DEV Community
Follow
I built a fully local AI assistant at 16 — no cloud, no API keys, runs on your GPU
A 16-year-old from Pune, India, named Shriisoot has developed O-AI, a fully local AI desktop assistant. Frustrated with cloud-based assistants compromising privacy, they created O-AI to function like JARVIS from Iron Man, prioritizing personal and private interaction. O-AI runs large language models locally using llama.cpp or Ollama, requiring no internet connection or external API keys. It features a self-learning core that remembers conversation facts and a fine-tuning pipeline for personalized model training. The assistant supports voice control in English, Hindi, and Marathi via a local Whisper implementation, responding in the user's spoken language. O-AI offers various modes, including a JARVIS-like interface, PC automation, and animated desktop pets. It boasts over 30 automation capabilities for tasks like opening applications, web searches, and media control. A multi-step agent system allows for complex planning, execution, and verification of actions, supported by various step types. The technology stack includes Python with Flask for the backend and Electron with vanilla JavaScript for the frontend. Significant challenges included implementing outcome verification for agent tasks and robust content validation to prevent errors.
