Anthropic Releases Open-Source Blueprint for Building Commerce Agents
Key point
Anthropic has released implementation code for shopping and merchant agents designed for retail and travel platforms.
Details
Anthropic has released an open-source blueprint to help build Claude-based commerce agents. This repository includes full implementations of shopping agents and merchant agents for use in retail, travel, telecommunications, and ticketing platforms.
Shopping Agent
Operates within user apps or websites, handling natural language requests.
- Multi-item search and cart construction: Understands complex requests like "tent, sleeping bag, and stove for weekend camping with two kids" and assembles appropriate products from the catalog.
- Personalization and customer service: Remembers customer preferences to provide recommendations and resolves CS queries such as order tracking and return policies directly within the conversation.
- Guardrails application: Restricts prices and products to actual catalog data and prevents manipulative upselling patterns.
Merchant Agent
Supports store operators by providing insights based on internal data.
- Sales analysis and inventory management: Analyzes sales status and proactively detects risks such as inventory depletion before promotions.
- Pricing and marketing suggestions: Proposes discount rates, promotions, and marketing campaign drafts based on its own sales history.
- Human approval process: When the agent suggests changes, they are executed only after human final approval, maintaining control.
Deployment and Ecosystem
This blueprint can be deployed to existing Claude development environments such as the Claude API, Amazon Bedrock, Microsoft Foundry, and Google Cloud Vertex AI. Additionally, it collaborates with partners like Accenture, Mastercard, and Visa to support adoption within client and merchant communities. The code is available on GitHub, along with industry-specific demos and technical deep-dive materials.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.