Home

Already have a RAG? Evaluate it → New here? Get started →

RAG lets a language model answer questions about your documents. It first finds the most relevant pieces of text, then asks the model to answer using only those pieces.

With rag-titan you get:

• Many RAG types, one API. Switch with one word (rag_type="adaptive"), and see every step a type took with run(). → RAG types
• Answers grounded in your files. The model answers only from your documents and cites the chunks and pages it used, running locally with Ollama.
• Chunking your way. Eight chunkers, from fixed word windows to whole sentences, paragraphs, heading sections or pages. → Chunking methods
• Measure, don't guess. Compare any chunker with any RAG type on your own questions in one command. → Evaluating RAG
• Find the stage that fails. Score chunking, embedding, retrieval, reranking, context and answers of any RAG system from its logged outputs, with a 0 to 100 scorecard and regression checks. → Evaluating every stage
• Already have a RAG? Evaluate it in 6 simple steps, no model needed. → Evaluation cookbook
• Watch it work in your browser. The RAG playground lights up every step on a live flow diagram; the chunk playground shows how text is cut.
• Easy to change. Swap the LLM, embedder, vector store, chunker or RAG type by inheriting one base class. → Extending rag-titan

from rag_titan import RAGTitan

rag = RAGTitan(rag_type="simple")
rag.index(["handbook.pdf"])
print(rag.ask("How many leave days do I get?"))

Where to go next

I want to...	Read
install it and ask my first question	Getting started
learn evaluation with copy-and-run recipes	Evaluation cookbook
run RAG types in my browser	RAG playground
choose a RAG type	RAG types
compare setups on my own questions	Evaluating RAG
score every stage of a RAG pipeline	Evaluating every stage
look up a metric and its formula	Evaluation metrics
understand each part	Building blocks
change models or chunk sizes	Configuration
add my own vector store or LLM	Extending rag-titan