xiry.

Development · GitHub repo

RAGLite

Lightweight Python toolkit for RAG using DuckDB or PostgreSQL

From the Xiry library · Updated 10 October 2026

Visit github.com ↗

RAGLite is an open-source Python library designed to build Retrieval-Augmented Generation pipelines without heavy frameworks like LangChain or PyTorch. It provides hybrid vector and keyword search on DuckDB or PostgreSQL, late chunking, reranking, and adaptive retrieval. It also includes an MCP server and an optional Chainlit interface.

What it's for

  • Build a local or server-based RAG application using DuckDB or PostgreSQL
  • Expose an internal document store to Claude desktop via Model Context Protocol (MCP)
  • Run hybrid vector and keyword search with late chunking and semantic splitting
  • Deploy a ChatGPT-like document assistant to Slack, Teams, or the web

Features

  • DuckDB (FTS + VSS) or PostgreSQL (tsvector + pgvector) database backends
  • Model support via LiteLLM and local execution through llama-cpp-python
  • Semantic chunking and multi-vector late chunking embeddings
  • Built-in FastMCP server for Claude desktop integration
  • Hybrid keyword and vector search with reranker integration (FlashRank default)
  • Optional ChatGPT-like UI via Chainlit
  • Closed-form linear query adapter using orthogonal Procrustes problem solving

Pricing

Open source

Details

Type
GitHub repo
Category
Development
Topic
LLM Frameworks
Open source
Yes
Platforms
Code library, Command line, macOS, Linux, Windows
For
Python developers building RAG pipelines who prefer minimal dependencies over monolithic frameworks
Alternative to
LangChain, LlamaIndex
Tags
rag, python, duckdb, pgvector, litellm, mcp, search

Why it was saved

  • @Sumanth_077, 9 December 2024: Configurable RAG stack (any LLM via LiteLLM, any reranker, PDF-to-Markdown ingestion, hybrid keyword+vector search) with only lightweight permissive dependencies — no PyTorch or LangChain.

Compare

Related

  • NLWeb

    Open protocols and tools for conversational website interfaces via MCP

  • Firecrawl

    Turn websites and documents into clean, LLM-ready markdown or structured data.

  • awesome-llm-apps

    Open-source collection of runnable AI agent, skill, and RAG templates

  • "Easy Mode" Open Source AI Stack

    An opinionated reference architecture for building full-stack open-source AI applications

  • Composio

    Tooling and authentication layer connecting AI agents to external applications

  • Context+

    MCP server providing structural AST indexing and semantic navigation for coding agents

Updated