Show HN: Embedding Explorer – compare text embedding models in your browser (github.com)

Embedding Explorer is a minimal, in‑browser web app for ingesting data, generating embeddings from multiple providers, and running fast similarity searches so you can evaluate embedding models side‑by‑side without spinning up a backend. Its four‑step workflow lets you connect datasets (CSV, SQLite, or samples), design input templates and preprocessing, configure multiple providers/models with API keys and parameters, then run batch embedding jobs and compare results through interactive k‑NN/cosine similarity views and per‑model result lists. It supports re‑embedding, progress tracking, and directed comparisons to help pick the right model for your retrieval or semantic search use case. Under the hood the app is written in Dart using the Jaspr framework, with background workers and data tasks also in Dart. It uses the experimental libSQL WASM package backed by the browser’s OPFS for persistent document/metadata/vector storage, runs vector queries (k‑NN/cosine) locally, and manages the JS toolchain (Vite, Tailwind) via a pnpm workspace — the Vite bundles are consumed by Dart through FFI. The approach speeds iteration, improves reproducibility and privacy (data stays in the browser), and makes model selection easier; practical limits include browser storage/quota and runtime performance for very large corpora, plus required network access to external embedding APIs.