MindShard: A Self-Hosted Semantic Memory Layer for Your Browsing (github.com)

MindShard is a new open‑source, self‑hosted “semantic memory” for your web browsing: a Rust‑based local HTTP(S) proxy that captures pages you read, converts them into vector embeddings, indexes them, and exposes a semantic search UI so you can find content by meaning (e.g., recall a quote or idea and retrieve the original passage and link). It’s designed to act like a photographic memory for browsing history while keeping processing local — the only external network access is the one‑time download of an embedding model from Hugging Face. From a technical and practical standpoint MindShard is notable because it bundles embedding generation, a vector index, and a web/proxy stack into a standalone tool (Rust 1.88+, OpenSSL). Typical deployment is via Docker or cargo run; it requires generating and trusting a self‑signed CA (mindshard.cer / mindshard.key), runs a proxy on port 8080 and a web UI on 3000, and persists data in a LibSQL database. Important caveats: it intercepts all HTTP traffic (so can capture sensitive data), is experimental/educational only, and requires careful certificate installation and proxy configuration. For the AI/ML community it’s a useful reference implementation of on‑device semantic indexing and private retrieval, but operators must weigh the privacy/security risks of local traffic interception.