Deploy 0d0cd6c

app/models/chat.py

@@ -22,6 +22,9 @@ class ChatRequest(BaseModel):
     # Bypasses the Gemini fast-path unconditionally so pill follow-ups
     # always produce cited, chunk-grounded answers rather than TOON summaries.
     is_followup: bool = False
+    # True when submitted from microphone/transcription path.
+    # Used only for response-style tuning (audio UX), never for retrieval scope.
+    audio_mode: bool = False
 
 
 class ChatResponse(BaseModel):

app/models/pipeline.py

@@ -82,6 +82,8 @@ class PipelineState(TypedDict):
     # Stage 1: True when submitted via a follow-up pill button.
     # Bypasses Gemini fast-path so pill follow-ups always produce cited RAG answers.
     is_followup: bool
+    # True when request came from speech-to-text path.
+    is_audio_mode: bool
     latency_ms: int
     error: Optional[str]
     interaction_id: Optional[int]