MetaClaw framework trains AI agents while you're in meetings by checking your Google Calendar

MetaClaw connects to multiple LLM providers via the OpenClaw platform.

An OMLS watches sleep periods, OS keyboard/mouse inactivity, and Google Calendar events to find idle training windows; a training window opens when the calendar shows the user is in a meeting.

On a failed task, a separate model distills a compact behavioral rule and injects it into the agent's system prompt for all future tasks.

The paper identifies three main rule types—normalizing time formats, making backups before destructive file operations, and following naming conventions—and training uses only post-change data.

Training runs as cloud-based reinforcement learning with LoRA fine-tuning, and rule updates are applied without taking the service down.