SEC / 001Reference implementationChapter 02

How XEREBROimplements ACO.

XEREBRO is an operating system for commerce: a neutral core and eight specialised applications that observe the shop, diagnose it, act, and prove the effect. It runs on Magento, Adobe Commerce, Shopware and Shopify with identical signal depth. Operated by DevQon GmbH, Zurich. This page shows how one implementation delivers on the characteristics of the category.

SEC / 002

Eight applications as ACO components.

The category describes four functions: observe, diagnose, act, prove. Each XEREBRO application covers one of them, all run on the same core and can be booked individually.

A / 01Observe

Qompare

Price radar

Matches every article by EAN against up to 14 competitors and reports price and availability changes as they happen.

A / 02Act

Qontext

Sales agent

Delivers cross-sell, up-sell and win-back from eight data sources, from order history to the shop source code.

A / 03Diagnose

Qapture

Session replay

Reconstructs sessions server-side on Swiss infrastructure and marks the moment a customer abandons the purchase.

A / 04Prove

Qontour

Uplift simulator

Calculates from real daily revenue and measured abandonment rate what a fix is worth in CHF per year.

A / 05Observe

Qontrol

Order-flow monitor

Checks six core signals of the order path every five minutes and alerts with the CHF value of the affected orders.

A / 06Act

Qookie

Consent layer

Manages consent server-side under revFADP, GDPR, ePrivacy and IAB TCF 2.2, without US JavaScript and without third-party cookies.

A / 07Prove

Quasar

Executive signal

Condenses the signals of all applications every Monday at 08:00 into a CHF ranking with three actions.

A / 08Act

Qonsul

AI product advisor

Advises in the shop chat exclusively on the basis of the shop’s own catalogue, prices and stock.

SEC / 003

Six core signals every five minutes.

Qontrol is the observation component. It reads the operating state of the shop at the system itself, not just at the HTTP status. The examples come from Magento and Adobe Commerce; the same signals are read on Shopware through the message queue and on Shopify through the Admin API.

YOUR ORDER FLOW HAS A PULSE. Qontrol listens around the clock: order flow, cron, indexer, payment, stock, search. ANOMALY ORDER-FLOW CRON INDEXER PAYMENT STOCK SEARCH LIVE
Your order flow has a pulse. Qontrol listens around the clock on six signals.
C / 01

Order Flow

Synthetic test order through the real order path, every five minutes.

  • Cart and add-to-cart
  • Address entry and shipping choice
  • Payment authorisation: Datatrans, SIX Saferpay, Payone, Klarna, Adyen, Stripe
  • Order confirmation with order ID
  • End-to-end time per step
C / 02

Cron Health

Scheduled jobs and message queues checked for stalls.

  • Example Magento: cron_group default, index, consumers
  • Example Shopware: message-queue backlog per consumer
  • Last-success timestamp per job
  • Runtime outliers against baseline
  • Blocked queue before the first customer call
C / 03

Indexer Status

Checks whether an index is stale and where the cause lies.

  • Example Magento: catalog_product_price, cataloginventory_stock
  • Example Magento: catalogsearch_fulltext, catalog_category_product
  • Example Shopify: catalogue sync lag via the Admin API
  • Trace back to the last catalogue diff
  • Frontend price state against system price state
C / 04

Mode and Cache

Operating mode and cache backend configured correctly.

  • Example Magento: MAGE_MODE production versus developer
  • Redis cache backend reachable
  • Varnish purge status
  • Full-page-cache hit ratio after every deploy
  • Session backend
C / 05

Search Health

Search index, aliasing and reindex windows in view.

  • Example Elasticsearch and OpenSearch: cluster status green, yellow, red
  • Index alias consistency
  • Reindex windows outside peak hours
  • Query latency p95
  • Empty result lists for top search terms
C / 06

Uptime and Performance

Baseline beneath all other signals. Not the answer, but the frame.

  • HTTP status home page and product page
  • TTFB catalogue and checkout
  • Response time p50 and p95
  • SSL certificate expiry
  • DNS resolution CH and EU
SEC / 004

Four examples a horizontal tool does not see.

