AI / ML

Jarvis - AI Voice Assistant

A local, agentic voice assistant inspired by Iron Man's JARVIS - a local Ollama LLM with a Kokoro TTS engine that calls tools to automate day-to-day tasks.

View Code

Overview

A fully local, privacy-first voice assistant inspired by Iron Man's JARVIS. It runs a local Ollama language model with a Whisper speech-to-text and Kokoro text-to-speech pipeline, and uses a plugin architecture where a central dispatcher routes requests to autonomous agents (calendar, weather, news, search, smart home, timers, tasks, and a browser agent).

Highlights

Multi-threaded audio pipeline with a Porcupine wake word (~50ms), real-time Whisper transcription, and sentence-buffered TTS streaming for fast, natural responses.
A lightweight router classifies requests across 15+ functions and manages GPU memory by loading and unloading models on demand.
PySide6 desktop UI with streaming chat and a vision-language browser agent (Playwright) for autonomous web tasks.
Plugin design lets new agents be added with zero changes to the core.

Built with

Python

Back to all projects