Live shopper and catalog graphs for retail AI agents

From raw clickstream to an agent-readable retail graph

E-commerce clickstreams and order tables grow fast and fragment intent across sessions. JetGraph folds streaming catalog and shopper events into compact relationships:

• Shopper behavior: USER -[:VIEWED]-> PRODUCT, CLICKED, COMPARED, ADDED_TO_CART, and PURCHASED, updated in place with recency, counts, value histograms, and 21-tick activity bitmaps.
• Catalog & basket structure: PRODUCT -[:BELONGS_TO]-> CATEGORY, MADE_BY -> BRAND, IN_PRICE_SEGMENT -> PRICE_RANGE, GOOD_FOR -> FEATURE, BOUGHT_WITH, and COMPATIBLE_WITH.
• Voice of the customer: PRODUCT -[:HAS_REVIEW_THEME]-> REVIEW_CLUSTER and POPULAR_WITH -> BUYER_SEGMENT.

Questions retail agents answer over MCP

• High-intent cart abandonment: use get_edge_state or cypher_read to compare recent VIEWED / COMPARED / ADDED_TO_CART activity against missing PURCHASED pairs.
• Basket affinity & cross-sell: walk BOUGHT_WITH and COMPATIBLE_WITH neighbors or run find_similar across shared PURCHASED and CATEGORY edges.
• Category & price-band drift: compare a shopper's get_histogram across 24 hours vs 30 days to see when a buyer shifts categories or cools off.
• Merchandising triage: ask why conversion dropped in a category by combining HAS_REVIEW_THEME links, stock/price properties, and ADDED_TO_CART vs PURCHASED counts.

Compact footprint on high-volume clickstreams

Millions of repeat views and cart events update existing (USER, PRODUCT) compact records rather than appending unbounded rows — keeping in-memory footprint proportional to unique shopper-product interactions and letting idle browsing edges expire via state_ttl_secs.