E.V. — Hands-Free Voice Assistant

E.V. is a hands-free personal productivity voice agent built for developers who can't afford to break their flow. Using AssemblyAI's Voice Agent API over a single WebSocket, E.V. listens always — no wake word, no button press, so you can capture a task, dictate a note, or ask for a spoken briefing without ever touching your keyboard. The agent runs as a Python CLI application, streaming mic audio to AssemblyAI's unified pipeline where speech recognition, LLM reasoning, and text-to-speech all happen server-side in one round trip. Barge-in is a native interrupt E.V. mid-sentence the moment you have a follow-up thought. E.V. ships with 14 tools: add, list, complete, and delete tasks with optional due dates; add and delete notes; keyword search across tasks and notes; a spoken briefing of overdue and due-today items; runtime voice switching across 12 voices (implemented as a clean session reconnect, the correct approach for AssemblyAI's immutable-per-session voice config); a clean session end; and persistent memory injected into the system prompt automatically on every future run. Two optional features round it out: read_screen takes a screenshot and calls Claude's vision API so E.V. can see and comment on your monitor; remember_fact writes to a local JSON store so E.V. greets you differently each session based on what you told it before. The very first thing E.V. says each session is a proactive briefing: overdue or due-today tasks are called out before you even ask. All storage is local JSON. No third-party dependencies beyond AssemblyAI and sounddevice.