LunarStudio – Lightweight Local AI Desktop App (C++ + llama.cpp)

LunarStudio is a fast, privacy‑focused offline AI assistant built using C++, Flutter, FAISS, SQLite, and llama.cpp. It performs vector search, embedding generation, and LLM inference fully offline inside your system.

---

🚀 Features

• Fast local inference using GGUF models
• Semantic search powered by FAISS
• Sentence embeddings with MiniLM
• Local database storage using SQLite
• 100% offline & private
• Lightweight compared to Electron or web view frameworks

---

📥 Download Required Models

Before running LunarStudio, download the following models into the models folder.

1. All-MiniLM-L6-v2 (Embeddings Model)

cd models/ wget https://huggingface.co/leliuga/all-MiniLM-L6-v2-GGUF/resolve/main/all-MiniLM-L6-v2.F16.gguf

2. Qwen2.5 0.5B Instruct (Main LLM) (We releasing fine tuned model soon..)

cd models/ wget https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q8_0.gguf

---

🔧 Building the Project

Dependencies

cmake make gcc faiss sqlite openblas lapack nlohmann-json llama.cpp

Build Steps

cd lunar-studio/engine mkdir build && cd build cmake .. make -j$(nproc)

---

🧠 How It Works

1. Embeddings generated using MiniLM
2. FAISS vector search
3. Qwen2.5 LLM generates response

---

🙌 Contributing

PRs and issues are welcome!

---