Viber, WhatsApp, app inbox Catalog triggers: price drop, back in stock Ads Audiences by predicted value to Meta, Google, TikTok Server-side conversions, one event ID, counted once Your systems Webhooks on journey steps and low NPS scores Feeds API, events API, the MCP server
How it proves it

A holdout, and orders counted once

The lift is measured in your own orders, not claimed. The mechanics are on the How it decides page.

VISITORS Split from day one A share you choose never sees the agent WITH THE AGENT The rest Search, pages, chat, messages, audiences HOLDOUT Your store as it was Same weeks, same promotions READ-OUT Revenue per visitor, both groups Orders counted once, with their clicks Downloadable from the agent Windows stated: opened in 1 day or clicked in 7; site clicks in 1 day Read in your Shopify or GA4
VISITORS Split from day one A share you choose never sees the agent WITH THE AGENT The rest Search, pages, chat, messages, audiences HOLDOUT Your store as it was Same weeks, same promotions READ-OUT Revenue per visitor, both groups Orders counted once, with their clicks Downloadable from the agent Windows stated: opened in 1 day or clicked in 7; site clicks in 1 day Read in your Shopify or GA4
Connecting your stack

What connects, and how long it takes

How it connectsTime to live
ShopifyApp install, two years of orders synced1 to 2 hours
WooCommerce, Magento, PrestaShop, OpenCart, CloudCart, CS-Cart, Gomag, MerchantProPlug-inSame day
Custom storeOne tag or Google Tag Manager, events API3 to 5 business days
Mobile appsSDKs: iOS, Android, Flutter, React NativeNext app release
Email toolRuns next to KlaviyoDay one
Customer data platformSegment, both directionsDays
Help deskHandoff to Gorgias, Zendesk, FreshdeskDay one
Your systems and agentsEvents API, webhooks, feeds, MCP serverSelf-serve
Where to look next

The details, page by page

APIs and the MCP server

Events in, feeds and webhooks out, and what an assistant can ask the agent.

For AI agents →

How it decides

The goal, the decisions it makes per person, and how the holdout reads.

Read the mechanics →

Every feature, grouped by job

The full list, including data and security.

All features →

Questions from technical owners

Where does Releva run?

On Google Cloud, with event-based ingestion. EU-only hosting on request, in-region cloud in Asia. Each brand's data is isolated. ISO 27001, GDPR.

Do we need a data warehouse or a CDP first?

No. Events, orders and the catalog go in; feeds and messages come out. If you run Segment, Releva connects to it in both directions.

How fast are the decisions?

Feeds are computed per request with the page or screen context. Catalog changes apply within a minute. Segments refresh every 10 to 30 minutes.

Where are language models used?

Only in the chat with the shopper and in the long-running processes a marketer starts by asking. Decisions and recommendations come from predictive models trained on the store's data.

What scale has it run at?

One client keeps 22 million profiles on Releva, recorded 131 million events in a month, had 22.6 million shoppers enter journeys in one day and delivered 8.7 million messages in a month.

Bring your engineer to the call.

One call with your store and your stack open. We walk through the data flow, the APIs and how the holdout reads in your orders.

See it on your store
GeorgiYavorBoryanaBoyanMariaIsaacNikoletaYou'll talk to one of us.