Building a House-Scale AI with Muse
A user shares their experience of creating a house-scale AI system using Raspberry Pi 5 and Home Assistant. The AI, named Muse, handled engineering tasks, fixed issues, and documented the process.
added by @n3PyVpnDDx79860
My AI moved into my house. I just shouted at the ceiling; it did the engineering.
For 3 days I built a house-scale AI with Muse — mostly by talking to it.
The setup:
- Raspberry Pi 5 + Home Assistant as the hub
- A living-room voice satellite — "Hey Jarvis" wake word, room mic + speaker
- Cloud DeepSeek as the brain
- Speech recognition & neural TTS on my DGX Spark (CPU-only, never touches my GPU work)
- Music: name any song, it searches Bilibili with my premium account and plays it on my smart speaker
The fun part is what broke — and who fixed it:
- The satellite disconnected every 8 seconds. Muse traced it to a ping/pong protocol mismatch and patched the open-source Wyoming code — 6 patches.
- The TTS voice sounded robotic, so Muse built a custom server wrapping neural voices.
- Bilibili has no Music Assistant provider, so Muse built the whole search, extract, DLNA pipeline from scratch.
- This morning the speaker "accepted" songs but stayed silent. Muse proved the CDN was 403-ing the speaker's fetch — and rebuilt the pipeline to relay audio through the Pi.
- The wake word ignored me, until Muse tapped the live audio stream, measured my real voice peaking at 0.45, and recalibrated the threshold.
Then it wrote the 13-page engineering document. With diagrams.
This is what a personal agent should be: not a chatbot — a collaborator that ships, debugs, and documents real systems in the physical world.
Trying Muse? My invite code: 54SBYS