From 66949c07d85d0df51e626cda347ec557eacbe1de Mon Sep 17 00:00:00 2001
From: Matt <matt@letsbe.solutions>
Date: Sat, 28 Mar 2026 14:42:04 +0100
Subject: [PATCH] fix: accumulate transcript chunks into single messages per
 turn

Streaming transcription words now merge into one bubble per agent/user turn
instead of creating separate entries for each chunk.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 .../configurator/VoiceAgentProvider.tsx           | 15 ++++++++++++++-
 1 file changed, 14 insertions(+), 1 deletion(-)

diff --git a/src/components/configurator/VoiceAgentProvider.tsx b/src/components/configurator/VoiceAgentProvider.tsx
index 1747953..ef6e099 100644
--- a/src/components/configurator/VoiceAgentProvider.tsx
+++ b/src/components/configurator/VoiceAgentProvider.tsx
@@ -132,6 +132,7 @@ export default function VoiceAgentProvider({ locale, children }: VoiceAgentProvi
   const [completedBrief, setCompletedBrief] = useState<string | null>(null);
   const [completedFormData, setCompletedFormData] = useState<WizardFormData | null>(null);
 
+  const turnCompleteRef = useRef(true);
   const wsRef = useRef<WebSocket | null>(null);
   const mediaStreamRef = useRef<MediaStream | null>(null);
   const audioContextRef = useRef<AudioContext | null>(null);
@@ -141,7 +142,15 @@ export default function VoiceAgentProvider({ locale, children }: VoiceAgentProvi
   const animFrameRef = useRef<number>(0);
 
   const addTranscript = useCallback((role: 'user' | 'agent', text: string) => {
-    setTranscript((prev) => [...prev, { role, text, timestamp: Date.now() }]);
+    setTranscript((prev) => {
+      const last = prev[prev.length - 1];
+      // Append to last entry if same role and turn is still ongoing
+      if (last && last.role === role && !turnCompleteRef.current) {
+        return [...prev.slice(0, -1), { ...last, text: last.text + text }];
+      }
+      turnCompleteRef.current = false;
+      return [...prev, { role, text, timestamp: Date.now() }];
+    });
   }, []);
 
   const trackAmplitude = useCallback(() => {
@@ -369,6 +378,10 @@ export default function VoiceAgentProvider({ locale, children }: VoiceAgentProvi
           if (msg.serverContent.outputTranscription?.text) {
             addTranscript('agent', msg.serverContent.outputTranscription.text);
           }
+          // Turn complete — next output starts a new transcript entry
+          if (msg.serverContent.turnComplete || msg.serverContent.generationComplete) {
+            turnCompleteRef.current = true;
+          }
         }
 
         // Tool call