Raggify Perception

Introduction

This document is an input reference for developing raggify-perception: the part of Raggify system.

Related Modules

Raggify App

• Developed as a web application

Web Server (raggify-app client)

• The client folder is the root of raggify-app client
• Frontend of raggify-app
• Uses React, TypeScript, etc.
• Main feature is chat-style querying over documents ingested into the knowledge base
• Chat UI aims for a NotebookLM-like experience
• RAG pipeline settings UI aims for a Dify-like experience

App Server (raggify-app server)

• The server folder is the root of raggify-app server
• Backend of raggify-app
• Python program
• Manages user information and routes requests to raggify server
• Sits between raggify-app client (front) and raggify server (back)
• raggify server is treated as a driver abstraction
• It has an agent implementation and uses the backend LLM server to generate answers

RAG Server (raggify server)

• The RAG foundation library described above
• Published as a PyPI library
• Handles upstream RAG processing (local file/URL ingestion, embeddings, index generation, etc.)
• Can selectively use local models and third-party APIs
• Supports several usage modes; in this project it runs as a resident REST API server
• This is where scaling should happen

Raggify Client

• Client program used when raggify server is accessed as a REST API server
• Published as a PyPI library
• Supports several usage modes; in this project it provides an API for requests from raggify-app server to raggify server

LLM Server

• Pure local LLM server.

Raggify Perception Library

• The development target of this project
• Provides REST API Client for OCR, ASR, and image, audio, and video captioning, used by the raggify server when ingesting documents and by the app server when generating queries