VectorBrain is a self-hostable, full-stack Retrieval-Augmented Generation (RAG) system built for working with important and sensitive files. You can ingest your own documents, query them using natural language, and get context-aware answers backed by embedded source content.
This project is ideal for internal knowledge bases, secure document analysis, and custom AI assistants—all while keeping your data private and under your control.
- ✅ Self-hosted: No external APIs required
- 🔎 Document-aware: Upload files like
.txt,.csv, and ask natural-language questions about them - 🔒 Secure: Keep all document processing local
- ⚡ Fast: Powered by local LLM inference with real-time streaming support
- 📂 Searchable: Built-in document management and multi-file support
- Backend: Python + FastAPI + LlamaCpp (local LLM)
- Frontend: React + Tailwind + Vite
- LLM Support: GGUF models running via llama.cpp
| Part Number | Part Name | Quantity | Unit Price (USD) | Total Price (USD) | Supplier | URL |
|---|---|---|---|---|---|---|
| PN-001 | Resistor 10kΩ | 100 | 0.01 | 1.00 | Digi-Key | https://www.digikey.com/en/products/detail/yageo/RC0603FR-0710KL/732542 |
| PN-002 | Capacitor 100uF | 50 | 0.05 | 2.50 | Mouser | https://www.mouser.com/ProductDetail/Nichicon/UFW1C101MPD?qs=sGAEpiMZZMtZ1n0r9vR22Q%3D%3D |
| PN-003 | LED Red 5mm | 200 | 0.03 | 6.00 | SparkFun | https://www.sparkfun.com/products/9590 |
| PN-004 | Microcontroller ATmega328 | 10 | 2.50 | 25.00 | Adafruit | https://www.adafruit.com/product/123 |
| PN-005 | Voltage Regulator LM7805 | 15 | 0.75 | 11.25 | Jameco | https://www.jameco.com/z/LM7805-Voltage-Regulator-5V-1A-TO-220-Linear_51263.html |
git clone https://github.com/yourusername/vectorbrain.git
cd vectorbrainpython -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
pip install -r requirements.txtuvicorn backend.main:app --host 0.0.0.0 --port 8000 --reloadMake sure to set your model path and config in backend/main.py or your .env file if applicable.
cd frontend
npm install
npm run devThis will start the frontend at http://localhost:5173
- Upload
.txtor.csvfiles via the Documents tab - Switch to Chat and ask questions like:
“Summarize the content of the uploaded document.”
“What does the safety protocol say about emergency stops?”
The system will retrieve the most relevant chunks from your files and stream a response from the LLM.
- You’ll need a compatible GGUF model. We recommend a quantized LLaMA 2 or Mistral model for fast local inference.
- Make sure
llama.cppis compiled with GPU support if you want fast response times on consumer GPUs (like 4070/4090).
MIT License © 2024