rag-titan

A Swiss Army knife for RAG (Retrieval-Augmented Generation).

rag-titan puts many types of RAG in one small, beginner-friendly Python package. Every part (loader, chunker, embedder, vector store, reranker, generator) is a small class you can read in a minute and replace with your own.

Features

More RAG types (corrective, hybrid, multi-query, graph, agentic, ...) are planned. See the roadmap.

• Runs 100% locally with Ollama, so your data stays on your machine.
• Reads PDF and DOCX files.
• Answers are grounded in your documents and cite the chunks they used.
• Every part is swappable. You only need to inherit from one base class.
• Includes a rag-titan command line tool.

Simple RAG

Adaptive RAG

Installation

pip install rag-titan

Then install Ollama and pull the default models:

ollama pull gemma3:4b          # chat model
ollama pull nomic-embed-text   # embedding model

Quick start

from rag_titan import RAGTitan

rag = RAGTitan(rag_type="simple")        # or "adaptive"
rag.index(["handbook.pdf", "policy.docx"])

print(rag.ask("How many leave days do I get?"))
# You get 20 leave days every year [chunk_0].

Change any setting in code:

rag = RAGTitan(
    rag_type="adaptive",
    settings={"llm_model": "llama3.2", "chunk_size": 200, "reranker": "none"},
)

Command line

rag-titan ask  --files handbook.pdf --question "How many leave days?"
rag-titan chat --files handbook.pdf policy.docx --rag-type adaptive

Configuration

Settings come from environment variables or a .env file. Copy .env.example to get started. Every setting has a default.

See the configuration docs for the full list.

How it works

Agents

A small tool-calling agent loop is included. See the agents docs.

Project layout

src/rag_titan/
├── app.py            RAGTitan: the easy, all-in-one class
├── cli.py            the `rag-titan` command
├── config.py         settings from .env / environment
├── factory.py        builds the right class for each setting
├── prompts.py        every RAG prompt in one place
├── llm/              chat models (Ollama)
├── loaders/          PDF and DOCX readers
├── chunking/         split text into chunks
├── embeddings/       text -> vectors
├── vector_stores/    save and search vectors
├── retrieval/        retriever + rerankers
├── generation/       write the final answer
├── pipelines/        indexing pipeline
├── rag_types/        simple and adaptive RAG
├── agents/           tool-calling agents
└── utils/            logging

Documentation

Full docs live in docs/. To view them locally:

pip install -e ".[docs]"
mkdocs serve

Contributing

Contributions are welcome. Please read CONTRIBUTING.md.

License

MIT