● Personalized Discovery – The agent analyzes browsing behavior, category preferences, price sensitivity, and purchase history to surface relevant recommendations and not merely search results. With L3 personalization enabled, the system continuously refines its understanding of each shopper’s preferences during the session, improving recommendations in real time.
● Agentic Checkout – Complete a purchase through chat including product selection, cart management, shipping, coupons, and payment for example:
"I need a look for a work interview."
From this single request, the agent can:
● Recommend relevant products
● Add items to the cart
● Collect shipping information
● Apply coupons
● Select shipping options
● Process payment
All without requiring the shopper to leave the chat interface.
● Voice Commerce –
The platform enables hands-free shopping using the browser’s native speech-to-text and text-to-speech APIs.
The agent:
● Listens to spoken queries
● Responds verbally
● Automatically re-activates the microphone after each reply
This creates a continuous voice conversation loop, functioning on any modern browser without additional hardware or subscriptions.
● Flexible Payments
Shoppers can complete purchases directly within the chat using multiple payment methods:
● Credit/Debit Card via Stripe (PCI DSS compliant)
● Google Pay / Apple Pay
● Klarna (Buy Now, Pay Later)
● PayPal
Klarna, Tamara, PayPal open in controlled popup windows that return the shopper to the conversation once the transaction is complete.
● Guest & Registered Flows –
The platform supports both guest and authenticated user flows. Guest shoppers can complete purchases without creating an account. Registered users can authenticate using:
● OTP verification
● Email and password
● Google / Facebook Sign-In
The agent automatically detects the user’s state and adjusts the interaction accordingly.
● Revenue Optimization Through Selling Style
The agent’s sales behavior can be configured per brand.
Example if ran in “aggressive” selling mode, encouraging upsells and cross-sells such as:
"Complete the look."
Other supported selling styles include:
● Luxury
● Consultative
● Educational
● Friendly
● Proactive
This allows the same AI agent framework to align with different brand identities and customer engagement strategies.
● Dynamic Personality Engine
The platform includes a calendar-aware personality engine that adjusts the agent’s tone and selling intensity based on:
● Seasonal campaigns
● Day of the week
● Time of day
This enables automated holiday promotions, weekend casual tone shifts, or campaign-specific messaging without manual updates.
● Recommendation Bias Monitoring
The system includes a bias detection mechanism that monitors recommendation patterns.
If the agent begins to overly promote a specific product or category beyond configurable thresholds, the system flags the behavior. This protects catalog diversity and brand integrity.
● Built-in Analytics and Admin Controls
The platform includes an integrated analytics dashboard and administration portal.
Key insights include:
● Conversion funnel analysis
● Shopper session behavior
● Average order value / Cost per conversation
● Bias detection reports
The admin portal allows real-time updates to:
● Selling style
● Personality rules
● Campaign schedules
● Voice settings
● Bias thresholds
All configuration changes can be made without redeploying code.
● Rapid Brand Onboarding - A retailer’s entire brand configuration — including identity, personality, voice settings, AI model, and payment integrations within 4 weeks of Go to Market.
● commercetools Integration - Adapter pattern supports commercetools, Salesforce Commerce Cloud, Shopware, and Shopify through a common RetailerStore interface.
● Zero‐Dependency Voice – Uses native browserspeech APIs without externalservices.
● Progressive Feature Activation – Voice, personalization, analytics, and bias detection can be enabled incrementally from admin.
● Built‐in Authentication – OTP login, email/password login,social media login.
● Multi‐Protocol Interoperability – Supports A2A/UCP endpoints, REST APIs, and OpenAI plugin manifests.
● Built-in Observability - Tracks conversation metrics, order events, conversion funnels, cost per conversation, and recommendation
Security is built into every layer of the platform, not bolted on after the fact.
· PII Isolation from LLM -
Customer personal data (credit card numbers, emails, passwords, addresses) is stripped before any data is sent to the AI model. The LLM never sees raw PII.
· Prompt Injection Protection -
Input sanitization detects and blocks attempts to override the agent's system prompt, extract internal configuration, or hijack the agent for non-shopping purposes.
· Catalog-Only Responses -
The agent only references products that exist in the retailer's live catalog via API. No product knowledge from LLM training data is surfaced. Every recommendation is validated against the catalog before being shown to the shopper.
· Price & Promotion Accuracy -
All pricing comes from the commercetools API in real time. The agent never fabricates or approximates prices, discounts, or promotions.
· Content Moderation -
Harmful, toxic, and off-topic inputs are detected and deflected before reaching the LLM. The agent stays strictly within the retailer's shopping context.
· Data Boundary Enforcement -
The agent blocks competitor references, data extraction attempts, social engineering, and out-of-scope questions. In multi-tenant deployments, retailer data is fully isolated.
· CORS & API Authentication -
Backend APIs are authenticated and rate-limited. CORS is restricted to the retailer's configured domains only.
· PCI-DSS Compliance -
Payment processing is PCI-DSS compliant via Stripe and other certified payment providers. Card data is never stored or processed by the agent.
· GDPR & CCPA Ready -
AI disclosure in the chat widget, privacy policy links, configurable data retention with automatic purge, and right-to-erasure tooling.
· No Training on Customer Data -
Conversation data is never used for AI model training. API configurations explicitly opt out of provider training programs.
