raglite-toolkit

Semantic search, cited answers, an MCP server and agent tools over your documents, in a few lines of Python.

Documentation · Getting started · Providers · MCP server · Troubleshooting · Changelog

RAGLite indexes PDFs, Word files, Markdown, text, JSON, folders and web pages, then searches them and answers questions with any LLM, citing the passages it used.

• Answers with citations, including the PDF page and Markdown section of each passage
• Hybrid search: vector search for meaning, BM25 for exact terms such as error codes, or both
• Metadata filters and reranking, and raglite eval to measure retrieval on your own questions
• MCP server for Claude Code, Claude Desktop, Cursor and VS Code, and a function-calling tool for OpenAI and Anthropic agents
• Any provider: OpenAI, Anthropic, Google, Mistral, Cohere, Groq, xAI and Ollama; install only the extras you use
• Fully local option: local embeddings and Ollama, no API key
• Content-hash caching, Qdrant and Pinecone stores, streaming, a FastAPI REST API and a CLI

Install

Requires Python 3.11 or later.

pip install 'raglite-toolkit[local,anthropic]'   # local embeddings (the default) + Claude answers
pip install 'raglite-toolkit[all]'               # every provider

Quick start

import os
from raglite import Document

doc = Document("./policy.pdf", {
    "llm": {"provider": "anthropic", "apiKey": os.environ["ANTHROPIC_API_KEY"]},
})

doc.build()

hits = doc.search("refund policy", top_k=3)

answer = doc.ask("What is the refund policy?")
print(answer.text)       # "Refunds are issued within 30 days [1]."
print(answer.citations)  # [Citation(n=1, source="policy.pdf", page=12, text="...")]

Learn more

• Collections: index folders, files and URLs together
• Citations: pages and sections for each cited passage
• Hybrid search: vector, keyword and hybrid retrieval
• Metadata filters: search one customer's documents or a range of pages
• Reranking: Cohere, Voyage, an LLM or your own function
• Evaluating retrieval: raglite eval scores recall@k, MRR and hit rate on your questions
• MCP server: give your documents to AI coding tools
• Agent tools: as_tool() for OpenAI and Anthropic function calling
• Streaming, progress and cancellation
• Vector stores: Qdrant, Pinecone or your own
• REST API and CLI
• How caching works
• Options and types

Choose the Python tab on any page; the choice is remembered.

TypeScript

The same library for TypeScript: raglite-toolkit on npm (source).

Contributing

See CONTRIBUTING.md. Security issues: SECURITY.md.

License

MIT © Piyush Anand