Automation

Every night, the system captures a snapshot from the pet room camera via RTSP, analyzes pet activity patterns throughout the day, and sends a summary to the owner. It tracks feeding, drinking, sleeping, and movement across five cats — making sure everyone is healthy and the food supply is adequate.

Full system backup runs twice daily — once in the quiet hours, once at noon. It snapshots all agent workspaces, memory databases, configuration files, and session histories. Old snapshots are automatically pruned and repacked to keep storage lean. Recovery is tested periodically to ensure backups are not just stored, but actually restorable.

Each specialized agent — Researcher, Coder, Content Strategist, and Trader — runs a self-review of its previous day's work. The review summarizes completed tasks, extracts lessons learned, identifies mistakes worth remembering, and writes findings to the agent's personal memory. Think of it as a daily retrospective, run by the agents themselves.

Housekeeping that prevents entropy. Scans for broken or errored cron jobs and removes them. Cleans up orphaned session files older than 3 days. Verifies that all active cron tasks are in a healthy state. The system stays lean so nothing silently rots in the background.

A daily health reminder sent via iMessage at a fixed time. Tracks medication cycles — when to take, when to pause — and adjusts the message accordingly. Simple but reliable: no app needed, no notification fatigue, just a gentle ping at the right moment.

The memory system runs a full pipeline: scan raw memory files, extract structured facts via LLM, deduplicate overlapping entries, classify into categories (personal, tasks, agents, system), assign time-to-live values, and export to the vector search index. This is how agents remember what matters and forget what doesn't. Built on a three-paradigm fusion of Mem0 (fact extraction), Zep (temporal decay), and MemGPT (agent self-management).

Once a week, the decay engine runs: expired memories are summarized into lessons learned before being archived. Archived entries are removed from the active index but remain searchable — nothing is ever truly deleted, just gracefully aged out. This prevents memory bloat while preserving institutional knowledge.

Checks all 9 plugin source repositories for new git commits. If meaningful changes are detected — new features, architecture changes, new tools — it generates a list of recommended website updates and notifies the owner. No automatic edits to the website; changes require human approval first.

An automated security sweep across the entire agent infrastructure. Reviews session histories for suspicious patterns — unauthorized data access, unexpected tool usage, sensitive file reads. Scans system processes for anomalies. Checks file permissions, SSH configurations, and network connections. Findings are logged and flagged for human review if anything looks off. Built on top of the Security Shield module.