ragify-docs is a RAG (Retrieval-Augmented Generation) powered AI assistant that lets you upload your documents, index their content locally, and ask questions with full context awareness using hybrid LLMs — either OpenAI or Ollama.

• Do not remember your documents unless you paste content manually.
• Consume more tokens, especially when repeating context in each query.
• Do not fine-tune or remember past documents unless part of ongoing sessions.
• Your inputs are sometimes used for model improvement (unless settings are disabled).

• 🧠 Injects context from your indexed documents dynamically — no retraining required.
• 🔒 Runs locally with Ollama, or optionally uses OpenAI (no document content sent to OpenAI APIs).
• 💸 Reduces token usage with smart chunking and retrieval.
• 📂 Supports PDFs, DOCX, TXT, MD, and Google Docs.
• 📈 OpenAI Completions API guarantees your content is not used for training when using API keys.

• 📄 Ingest your private documents into a local vector store (ChromaDB)
• 🤖 Ask natural language questions and get answers from context
• 🔁 Session history with saving and loading
• 💬 Hybrid LLM inference (Ollama local or OpenAI fallback)
• 🧠 Smart embedding fallback (Ollama or SentenceTransformer)
• 📤 Export conversations as Markdown or TXT

git clone https://github.com/anandan-bs/ragify-docs.git
cd ragify-docs

# Install dependencies
pip install -r requirements.txt

# or using make
make setup

Create a .env file (or pass env vars directly):

# Required for OpenAI API access
OPENAI_API_KEY=your_openai_api_key

# Required for Hugging Face model downloads
HUGGINGFACE_TOKEN=your_huggingface_token

# Optional: Set to 'false' to disable tokenizer parallelism
TOKENIZERS_PARALLELISM='false'

These can also be overridden in ragify_docs/config.py using pydantic.BaseSettings.

ragify-docs/
├── ragify_docs/
│   ├── core/                  # Core functionality modules
│   │   ├── __init__.py
│   │   ├── chunk.py          # Text chunking utilities
│   │   ├── embed.py          # Embedding functionality
│   │   ├── load_doc.py       # Document loading utilities
│   │   ├── ollama.py         # Ollama API integration
│   │   ├── openai.py         # OpenAI API integration
│   │   ├── store.py          # Vector store operations
│   │   └── tokenize.py       # Tokenization utilities
│   ├── __init__.py
│   ├── config.py             # Application configuration
│   ├── inference.py          # RAG inference logic
│   ├── ingest.py             # Document ingestion pipeline
│   └── main.py               # Gradio UI and application entry point
├── .data/                    # Local model storage, auto-generated
├── Makefile                  # Development commands
├── README.md                 # This file
├── requirements.txt          # Python dependencies
├── setup.py                  # Package installation
└── setup_model.py            # Model setup script

flowchart TD
    %% Central Chroma DB
    DB[(Chroma Vector DB)]

    %% Document Ingestion Pipeline
    A[User Uploads Document] --> B[Load Document]
    B --> C[Chunk with Token Awareness]
    C --> D[Generate Embeddings]
    D --> DB

    %% Query Processing Pipeline
    E[User Asks Question] --> F[Embed Question]
    F --> G[Query Chroma DB]
    DB -->|Retrieve| G
    G --> H[Generate LLM Prompt with Context]
    H --> I[Display Answer in Gradio UI]

    %% Styling for better visualization
    classDef db fill:#4db6ac,stroke:#333,stroke-width:2px,color:white
    class DB db

# Navigate to the project root (if not already there)
cd ragify-docs
python3 setup_model.py
python3 ragify_docs.py


# or using make
make run

Then open your browser at http://localhost:7860

• Ollama must be installed and running if use_ollama = True
• OpenAI completions use API key securely and do not send full documents
• Chroma vector DB and document indexing happen entirely locally

Anandan B S

• Email: anandanklnce@gmail.com
• GitHub: @anandan-bs

MIT License