Rag-System 1.0.1

A production-quality Retrieval Augmented Generation (RAG) Python Framework.

Rag-System

A production-quality Retrieval Augmented Generation (RAG) Python Framework.

Features

• Multi-Format Ingestion: Recursively ingest and parse .pdf, .docx, .doc, .txt, .md, .csv, .xlsx, .xls, .json, .html, .htm, and .xml.
• Hybrid Search: Combines ChromaDB vector similarity with native keyword filtering ($contains).
• Flexible LLM Selection: Out-of-the-box support for Groq, OpenAI, Anthropic, Gemini, Ollama, and OpenRouter.
• Session Memory: Persistent user facts serializer (memory.json) and token-budget-aware sliding conversation history window.
• Advanced Document Filtering: Expand wildcard glob paths and folders to target retrieval down to specific files and page ranges.
• Unified Developer Namespace: Exposes a clean, high-level user interface.

---

Installation

Install using pip:

pip install Rag-System

---

Quick Start

from Rag_System import RAG

# Initialize
rag = RAG(
    model="llama3-8b-8192",
    provider="groq",
    data_dir="data",
    auto_ingest=True,
    verbose=True
)

# Ask a question
response = rag.ask("Which departments have the lowest CAP1 MCA cutoffs?")
print(response.answer)

# Print citations
for citation in response.citations:
    print(f"- {citation.filename} (Page {citation.page_number}) similarity={citation.similarity:.4f}")

---

Configuration

Settings are parsed in the following priority order:

1. Constructor arguments
2. Environment variables from .env
3. Defaults (Rag_System/constants.py)

Environment Settings (.env)

GROQ_API_KEY=gsk_...
OPENAI_API_KEY=sk-proj-...
ANTHROPIC_API_KEY=sk-ant-...
GEMINI_API_KEY=AIzaSy...
OLLAMA_BASE_URL=http://localhost:11434

---

API Reference

class RAG

Arguments:

• model: Model name identifier.
• embedding_model: Text embedding model name. Defaults to BAAI/bge-small-en-v1.5.
• provider: Target LLM provider (groq, openai, anthropic, google, ollama, openrouter).
• data_dir: File directory path to scan.
• auto_ingest: If True, run index scans upon setup.
• verbose: Toggle framework log outputs.

Methods:

• ask(question, ...): Solves intent routing, runs retrival, and returns RAGResponse.
• chat(message): Run interactive conversation storing chat history.
• ingest(force_rebuild=False): Incremental document file indexing.
• retrieve(query, **kwargs): Return candidate nodes with similarity scores.
• search(query, **kwargs): Alias for retrieve.
• summarize(query=...): Summarize indexed documents.
• compare(query=...): Compare documents.
• statistics(): Retrieve document counts and database counters.
• list_documents(): Return list of uniquely indexed filenames.
• delete_document(filename): Wipe document contents matching name.
• update_document(filename): Reindex file.
• clear_memory(): Reset conversation states.
• add_memory(key, value): Remember user preference तथ्य.
• reset_vectorstore(): Clean all collections.

---

CLI Usage

The package installs a rag subcommand group:

# Ingest documents
rag ingest --force-rebuild

# Query
rag ask "Which colleges offer MCA?" --files "data/pdfs/*.pdf"

# Stats
rag stats

# Documents list
rag documents