ShopX unifies intent understanding and item-space operations for agentic shopping

The authors propose ShopX, a foundation model designed to bridge the gap between language understanding and item fulfillment in agentic shopping workflows. Unlike existing approaches that wrap LLMs around separate search pipelines, ShopX uses semantic IDs (SIDs) to allow models to directly operate within the item space.

• ShopX unifies intent understanding, execution planning, and flexible SID-native operations into a single model.
• The framework includes a serving harness with a model-facing action protocol for context access and state management.
• It enables complex tasks like SID beam-search retrieval, listwise ranking, and product bundling.
• Evaluation on Taobao production logs shows improved performance on complex or ambiguous requests compared to tool-mediated systems.

This model-centric design reduces lossy hand-offs between agent orchestration and item-space execution, improving overall framework behavior.