Bessouat40/RAGLight

RAGLight is a modular framework for Retrieval-Augmented Generation (RAG). It makes it easy to plug in different LLMs, embeddings, and vector stores, and now includes seamless MCP integration to connect external tools and data sources.

What it solves

RAGLight provides a lightweight, dependency-free toolkit for building Retrieval-Augmented Generation (RAG) and agentic pipelines. It solves the problem of bloated framework dependencies by using structural protocols (duck typing) instead of rigid inheritance, allowing developers to swap out LLMs, vector stores, and embedding models without being locked into a specific vendor's SDK.

How it works

The library is built on a core of dependency-free contracts. You define your components (LLM, vector store, etc.) by implementing simple methods, or you can use the library's built-in integrations for common tools like OpenAI, Ollama, Chroma, or Qdrant. The pipeline is stateless by default, but supports session-based history and conversational query reformulation. Agentic workflows are handled by a separate AgenticPipeline that manages tool execution and decision-making loops.

Who it’s for

This is for Python developers who want a clean, modular, and type-safe way to build RAG applications without the overhead of heavy, opinionated frameworks. It is particularly well-suited for those who prefer dependency injection and want to maintain full control over their pipeline components.

Highlights

  • Zero-dependency core: The core package has no third-party runtime dependencies.
  • Structural protocols: Custom components do not need to inherit from specific base classes, making integration easy.
  • Agentic support: Includes a dedicated namespace for agentic workflows, including support for MCP (Model Context Protocol) tools.
  • Flexible configuration: Supports both direct object injection for code-first workflows and registry-backed configuration for loading pipelines from JSON or TOML files.
  • Native tracing: Built-in integration with Langfuse for observability across both classic and agentic RAG pipelines.

Related

  • Project
    superlinear-ai/raglite

    A lightweight Python toolkit for Retrieval-Augmented Generation (RAG) that supports DuckDB or PostgreSQL and integrates with various LLM providers.

  • Project
    weiwill88/Local_Pdf_Chat_RAG

    A modular Python implementation of a RAG pipeline designed for learning and experimentation, featuring hybrid retrieval and support for multiple LLM backends.

  • Project
    rag-web-ui/rag-web-ui

    An intelligent dialogue system that allows users to build custom Q&A services by combining their own document knowledge bases with LLMs using Retrieval-Augmented Generation.

  • Project
    neuml/rag

    A Streamlit application backed by txtai that implements both Vector and Graph RAG to provide factually grounded LLM responses.

  • Project
    Raudaschl/rag-fusion

    RAG‑Fusion is a search pipeline that expands a user query with LLM‑generated rewrites, runs each rewrite through both BM25 and dense‑vector search, fuses the ten result lists via Reciprocal Rank Fusion, and optionally reranks the pooled candidates with a cross‑encoder or decision model. On a medical‑corpus benchmark it improves NDCG@10 by ~+0.05 and MRR by ~+0.09 over a plain vector baseline, at a modest latency (~1.5 s) and API cost. Best for domains with terminology mismatch and recall‑heavy needs; not suited for ultra‑low‑latency or precision‑only use‑cases.