An uptime monitor reports green. The customer abandons. The difference lies in the depth of view and in what happens next: the agent acts within stored policies. Without a policy it waits for a click. Every action is written to the audit log.

THE AGENT SEES THE OUTAGE BEFORE IT COSTS CHF. CRON INDEXER PAYMENT payment_provider: timeout Policy: fallback provider → active Audit log · recorded #1#2!#4#5#1#2!#4#5 Check green · orders flow Anomaly · action within the policy, otherwise proposal with CHF estimate
The agent sees the outage before it costs money: anomaly detected, action within the policy, logged.
Example 01

Payment timeout in checkout

The payment handler responds after 4'862 milliseconds with ETIMEDOUT. The uptime monitor sees HTTP 200 on the home page. The customer abandons. XEREBRO reads the payment log, recognises the timeout pattern and activates the fallback provider by policy. Without a policy the proposal waits for approval with the log excerpt attached.

Example 02

Price index stale

After a large catalogue import the price index sits on invalid for hours. The category page shows old prices, the discount campaign does not apply. XEREBRO identifies the cause in the catalogue update, triggers the targeted reindex within the policy and quantifies the loss per hour.

Example 03

Message queue stalled

Orders come in, but order confirmations do not go out. Customers call the next day. The fault lies in a stalled queue, not in the web application. XEREBRO distinguishes between job groups, reports the blockage precisely and restarts the consumer if the policy allows it.

Example 04

Search index rebuild during peak sale

On Black Friday afternoon a cron job starts the search index rebuild. Search returns empty results for seven minutes. XEREBRO sees the alias switching, moves the rebuild out of the peak window by policy and logs the action for the review.

SEC / 005

Execution model and evidence.

Three properties by which an ACO implementation can be measured: how it acts, what it draws its decisions from, and how it proves the effect.

WITHOUT A POLICY: A PROPOSAL. WITH A POLICY: AN ACTION. AKTION · #0412 · POLICY: REINDEX.ALLOWED Indexer «catalog_product_price» invalid [ERROR] main.CRITICAL: Reindex window exceeded · elasticsearch timeout 30s Causal chain: cron backlog → index invalid → stale prices Estimated impact: CHF 3'800 / h EXECUTED 07:14 · AUDIT-LOG #0412 · ABORT ANYTIME LOGGED An audit log per action. Abort anytime.
Without a policy you get a proposal, with a policy an action. Every action with an audit log, abort at any time.
E / 01

Policy-governed autonomy

The operator sets the framework and policies. The agent acts autonomously within them, every action is written to the audit log. Without a policy it waits for a click. Code access stays read-only, XEREBRO writes nothing into the system. This delivers the human oversight required by EU AI Act Article 14.

E / 02

Eight data sources fused

Shop, ERP, order history, market prices, server metrics, payment events, social signals from Meta, TikTok and X, and the shop source code feed into one decision. No widget built from two data points.

E / 03

CHF evidence within three months

Each application delivers a first proven CHF amount within three months, traceable in the Value Ledger with before-and-after value, date and source. If the evidence does not materialise, the operator cancels without commitment.

SEC / 006

Availability and operation.

Generally available since Q3 2026, bookable monthly after the 3-month evidence. The target group is shops from CHF 500'000 online revenue per year; below that the economic lever is missing.

F / 01

Booked individually

Eight applications, each booked individually and cancellable monthly once the first CHF evidence after three months is delivered or fails to appear.

F / 02

Token quota included

Every configuration includes a token quota for the reasoning. No overage without the operator's approval.

F / 03

Scales with the shop

Configured by online revenue, shop system and number of applications. Large shops get individual token quotas, multi-store and SLA.

F / 04

Two zones, 18 certificates

Critical data at Infomaniak in Geneva and Zurich under Swiss law. Reasoning at an EU-certified cloud provider in the Frankfurt region. 18 verifiable certificates across both zones.