Nash Life Manager

Overview

Nash is a local-first AI assistant I am developing to turn voice and text into structured memories, goals and scheduling workflows. It combines a FastAPI backend, LangGraph agents, SQLite persistence and a React interface.

Voice to durable context

Local speech transcription feeds diary entries and structured semantic units. The application combines chronological memory with local vector retrieval so that useful context can persist across conversations.

Explicit control over agent actions

The calendar engine distinguishes direct instructions from suggestions inferred from diaries or goals. Suggestions remain pending until approved, fixed commitments are protected during replanning, and schedule changes have an audit history.

System boundaries

Storage, audio, transcription and vector retrieval stay local. The default reasoning path uses the OpenAI API with pseudonymization, so this is not an entirely offline system. Ollama remains an optional alternative.

Current state

This is an active local prototype, not a hosted service. Implemented modules include diary and memory, goal discovery, calendar suggestions and personal tracking. A separate encrypted phone voice inbox has been implemented locally but is not deployed or integrated into the main processing flow.

The repository contains the committed development snapshot; later local experiments are described separately. No personal recordings or databases are included.

Technologies