perceive
Semantic search for your life
Perceive is a side project which I started to become more familiar with ML embeddings and semantic search. It has the ability to import browser history and filesystem sources, and index them for searching.
Features
- Read browser history and fetch the content
- Import bookmarks from Chrome
- Scan the local filesystem and index text files
- Supports multiple sources at once
- All indexing happens locally -- no need to send your data to someone else's server
Adding Sources
The browser history and bookmark source types use the Chromium browser directory as the location. For Brave Browser on
macOS this is ~/Library/Application\ Support/BraveSoftware/Brave-Browser/Default. Other Chromium-based browsers will
have similar locations.
Future Features
- Bookmark management
- More UI work
- Better article content extraction
- Browser extension for better page capture
Installation
As of now there is no prepackaged binary with everything included.
First, you'll want to run just install-models, using just. This will download the
model from Huggingface and convert it to the format used by Rust's version of Torch.
Building on macOS
You'll need to have libtorch installed. You can use pip3 or homebrew to install pytorch, and then
set the LIBTORCH environment variable to the path where libtorch is installed.
For example, when installed via homebrew you would add export LIBTORCH=/opt/homebrew/opt/pytorch to your ~/.zshrc or
~/.zprofile.
At some point I'll try to add some better way of installing a local copy too, to make it easier to match the version of libtorch with the version expected by the bindings.