What is always-here?

This project builds a personal AI companion that tracks your Apple Watch health data and phone usage to decide when to check in with you on its own schedule. Instead of sending messages at fixed intervals, the AI monitors signals like your heart rate, sleep quality, and how recently you were active across apps, then decides whether to send a message, what tone to use, and what to say. The system has four main behaviors. First, a scheduling loop runs every 20 minutes and checks how long it has been since you last interacted with the AI, what time it is, and how active your phone has been. Based on those signals, it decides whether to call an AI language model to generate a message and push it to your phone. The AI can also choose to stay silent. Second, a late-night guard watches for app usage between 1 AM and 5 AM and sends a prompt to go to sleep if you are still on your phone. Third, health data from Apple Watch (heart rate, heart rate variability, step count, sleep stages) is synced to the server via iOS Shortcuts and influences the tone of messages. Low heart rate variability leads to gentler messages, a good night of sleep allows more energetic ones. Fourth, the AI writes two short private diary entries per day based on your conversation history and health data, stored in a database for you to read. The technical setup involves iOS Shortcuts automations that send data to a self-hosted Node.js server, which stores events in SQLite and coordinates calls to an external AI API. The server also acts as a message gateway, storing a unified conversation timeline so the AI maintains context even if you talk to it from different apps or platforms. The README is written entirely in Chinese. The code examples are in JavaScript and Python.