Hermitage is a local first Retrieval Augmented Generation system allowing you to quickly
query ranges of documents on the go. I build the first version of Hermitage in preparation
of and while writing my bachelors's thesis. The goal was to read my sources once or twice,
then pass them onto a system i can query the details and most important ideas again and again.
Features
The feature list grows and shrinks constantly right now as i am actively using and developing
it. The current and most stable features include:
Fully featured cli and web frontend
Section aware chunking
Multimodal ingest and queries
Information separation by domain
Information organisation in collections
Long term memory
Focus on scientific sources and standards
Support for local models (LLM, embedding, visual analysis and transcription) first
More to come as i figure it out.
Sreenshots
The following images are screenshots of Hermitage, the cli and web frontend: Starting view of the web frontend with the domain and collection
selection open.User query and response.Query in the cli frontend.
Technology
Hermitage is using Python and an assortment of libraries. For local LLMs i am using llama.cpp.
The frontend is built using SvelteKit.
