Short answer: Aria absolutely should own this
The video Topher shared is the right unlock: Homey Pro now sells the software that ran in their hardware hub, and it runs on a Mac. The M3 is already the right place for it — it's awake, it's local, it's overprovisioned (96 GB RAM, 26 cores, < 2% disk used).
The question isn't "build vs buy." It's "which bridge software, and where does Aria sit on top." Aria's job is the conversational + cross-lane layer that turns "I'm going to bed" into 14 device actions across 5 protocols + a calendar block + a notify to Jaime if she's still up.
And the vibe-code framework (see /info/vibe-code) maps perfectly: voice memo of what you want → Aria → Homey flow → working automation. Same shape as the 3D printing pipeline.
🏗️ Recommended architecture (5 layers)
┌─────────────────────────────────────────────────────┐ │ USER — Topher voice memo, Jaime Siri, kids tap │ └────────┬───────────────────────────┬────────────────┘ ▼ natural-lang ▼ Siri / HomeKit app ┌──────────────────────┐ ┌────────────────────────────┐ │ ARIA (M4) │ │ APPLE HOMEKIT │ │ orchestrator + LLM │◄──►│ Siri / Home app │ └────────┬─────────────┘ └────────────┬───────────────┘ ▼ REST/WS ▼ HomeKit ┌──────────────────────────────────────────────────────┐ │ HOMEY PRO BRIDGE on M3 │ │ device unification: Zigbee + Z-Wave + Wi-Fi + │ │ Thread + Matter + BLE + cloud-API + Spotify │ └──┬─────────┬────────┬────────┬──────────────────────┘ ▼ ▼ ▼ ▼ ┌─────┐ ┌─────┐ ┌──────┐ ┌──────────┐ │ Hue │ │LIFX │ │Aqara │ │ Spotify │ ... └─────┘ └─────┘ └──────┘ └──────────┘ AUTONOMOUS LAYER (next to Aria): Hermes on M3 — qwen3.6/llama4:scout for always-on routines, zero token cost.
Homey Pro Bridge on M3
Easy UX, broad device support, local-first, video-validated
HomeKit + Siri for family-facing controls
Jaime + kids already know how to use Siri; HomeKit covers most modern devices natively
Aria (Claude on M4) — natural language + cross-cutting logic
Aria reads Homey state, asks Topher in plain English, coordinates with the rest of his life context
Hermes on M3 — for autonomous routines that need reasoning
Zero token cost for always-on jobs like "every 30 min, check if anyone's home and adjust"
Home-grown only for high-leverage moments
E.g., "if vibe-code-3d-print job completes, send a HomeKit notification + flash the workshop LED"
⚖️ The four options Aria considered
Homey Pro Bridge (self-hosted)
- Runs on
- M3 (Mac/PC/NAS/Docker/Pi all supported)
- Devices
- Zigbee · Z-Wave · Wi-Fi · Thread · Matter · BLE · Infrared via included radios; software-only on Mac means Wi-Fi/cloud devices + Zigbee/Z-Wave via USB stick (~$35)
- UX
- Best-in-class. Visual flow builder ("if vibration sensor → play Spotify Frank Sinatra"). Wife/kids can use it without training.
- API
- Full REST + WebSocket. Skills for Alexa + Google. Aria integration is straightforward.
- Cost
- Software self-hosted appears free (per the video); confirm at homey.app/en-us/homey-pro/. ~$35 for a Zigbee USB stick if going hardware-radio.
- Lock-in
- Low. Devices stay in their native protocols; Homey just orchestrates.
Why top/bottom of list: Topher just watched the video, the demo was the exact use case ("vibe sensor → Spotify"), the creator literally ran it on a Mac. This is the highest-velocity option.
Home Assistant
- Runs on
- M3 (Docker container or HA OS in VM)
- Devices
- 3,000+ integrations — broadest device support in the world
- UX
- Powerful but UX-hostile. The Spotify-via-Alexa flow Homey did in 30 seconds takes hours in HA (OAuth + media_player bridge + service call). Topher would not delegate this to Jaime.
- API
- Full REST + WebSocket + native MQTT. Aria integration is straightforward; ecosystem is huge.
- Cost
- Free + open-source. Optional Nabu Casa cloud at $7.50/mo for remote access (Aria-on-Tailscale removes this need).
