semtex is semtantic search for your desktop. Fully offline and private.
WARNING: semtex is alpha-quality software (on a good day). Any data stored may get lost or be corrupted.
-
Download and run the desktop application (currently Linux only).
-
Install the browser extension. Currently, this is only supported on browsers which allow you to install unsigned extensions (e.g. Firefox Nightly).
- a rust-based backend to index text into a local vector index (usearch) and database (sqlite)
- a browser extension which tracks local browsing activity;
- a desktop application to search and manage your data.
semtex's browser extension inspects the content of each page you visit and determines whether it appears to be article-like text. It uses Mozilla's Readability library to transform websites into plain text and sends these (via localhost) to the desktop application.
The desktop application receives these text snippets and:
- stores them in a local
sqlite
database; - transforms them into a vector embedding representation (using MiniLM-L12-v2);
- writes the embeddings to a local vector database (
usearch
).
The embedding model used has been chosen to work well without GPU acceleration.
To search your browing history, open up the desktop app and type a free-form semantic search query. This will similarly be converted to a text embedding and then used to search the local vector index and sqlite database.
semtex is an open source project built by Scalar. We are experts in building LLM-powered data processing and search systems. Get in touch hello@scalar.dev.