vec.blue
A decentralized search engine on the AT Protocol.
No central crawlers. No SEO spam. Just a Web of Trust.
Social Graph Scoping
Search is strictly limited to you and the people you follow. A high-signal, human-verified knowledge base.
Decentralized
Pages you save are chunked, vectorized, and stored as binary blobs on your Personal Data Server. You own the index.
Hybrid Vector Search
Fuses semantic context (Dense) with exact keyword matching (Sparse).
Disclaimer: Highly Experimental
vec.blue is strictly a personal experiment. The engine might crash, the service could halt permanently, and your synced indexes may vanish without any prior notice.
How It Works
Extract & Compute (Local)
You click the Ingest button in the extension. The browser extension instantly chunks the text and calculates Dense/Sparse vectors directly on your machine. No cloud APIs. No server costs.
Publish to PDS
The generated vectors are bundled into a lightweight JSON Blob and synced to your AT Protocol Personal Data Server.
Ingest to Vector DB
A vec.blue engine fetches your Blobs and flattens them. Every single chunk becomes an point inside a Qdrant database.
Search
When you type a query, the embedding model runs directly on your machine to generate Dense and Sparse vectors in real-time. It then queries the database.
Roadmap & Ideas
Ingestion Browser Extension (WIP)
The core crawler and embedding engine live in a browser extension. It is currently in active development—which means nobody can actually use this yet. Stay tuned.
Offline-First & Local Execution
Sync PDS vector blobs directly to your own machine. Execute hybrid searches, entirely offline.
Bring Your Own Index
A paradigm shift: Instead of begging Google to crawl your site, simply publish your own pre-calculated vector index for anyone to seamlessly subscribe to.
Local Auto-Crawler
A background daemon to continuously monitor your saved items, automatically re-calculate vectors on content updates, and silently purge dead links (404s).
Vector Portability
Embeddings are computationally expensive. By making vector blobs freely cloneable, communities can fork and share massive search indexes without wasting CPU cycles.
Semantic Query Cache
Bypass the heavy compute for query expansion (HyDE). Instantly map short keywords to pre-calculated, high-dimensional context vectors using your local dictionary or community-curated remote expansion packs.