hayagriva 0.1.0

Modular Retrieval-Augmented Generation (RAG) framework.

Hayagriva

Illuminate what was forgotten. Retrieve what matters.

Hayagriva is a lightweight, modular Retrieval-Augmented Generation (RAG) framework designed to help developers build production-ready AI systems that combine large language models with real-world contextual knowledge. Inspired by Hayagriva, the divine preserver and restorer of the Vedas, this library focuses on retrieving truth from stored knowledge and generating accurate, grounded responses.

Features

• Pluggable retrievers (vector databases, keyword search, hybrid search)
• Model-agnostic generation (choose any LLM backend)
• Flexible embeddings and rerankers
• Built-in document loaders and chunking utilities
• Minimal and extensible architecture
• Easy integration with existing pipelines
• Lightweight and fast
• Basic CLI and Gradio UI for experimentation

Project Structure

hayagriva/
├── core/           # Core RAG building blocks (chunker, embeddings, vector store, retriever, generator)
├── ingestion/      # Loaders and parsers for documents
├── ui/             # Gradio UI components
├── cli/            # Command line interface
├── utils/          # Utilities for logging, validation, etc.
├── config.py       # Configuration dataclasses
├── exceptions.py   # Custom exception types
└── version.py      # Version metadata

Quick Start

Python API

from hayagriva import Hayagriva, HayagrivaConfig

config = HayagrivaConfig()  # customize models or retrieval parameters if needed
rag = Hayagriva(config)
rag.add_documents(["Hayagriva restores forgotten knowledge."])
response = rag.ask("Who retrieved the lost Vedas?")
print(response)

CLI

# Launch the Gradio UI
hayagriva ui --port 7878

# Ingest local text files
hayagriva ingest ./docs

# Ask a question from the terminal
hayagriva query "Explain retrieval-augmented generation"

Gradio UI

Running hayagriva ui starts a simple two-panel UI: left for ingestion, right for chat-based Q&A. It is a placeholder designed for future expansion.

Minimum Requirements

• Python 3.10+
• Dependencies: sentence-transformers, faiss-cpu, openai>=1.0.0, gradio

Set OPENAI_API_KEY in your environment or configure it programmatically when instantiating Hayagriva.

Installation

Install the package and dependencies in a Python 3.10+ environment:

pip install -e .

Ensure you have system build tools available for faiss-cpu (the default vector store) and that OPENAI_API_KEY is set before running generation.

Roadmap

• Built-in document ingestion for PDFs, DOCX, JSON
• Streaming responses and multi-turn chat memory
• Additional vector stores (Chroma, Qdrant, Pinecone, Milvus)
• Hybrid retrieval and reranking
• Rich UI controls and analytics
• Agentic tool-use within the pipeline

License

MIT License