> ## Documentation Index
> Fetch the complete documentation index at: https://docs.raglight.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Introduction

> A lightweight, local-first framework for rapidly prototyping and experimenting with Retrieval-Augmented Generation systems.

# RAGLight

**RAGLight** is a lightweight, modular, and **local-first** Python framework designed to
**rapidly prototype, test, and experiment with Retrieval-Augmented Generation (RAG) systems**.

It is built for developers and researchers who want to iterate quickly on ideas,
run everything locally, and keep full visibility over how data is ingested,
retrieved, and injected into Large Language Models.

RAGLight favors **clarity over abstraction** and **composition over hidden magic**.

## Features

<CardGroup cols={2}>
  <Card title="Modular pipelines" icon="cubes">
    Swap LLMs, embeddings, and vector stores without touching the rest of your pipeline.
  </Card>

  <Card title="Hybrid search" icon="magnifying-glass">
    BM25 + semantic search fused with Reciprocal Rank Fusion — works on both Chroma and Qdrant.
  </Card>

  <Card title="Streaming" icon="bolt">
    Token-by-token output via `generate_streaming()` on all LLM providers.
  </Card>

  <Card title="Conversation history" icon="comments">
    Full multi-turn history across all providers, with optional `max_history` cap.
  </Card>

  <Card title="Agentic RAG" icon="robot">
    Tool-calling agent with MCP support — goes beyond simple retrieval.
  </Card>

  <Card title="REST API + Chat UI" icon="server">
    `raglight serve --ui` deploys your pipeline as an API and a Streamlit chat interface.
  </Card>

  <Card title="Langfuse observability" icon="chart-line">
    Trace every retrieve → rerank → generate call end-to-end in your Langfuse dashboard.
  </Card>

  <Card title="AWS Bedrock" icon="aws">
    Claude, Titan, Llama and more — plug into your existing AWS credentials.
  </Card>
</CardGroup>

## What RAGLight focuses on

RAGLight is intentionally minimal and explicit. It focuses on:

* Explicit and controllable data ingestion (files, folders, repositories)
* Pluggable embedding models and providers
* Interchangeable vector stores (Chroma, Qdrant)
* Simple, inspectable RAG pipelines
* Agentic and reasoning-oriented extensions (Agentic RAG)
* Local experimentation with LLMs (Ollama, LMStudio, vLLM)

Every step of the pipeline is configurable and observable, making it easy to debug,
benchmark, and iterate.

## Who is RAGLight for?

RAGLight is designed for developers who want:

* Full control over their RAG stack
* A framework optimized for **local prototyping and experimentation**
* Minimal abstractions with predictable behavior
* The ability to swap components without refactoring everything
* A clean path from prototype to more advanced setups

It is particularly well suited for:

* Research and proof-of-concept work
* Testing new RAG or Agentic RAG ideas
* Local-first workflows and offline environments
* Developers who prefer explicit code over black-box solutions

## What RAGLight is not

RAGLight does **not** try to be:

* A no-code or SaaS platform
* A fully managed production system
* An opinionated end-to-end product

Instead, it provides a **solid, extensible foundation** for building and exploring
RAG-based systems, without getting in your way.