AnovaX: A Local, Multi-Agent Voice Assistant with LLM Planning, Typed Executors, and Adaptive Recovery

Short summary

AnovaX is a local-first, multi-agent desktop voice assistant that runs entirely on the user's computer, using a Gemini LLM planner to emit JSON tool-call plans executed by typed child agents (AppAgent, BrowserAgent, TypingAgent, etc.). It includes a whitelist safety layer, adaptive recovery loop for failed steps, and a Flask server for phone-based remote control over local WiFi. The project demonstrates that a few-thousand-line assistant can open apps, type, search, coordinate concurrent actions, and recover from failures without sending audio to the cloud.

• •Local-first voice assistant runs entirely on-device with Gemini as the LLM planner
• •Typed child agents handle apps, typing, browsing with individual timeout and retry policies
• •Phone-friendly Flask remote mirrors agent events and streams screen over local WiFi

Generated with AI, which can make mistakes.