Gist | Devpost

example of a completion based on some of the resources you selected
link selection mode to save links, including the context relating to the content saved
resouce panel

Inspiration

As one that enjoys esoteric but contemporary memetic ramblings, I found myself vastly disappointed at the quality of twitter's search in regard to calling posts you've bookmarked. I thus came up with the idea of a qualitative search algorithm that allows you to describe things you want to retrieve

What it does

Gist embeds and saves entries, allows you to retrieve anything you've saved by describing it. It uses rag and semantic search under the hood, using local embedding models so your data never leaves your machine.

How we built it

We use transformers.js for embedding and qdrant local vector database to retrieve data via semantic search. It also ships with an AI panel that lets you search

Challenges we ran into

Getting a lot of the UI features to work was very finnicky. Most websites don't let you run content scripts to place labels on DOM elements, so getting around this needed lots of googling.

Accomplishments that we're proud of

Having the mouse be optional to the extensions function is a nice touch for keyboard absolutists,

What we learned

Hot Module Reloading is great until it isnt'. Cached browser states often screwed up what was otherwise a smooth development session

What's next for Gist

more browser integrations, AI browser navigation through websockets, so you have a search agent that looks through pages to find even more context relating to your saved links