- Lock-in
- Zero. Everything is YAML/JSON, self-hosted.
Why top/bottom of list: Right answer if Topher wants to go DEEP and stay there for years. Wrong answer if the goal is "vibe-code my home" — the UX friction is the whole problem the video sets up.
Apple Native (HomeKit + Shortcuts + AppleScript)
- Runs on
- Apple TV / HomePod / iPad as the hub; AppleScript on M3 for orchestration
- Devices
- HomeKit/Matter only — narrower but growing. Most major brands now ship HomeKit-compat or Matter.
- UX
- Clean for HomeKit devices, friction-y the moment you cross into Wi-Fi-only or off-protocol gear.
- API
- Apple Shortcuts SDK + AppleScript + the new Home API (iOS 18+). Aria can hit these via macOS scripting on M3.
- Cost
- Free, devices already in house.
- Lock-in
- High to Apple ecosystem.
Why top/bottom of list: Use as the FRONTEND for the family (Siri "turn off the bedroom lights"), not the orchestration brain. Pair with Option A or B for the heavy lifting.
100% home-grown (Aria writes every integration)
- Runs on
- M3, custom Python services
- Devices
- Only what Aria writes drivers for. Every brand is a new project.
- UX
- Depends entirely on what we build. Could be amazing for the demos we focus on; will always have a long tail of devices that don't work.
- API
- We are the API. Native MCP integration with Aria from day one.
- Cost
- Free in dollars; expensive in Aria-hours.
- Lock-in
- None.
Why top/bottom of list: Wrong answer for everyday automation. Right answer for specific high-value moments (the vibe-code-3d-printing pipeline IS this for one domain). Use sparingly, on top of A/B.
🎬 First 5 demos to ship (easiest → hardest)
1. Welcome-home routine
"When my car arrives home → lights on in office + Spotify resumes my morning playlist + ChatGPT recap of the day waits on the screen"
Complexity: Easy — Homey + LIFX/Hue + Spotify integration
2. 3D-print finished alert
"When Bambu P2S finishes a print, flash the workshop LED green + Slack the family + log to vault"
Complexity: Medium — Bambu API + Homey + Slack webhook
3. Workspace mode
"At 3pm crash time (per topher/health.md), dim office lights to focus mode + start a focus playlist + decline new Slack pings"
Complexity: Easy — Homey scheduled flow + scribe-health-style trigger
4. Quiet hours guard
"After 10pm, mute Alexa + AC down 2 degrees + scribe-health goes silent on notifications"
Complexity: Easy — Homey time trigger
5. Front door = video brief
"When the doorbell rings, capture camera feed + run Hermes vision skill: who is it, expected delivery, just a walk-by?"
Complexity: Hard — needs camera + vision model + reliable face/scene recognition. Long-tail target after the easy wins."
🤔 Concerns + Aria's answers
Privacy
Everything local on M3 / HomeKit. No data leaves the house. The Homey software model in the video specifically emphasized this.
Reliability
M3 has 22+ days uptime; Hermes/Ollama have run continuously since 4/24. Adding the home-automation load is small. Failover: if M3 dies, HomeKit + Siri continue to work for everything Apple-native.
Wife buy-in
Homey UX is the win here vs Home Assistant. Jaime should be able to make a flow herself in 60 seconds.
Cost overrun
Software is free. Hardware spend is ONLY if we want Zigbee/Z-Wave (~$35 USB stick) or to buy specific devices. No subscriptions required.
Lock-in
Devices stay in their native protocols. If Homey ever folds, we swap the bridge layer (HA, OpenHAB) and the device inventory transfers.
🚀 Next moves (decision-ready)
1. Topher confirms the recommended stack (Homey on M3 + HomeKit + Aria above). If yes, Aria handles install + integration. Time: ~2 hrs one-shot.
2. Aria inventories the house — list every smart device in the home, by protocol. Builds the actual device matrix so we know what's actually here. Topher can dump a list verbally or Aria can scan the LAN.
3. Pick the first demo — recommend "Workspace mode at 3pm" (highest leverage given the documented 3pm crash) or "Welcome-home routine" (highest emotional win). Topher's call.
4. Open OPS-NN: Smart Home — Aria as orchestrator (live build) — tracks the install + demos + device inventory. Aria can file this on confirmation.
5. Update the Smart Home lane card on the dashboard to reflect this becoming an active project, not just an infra lane.