To get started on building it yourself, you could consider a stack like: Python as the base, Ollama for local LLM serving, Whisper (e.g., via whisper.cpp) for STT, Langchain for the agent orchestration and tool integration, and something like Piper for TTS. For Notion/GCal, their official Python SDKs wrapped as Langchain tools would be the way.

This selection prioritizes a pragmatic balance between capability and a more manageable entry point for an individual project, rather than focusing strictly on the absolute SOTA or most experimental options for each component. It should provide a solid foundation for iterative development, and you can always swap out/upgrade specific parts later as you see fit