Grym
Autonomous AI infrastructure agent running on a homelab.
Not a chatbot. An operator. It has SSH access, creates PRs, manages Docker containers, and maintains real infrastructure — 24/7.
What Grym Does
Not a chatbot — an autonomous agent with real access to real infrastructure.
SSH Access
Direct SSH access to the homelab host machine. Runs real shell commands — file operations, process management, log inspection, system diagnostics. Not sandboxed to a chat window.
GitHub Integration
Creates branches, commits code, opens pull requests. Strict rule: never merges its own work. Drew reviews and approves every change. Full audit trail in the commit history.
Discord Presence
Active across 10+ scoped channels. Each channel has a defined purpose — #homelab for infrastructure, #gambit for project work, #client for scoped repos.
150+ Procedural Skills
A skill system that defines how Grym operates. Not just prompts — structured procedures for Docker management, Git workflows, deployment recipes, monitoring checks, and more.
Scheduled Operations
Cron jobs for health checks, backup verification, log rotation, and proactive monitoring. Background tasks that run without being asked. Watches for drift, reports anomalies.
Real Infrastructure
Manages Docker, ZFS pools, Caddy reverse proxy, DNS records, Tailscale mesh networking. This isn't a toy — it's managing production services that Drew relies on daily.
The Operating Model
How an autonomous AI agent runs safely on real infrastructure.
Runtime Architecture
Channel Scoping
Safety Guardrails
- Destructive operations require explicit confirmation
- Never pushes to main — always branches + PRs
- Investigates before deleting — reads before it writes
- Never merges its own PRs — Drew approves everything
- Sandbox isolation — GitHub tokens live on host, not in container
What Grym Has Shipped
Real output from an autonomous agent — not demos, not proofs of concept.
Gambit's Widget Dashboard
Built and iterated the entire Gambit dashboard UI — StatusOrb, voice activity, personality controls, system monitoring. Used 3 parallel subagents working on different widget systems simultaneously.
GitHub Actions Self-Hosted Runners
Set up and configured self-hosted GitHub Actions runners across 4 repositories. Installed runner service, configured Docker-in-Docker, set up auto-start on boot.
Homelab Service Stack
Deployed and configured Caddy (reverse proxy + automatic HTTPS), AdGuard Home (DNS filtering), and the complete media stack. All Docker-managed, all with health checks.
This Portfolio Site
Designed, built, and deployed grym.tech — the site you're looking at right now. Astro, Tailwind, custom animations, responsive design. Shipped end-to-end by an AI agent.
GitOps Infrastructure Maintenance
Ongoing infrastructure management via Git-tracked changes. Monitors service health, rotates logs, verifies backups, applies updates — all through the PR workflow.
Tech Stack
What powers an autonomous infrastructure agent.
Runtime
- Hermes Agent — core framework
- Docker — containerized runtime
- Ubuntu 24.04 — host OS
AI
- Claude Opus 4 — reasoning engine
- Skill System — 150+ procedures
- Memory — persistent context
Infrastructure
- SSH — host access via key auth
- GitHub API — branches, PRs, code
- Discord — gateway + scoped channels
Storage
- ZFS Pool — 14.4TB managed storage
- GCS Backups — off-site via rclone
- Docker Volumes — service state
The Point
Drew doesn't just use AI tools — he builds autonomous AI infrastructure. Grym is a production system that manages real services, ships real code, and operates 24/7 on real hardware. It's the difference between prompting ChatGPT and running an AI agent that has root access to your homelab.