Autonomous Agent Operations Team
Autonomous, multi-agent operations for the home lab · LLM + Linux automation
Summary
An autonomous operations team made of LLM agents that monitor and self-heal my home lab around
the clock. Eight specialist "L2" agents each own a domain (security, DNS, networking,
containers, backups, updates, media, storage), a manager agent supervises them, and an
"L1 service desk" handles alerts and chat over WhatsApp. The structure mirrors a real L1 to L3 IT support model, with me as the final escalation point.
01How it runs
Each agent runs on its own schedule, performs health checks over SSH against the server, catches configuration drift, remediates common failures automatically (restarting a wedged container, remounting a dropped share, flagging a stale backup) and writes a report plus a change-log entry for everything it touches. The manager runs every 15 minutes doing freshness checks, gap analysis and a job-description audit so no domain silently goes unmonitored. Anything needing a human bubbles up to the WhatsApp desk.
02The agents
| Agent | Cadence | Responsibility |
|---|---|---|
| cyber | 2h | Security posture: SSH, firewall (UFW), gateway audit. |
| docker | 2h | Containers, the smart-home stack, disk & GPU health. |
| media | 3h | Jellyfin media server health. |
| network | 30m | Internet uplink, mesh Wi-Fi nodes, latency. |
| dns | 8h | Pi-hole DNS resolution & blocklist health. |
| backup | 6h | Verifies config backups to the NAS are recent. |
| updates | daily | OS packages & Docker image staleness. |
| nas | 4h | NAS capacity & share availability. |
| manager | 15m | Supervises the fleet: freshness, gap analysis, audits. |
03Under the hood
| Component | Tool | Role |
|---|---|---|
| Reasoning | Claude + local models | The agents' decision-making; local models served by the private Ollama stack for offline tasks. |
| PA services | lanky-pa, pa-agent | Containerised assistant/agent services (personal-assistant + agent runner) exposing the team's HTTP endpoints. |
| Scheduling | systemd timers | Fire each agent on its cadence; the manager on a 15-minute timer. |
| Execution | Bash + SSH | Agent scripts run health checks and remediations against the host over SSH. |
| Service desk | WhatsApp gateway | The L1 layer: pushes alerts and takes chat requests from my phone. |
| Audit trail | Reports + change-log | Every run writes a latest-report file and appends a structured change-log entry. |
04What this demonstrates
- Practical AIOps / SOC automation: turning monitoring and triage into an autonomous, tiered agent workflow.
- LLM orchestration: multiple cooperating agents with defined roles, a supervisor, and human escalation.
- Linux / infrastructure ops: SSH automation, systemd scheduling, self-remediation and auditable change tracking.
- Reliability engineering mindset: the same detect → remediate → report loop a real ops team runs, applied to a live 35-container estate.