
Episode #348
The Brain Asks, The Code Decides
When you ask your smart home to add milk to the list, the AI doesn't add anything—it just makes a request. This episode reveals how guist, the open-source home server, keeps your smart speaker safe by giving the language model a menu of actions instead of the master key. We trace one voice command from wake word to spoken response, explore the three rules that prevent the AI from controlling your home, and examine why code, not algorithms, should always decide what happens next. 00:00:00 - Language models don't execute commands, they ask 00:01:35 - How the Dot listens without always recording 00:03:19 - The receptionist and the master key: permission model 00:04:53 - Three rules that prevent the AI from breaking out 00:06:25 - Menu, not the keys: the core design philosophy 00:07:58 - Every change tracked in git 00:09:29 - One callback powering the whole flow 00:11:14 - Tools on a shelf: architecture as metaphor --- Sources & further reading: • [internal, read 2026-10-03] home/brain/brainproxy.py, todos.py, sttgemini.py • turnlog.py. • [internal] home/README.md (voice assistant diagram), home/HANDOFF.md ("Next"). • [internal] home/docs/podcast/PODCAST-CONTEXT.md §1.5, §1.7, §2, §3, §4, §5.3. • [internal] guist-bot/docs/specs/desk/spec.md ("What it is", item 2). **Contains business • details outside this episode's scope. Quote only the shape.** This podcast episode was fully generated by AI — research, script, voices, and production. Built with Claude, Piper TTS, and automated pipeline tooling.

