Local Minds is a lightweight, CPU-friendly conversational AI interface built on llama-cpp-python. It includes a memory system for personalized, context-aware interactions — even on older hardware.
- Lightweight and optimized for CPUs (Raspberry Pi, old laptops)
- Runs local GGUF models (e.g., Gemma, Qwen)
- Long-term memory with context retrieval via
sentence-transformers - Multi-session chat memory support
- Minimalist yet modern front-end
- Fast boot with Flask backend
project/
├── app.py # Flask backend
├── memory.py # Memory logic with embeddings
├── requirements.txt # Python dependencies
├── templates/
│ └── index.html # Frontend HTML
├── static/
│ └── assets/ # Frontend
├── models/ # Place your .gguf model file here
└── memory/ # Stores session memories (auto-created)
git clone https://github.com/lynthera/bitsegments_localminds
cd bitsegments_localminds
pip install -r requirements.txtYou must have Python 3.10+ installed. CPU usage is default; GPU support can be enabled by changing n_gpu_layers=-1.
If you are using Windows, make sure that cmake and nmake are installed on your system.
python app.pyThen visit: http://localhost:5001
Place your .gguf model file under the models/ directory. Please place only one model in the directory. Switching between models is not possible at the moment.
The assistant uses sentence-transformers to embed past conversations and retrieve the most relevant memories per session.
Memory is stored in JSON files under ./memory/ by session and chat ID.
- llama-cpp-python
- flask
- flask-cors
- sentence-transformers
- scikit-learn
Install via:
pip install -r requirements.txt
Used Google Gemma-2-2B-IT(IQ3_M) Model
If you have similar projects and would like to work together, contact me. I'd be happy to build something together :) Email:beratdurakuniversity@gmail.com
MIT License