Welcome to RAGged Edge Box, an open-source Retrieval Augmented Generation (RAG) system designed to help you find and manage your confidential documents with ease. The state-of-the-art semantic search and AI-powered question-answering techniques make it simple to locate and access the information you need, all within the privacy of your own laptop.
Learn more: https://textualization.com/ragged/
To build the image you'll need:
- Debian GNU/Linux
- git
- wget
- Docker
./get-dependencies.shas root (or using sudo):
./make-image.shThe final output is RAGged_Edge_Box.ova in the current folder.
Make sure you have plenty of disk available for Docker and the intermediate disk images.
Under a Debian GNU/Linux bookworm install, compile the dependencies following the first image in the Dockerfile. That includes installing the dependencies system-wide.
Copy the SQLite3 extension binaries (vss0.so and vector0.so) to /usr/local/lib/sqlite3 (you might need to create the folder), then set sqlite3.extension_dir to /usr/local/lib/sqlite3 under [sqlite3] in the php.ini for your web server (e.g., /etc/php/8.2/apache2/php.ini).
You might want to set other entries in your web server php.ini according to box/php.ini in this repo (enabling the pdo_sqlite and ffi extensions, increasing memory limits, execution times, file upload sizes, etc).
Install the composer dependecies following the relevant entries in the Dockerfile (composer install and subsequent lines).
Then set a symlink from /var/www/ragged to the site/ folder in this repo.
RAGged Edge Box will be available at http://localhost/ragged.
To have the LLM running, build llama.cpp following the instructions in the Dockerfile (you only need to compile the binary version for your local CPU) and launch it following box/launch-llama.sh (the GGUF model will be located someplace else, probably in download folder if you had run make-image.sh.