VOX Scheduler

VOX Scheduler is AR Labs' design for a voice-first AI calendar assistant for an existing Google Calendar. It is designed to have no calendar of its own: the person speaks or types what they want ("move my three o'clock to tomorrow", "what does my day look like?") and the assistant reads and changes events in Google Calendar, then confirms by voice or text.

The first prototype was an AI agent (a language model run in a loop with tools) built with LangGraph, using the ReAct pattern of reasoning and tool calls, with Google Calendar tools exposed through the Model Context Protocol (MCP) to list, create, update and delete events, and a basic Telegram bot as the way in. A later design moved to a web app built on the same voice pipeline as Salesbot, our browser voice agent, after a pre-mortem found that calling speech-to-text, the model and text-to-speech one after another would leave the user waiting several seconds per reply. A streaming pipeline overlaps those steps.

flowchart LR
  U["User: voice or text"] --> W["Web app"]
  W <-->|"WebRTC audio"| P["Pipecat voice pipeline"]
  P --> M["Claude with calendar tools"]
  M <--> G["Google Calendar API"]
  P -->|"spoken confirmation"| W

The web design is documented, with architecture decision records and a slice-by-slice build plan; the backend and client code for it is a scaffold, and the build has not started. The direction has since widened from a calendar assistant to a personal planning assistant that connects goals and priorities to calendar time, with voice as the way of using it. The planning cascade that product would carry, from quarterly objectives through monthly goals and weekly priorities to calendar hours, already runs in VOX, the lab's working system, as a set of skills, packaged instructions an agent loads only when a task needs them (see scheduling by domains and modes). The product itself is paused, and nothing is deployed.