Merge remote-tracking branch 'origin/main'

docs/reference/node-types.md

@@ -509,10 +509,16 @@ living room setpoint to 22."), so the model is asked once per request. A
 target with `values` is offered only those, each explained to the model, and
 confirmed in its word ("Canopy open."). Any other value is refused.
 
-Every request ends in a reply:
+Questions about the home are answered from the context inputs, other
+questions from what the model knows.
 
-- When no tool fits or the request makes no sense, the model says so, as
+Every request ends in a reply, except talk the panel overheard:
+
+- When no tool can make the change asked for, the model says so, as
   "Sorry, I can't do that.".
+- Words that are neither a request nor a question, such as a TV or people
+  talking to each other, get no reply. The model answers them "Ignored.",
+  which is not said.
 - When no provider is set up, the connection is refused, it takes over 3
   seconds to connect or 15 seconds to answer, the reply is "The assistant
   isn't available right now." and the node reads degraded. A model that has to
@@ -543,6 +549,10 @@ the assistant's prompt, both `reply` outputs go to `tts`, and the device's
 | `partial` | `str` | the words agreed on so far, as they come, then the final text |
 | `text` | `str` | what was said, once per utterance |
 
+| Setting | Default | Notes |
+|---|---|---|
+| `prompt` | empty | words the speech is likely to hold, such as `Living room, kitchen, driveway, appliances.`, sent as the transcription prompt |
+
 The server is any that speaks OpenAI's audio API, such as `parrot serve`. Each
 utterance is one realtime transcription session (`{url}/v1/realtime`), opened
 when `listening` turns true or at the first chunk and committed when it turns
@@ -552,6 +562,11 @@ chunk waits for it. An utterance that comes back empty, fails, or gets no
 chunk for 10 seconds publishes "Didn't catch that." on `partial` alone, so a
 caption shows it and nothing speaks it. A lost one is logged.
 
+A `prompt` naming the installation's rooms and devices helps Whisper with
+words it rarely hears. On the recordings of one wall panel, small.en with such
+a prompt heard 30 of 42 hard commands right instead of 8, as "driveway" for
+"drive bay" and "bathroom" for "best room".
+
 `listening` has to hold a value before a chunk wakes the node. Declare it as a
 flow input with the initial value `false` when a device may never send it,
 such as one that sends whole clips.