EdgeReco: search and recommendations on the shopper's device EdgeReco is an open-source product search and recommendation engine that runs in the shopper's browser. A store publishes one signed catalog bundle; the browser verifies it, stores it locally, and serves search, ranking, and personalized recommendations without a per-query backend. How it works The catalog is content-addressed and signed with Ed25519. EdgeReco checks the signature and every SHA-256 chunk before using the data. It then runs BM25 keyword search and vector search on-device, combines their results with Reciprocal Rank Fusion, and reranks products using only signals from the current browser session. What the demo proves Nimbus is the live example storefront. It downloads the same static bundle any CDN can host, performs retrieval and ranking locally, and can reload offline after the required assets are cached. Product interactions stay in the browser unless an operator explicitly configures an optional analytics uplink. View source on GitHub · Architecture · EdgeProc substrate · Frequently asked questions Nimbus is the demo store. The interactive demo requires JavaScript.