ATLAS: Our Personal Offline JARVIS
The Spark We love cloud AI, but we hate the idea of a server constantly listening to our room or reading our private code. We wanted our own JARVIS—something fast, capable, and 100% private. So, we built ATLAS: a fully offline, voice-controlled AI assistant that runs right on our everyday hardware.
What It Does ATLAS is basically a hands-free coding buddy. You just say "Atlas," and it wakes up to help you create, read, or edit files locally. It manages tasks, launches our IDE, and even remembers our preferences across sessions—always asking permission before storing anything personal.
Under the Hood We wrote the core system in Python and glued together some amazing open-source tech: Ollama for the local LLM, faster-whisper for offline speech-to-text, and pyttsx3 for the voice. We hooked all of this up to a SQLite database so it has actual long-term memory.
The Hardest Part Getting a wake-word listener, an STT engine, and an LLM to run at the same time on a standard CPU without melting our computers was incredibly tough. We had to build a custom threaded event loop just to stop the microphone inputs from crashing into each other.
Why We're Proud We actually built a functional, local JARVIS with a low CPU footprint. It feels amazing when ATLAS naturally asks, "Should I remember that?" and actually brings that context up in a conversation a week later.
What's Next We want to add local codebase embeddings next. That way, ATLAS won't just write individual scripts—it will understand the entire architecture of whatever project we're working on.
Log in or sign up for Devpost to join the conversation.