Hermes Incident Commander

What Hermes Incident Commander does

The project was built on Hermes Agent for the 'Show us what Hermes Agent can do' hackathon. The repository description lists memory, skills, cron, gateway, subagents and Atropos RL among the Hermes capabilities it uses. It later gained a standalone watchdog that runs on any real Linux host with only an Anthropic API key and no Hermes installation, and the README says the agent learns from each incident it resolves.

Recent additions include notification channels for Discord, Slack, PagerDuty, Microsoft Teams and a generic webhook with a configurable JSON template, quiet hours and maintenance windows, flapping detection with suggested threshold fixes, and Prometheus gauges on /metrics. An offline dashboard filters incidents and exports CSV, incident_db.py provides stats and a --prune option with an optional --archive DIR, and install-watchdog.sh --with-prune-timer sets up a daily systemd timer.

Key features

• Detection, diagnosis and self-healing of production incidents
• Standalone watchdog that needs no Hermes installation
• Notifications to Discord, Slack, PagerDuty, Microsoft Teams and generic webhooks
• Quiet hours, maintenance windows and flapping detection
• Prometheus gauges on /metrics and an offline CSV-exporting dashboard
• Retention tools: incident_db.py --prune with optional archiving

When to use it

• Watching a Linux host for CPU, memory and disk breaches and getting a diagnosis report
• Suppressing alerts during a known nightly batch window
• Feeding watchdog state into an existing Prometheus and Grafana setup

Who it is for: SRE and DevOps engineers who want an AI agent to triage incidents, with or without a Hermes install.

How it fits with Hermes Agent

Originally built on Hermes Agent for a hackathon; the standalone watchdog can also run without Hermes.

Requirements: For the standalone watchdog, a Linux host and an Anthropic API key; no Hermes installation is needed

FAQ