ragdb 0.1.2

Lightweight embedded document store with TF-IDF search

RAGdb

RAGdb is a lightweight, embedded multimodal database for Retrieval-Augmented Generation (RAG) systems. It stores extracted text, metadata, and searchable vectors — all inside a single SQLite file.

⚡ RAGdb is the world’s first lightweight, SQLite-based, embedded multimodal RAG index with zero heavy dependencies and no servers.

No servers. No GPU. No vector database. Just:

pip install ragdb

…and you have a full local RAG index.

---

🌟 Why RAGdb?

• Embedded & portable — everything inside a single .ragdb SQLite file
• Multimodal — supports text, PDFs, Word, CSV, JSON, Excel, images, audio, video
• Fast local search using TF-IDF + cosine similarity
• Zero heavy ML dependencies (no PyTorch, no Transformers)
• No file storage — RAGdb stores extracted content, not raw files
• Natural-language ready — plug into GPT, Claude, Llama, or any LLM
• Works fully offline
• Small footprint — ideal for laptops, VMs, containers, or edge devices

RAGdb is the only open-source project that provides a full multimodal RAG index in a single file without requiring a server or vector database.

---

✨ What RAGdb Stores

RAGdb is a search index, not a backup system. It does not store your original file bytes.

For each ingested file, RAGdb stores:

• extracted text (where applicable)
• a TF-IDF vector
• metadata as JSON
• a short human-friendly preview
• absolute file path

Your actual files remain on disk or cloud — RAGdb holds only the RAG-ready representation.

---

📂 Supported Formats

📝 Full text extraction

• .txt
• .pdf (via PyPDF2)
• .docx
• .json
• .csv
• .xls, .xlsx

🖼 Images

• .png, .jpg, .jpeg, .webp, .bmp, .gif

Stored data:

• size, width, height, mode
• OCR text (requires Tesseract installed)
• human-friendly preview

Version 0.2.0 will include a built-in OCR engine (no Tesseract required).

🔊 Audio (metadata-only)

• .wav, .mp3, .ogg, .flac, .m4a

🎥 Video (metadata-only)

• .mp4, .mov, .mkv, .avi, .webm

---

🚀 Installation

pip install ragdb

Or from source:

pip install -e .

Dependencies:

• numpy
• pillow
• PyPDF2
• pandas + openpyxl
• python-docx
• pytesseract (optional OCR)
• fastapi + uvicorn (optional API server)

---

🧠 Basic Usage

from ragdb import RAGdb

# Create or load database
db = RAGdb("knowledge.ragdb")

# Ingest an entire folder
db.ingest_folder("docs")

# Search your RAG database
results = db.search("machine learning tax changes")

for path, score, media_type, preview in results:
    print(f"{score:.4f}  {media_type}  {path}")
    print("   ", preview)

---

🤖 Using With an LLM (Natural Language RAG)

RAGdb handles retrieval. The LLM handles reasoning.

Typical pattern:

1. User asks a natural-language question
2. Query RAGdb for top-N relevant pieces
3. Feed results into GPT/Claude/Llama
4. Generate an answer grounded in retrieved context

This provides semantic behavior without embedding heavy ML models.

---

🌐 Optional: FastAPI Server

Expose your .ragdb file over HTTP:

export RAGDB_PATH=/path/to/file.ragdb
export RAGDB_API_KEY=secret-token

uvicorn ragdb.server:create_app --factory --reload

REST endpoints:

• POST /ingest
• GET /documents
• GET /search

---

📌 Notes on Word Files

Old .doc files are not supported. Save them as .docx before ingestion.

---

📄 License

RAGdb is released under the MIT License.

---

💡 Coming Soon (v0.2.0)

• Built-in tiny OCR (no Tesseract required)
• Media extension (audio/video transcription, CLIP embeddings)
• Cloud embedding helpers
• Optional semantic search layer

---