Every node, every room, one screen.

What to look at

1. Transcript and replyEach row is one turn: which flow handled it, what was heard and what was answered.
2. Latency per stageThe bar splits each turn into speech to text, model and speech; open a row for the exact timings.
3. Listen to itPlay what was heard or what was answered; open the turn to grade it or run it again.

What to look at

1. One card per DomiaName, temperament, role, latency and volume at a glance; the full configuration is edited live on its page.
2. Open a DomiaEach card opens that Domia's page, where you apply a template or edit its configuration; most edits reload live.
3. Thin clientsA thin client keeps wake word and audio local and delegates the rest; its routines are set on its own page and its satellites on the Satellites screen.

What to look at

1. A system templateStandalone runs the whole voice chain on one machine; inference hub, thin client and tuned variants ship beside it.
2. What a template carriesThe chips preview model and voice; underneath sit engines, delegation and skills.
3. Apply to a DomiaPick a Domia and apply; the node restarts with its new role, and later edits reload live.

What to look at

1. Three layersFacts, episodes and what each Domia knows about you sit in their own tabs; search or narrow to one Domia just below.
2. A factEach fact names the Domia that learned it, when, and how confident it is.
3. Trace backConfidence plus a link to the turn that produced it; episodes and the user model are checked the same way.