
cmd	feat: containerd runtime with restart policy watchdog, exec user support, and systemd-native install	7 days ago
internal	feat: containerd runtime with restart policy watchdog, exec user support, and systemd-native install	7 days ago
.gitignore	chore: initial scaffold for aegis-agent	7 days ago
Dockerfile	feat: containerd runtime with restart policy watchdog, exec user support, and systemd-native install	7 days ago
README.md	docs: add README	7 days ago
go.mod	refactor: replace Docker REST proxy with containerd runtime abstraction	7 days ago

aegis-agent

Node agent for Aegis — runs on each managed host and connects back to the Aegis control plane over an mTLS WebSocket tunnel.

What it does

Manages containerd containers on the local host (deploy, start, stop, restart, remove)
Enforces container restart policies via a watchdog goroutine
Streams exec sessions (interactive terminal) back to the Aegis UI
Sends heartbeats with node metrics (CPU, memory, GPU, container counts) every 30 seconds
Optionally manages existing Docker (moby) containers during migration to containerd-native

Requirements

Linux host with containerd installed and running
ctr CLI (bundled with containerd)
systemd (for agent process management)

Installation

Generate a join token from the Aegis UI (Nodes → Add Node), then run the displayed install script on your target machine. It will:

Pull the agent image via ctr
Install a systemd service (aegis-agent.service)
Join the node to your Aegis instance

No Docker required.

Architecture

Aegis (control plane)
    │  mTLS WebSocket (port 8766)
    ▼
aegis-agent
    │
    ├── containerd socket  (/run/containerd/containerd.sock)
    └── logs               (/var/log/aegis/)

Agent config is persisted to /var/lib/aegis-agent/agent.json after the initial join so the agent survives restarts without re-joining.