A Synology NAS serves this public site. A Mac mini on my internal network runs an AI agent that uses several models, devices and messaging apps. This page shows how the pieces fit.
Two machines, split by who can reach them.
Runs nginx in a Docker container and serves this site as static files, behind Cloudflare.
Stays on my internal network. It is the AI playground: the agent runs here, and it maintains the NAS over the local network.
One primary model, two secondary ones, and a router on the way.
primary The default model for the agent.
secondary Runs alongside the primary.
secondary Runs alongside the primary.
in progress A router decides which model handles each task.
The agent is connected to the devices I use, so it can act where I do.
Send texts and use messaging apps.
Place phone calls.
Full control over the connected devices.
Jobs start on a timer and report back to me.
The same setup, mapped onto the layers in AI Agent Anatomy.
| Layer | In my setup |
|---|---|
| 1 Model | Anthropic Sonnet (primary), Anthropic Fable 5.1, OpenAI Astra 6 |
| 2 Router | In progress |
| 3 Harness | Hermes, running on the Mac mini |
| 4 Tools | Shell, messaging, calls, and the connected devices |
| 5 Memory / Permissions | Hermes memory and saved skills, with approval prompts |
| 6 Orchestration | Scheduled jobs and chat messages that wake the agent |