feat(react): wire user-llm-text into usePipecatConversation

client-js/client/client.ts

@@ -38,6 +38,7 @@ import {
   TranscriptData,
   TransportState,
   UICancelJobGroupData,
+  UserLLMTextData,
   UICommandData,
   UIEventData,
   UIJobGroupData,
@@ -149,6 +150,7 @@ export type RTVIEventCallbacks = Partial<{
   onUserMuteStarted: () => void;
   onUserMuteStopped: () => void;
   onUserTranscript: (data: TranscriptData) => void;
+  onUserLlmText: (data: UserLLMTextData) => void;
   onBotOutput: (data: BotOutputData) => void;
   /** @deprecated Use onBotOutput instead */
   onBotTranscript: (data: BotLLMTextData) => void;
@@ -1144,6 +1146,12 @@ export class PipecatClient extends RTVIEventEmitter {
         this._options.callbacks?.onUserTranscript?.(TranscriptData);
         break;
       }
+      case RTVIMessageType.USER_LLM_TEXT: {
+        const llmTextData = ev.data as UserLLMTextData;
+        this._options.callbacks?.onUserLlmText?.(llmTextData);
+        this.emit(RTVIEvent.UserLlmText, llmTextData);
+        break;
+      }
       case RTVIMessageType.BOT_OUTPUT: {
         this._options.callbacks?.onBotOutput?.(ev.data as BotOutputData);
         break;

client-js/rtvi/events.ts

@@ -21,6 +21,7 @@ import {
   TranscriptData,
   UICommandData,
   UIJobGroupData,
+  UserLLMTextData,
 } from "./messages";
 
 export enum RTVIEvent {
@@ -65,6 +66,7 @@ export enum RTVIEvent {
   BotTranscript = "botTranscript",
 
   // llm events
+  UserLlmText = "userLlmText",
   BotLlmText = "botLlmText",
   BotLlmStarted = "botLlmStarted",
   BotLlmStopped = "botLlmStopped",
@@ -150,6 +152,7 @@ export type RTVIEvents = Partial<{
   botTranscript: (data: BotLLMTextData) => void;
 
   // llm events
+  userLlmText: (data: UserLLMTextData) => void;
   botLlmText: (data: BotLLMTextData) => void;
   botLlmStarted: () => void;
   botLlmStopped: () => void;

client-js/rtvi/messages.ts

@@ -153,6 +153,10 @@ export type BotLLMTextData = {
   text: string;
 };
 
+export type UserLLMTextData = {
+  text: string;
+};
+
 export type BotTTSTextData = {
   text: string;
 };

client-react/src/conversation/conversationActions.ts

@@ -130,6 +130,7 @@ export const mergeMessages = (
       lastMerged.role === currentMessage.role &&
       currentMessage.role !== "system" &&
       currentMessage.role !== "function_call" &&
+      !lastMerged.final &&
       timeDiff < MERGE_WINDOW_MS;
 
     if (shouldMerge) {

client-react/src/conversation/useConversationEventWiring.ts

@@ -309,6 +309,29 @@ export function useConversationEventWiring() {
     )
   );
 
+  // user-llm-text fires when the user's input reaches the LLM — both in voice
+  // mode (after STT) and in chat mode (typed text, no STT). In voice mode
+  // user-transcription has already created an in-progress user message, so we
+  // only need to finalize it. In chat mode no user message exists yet, so we
+  // create one from the LLM text first, then finalize.
+  useRTVIClientEvent(
+    RTVIEvent.UserLlmText,
+    useAtomCallback(
+      useCallback((get, set, data) => {
+        const messages = get(messagesAtom);
+        const lastUserIdx = findLastIndex(
+          messages,
+          (m: ConversationMessage) => m.role === "user"
+        );
+
+        if (lastUserIdx === -1 || messages[lastUserIdx].final) {
+          upsertUserTranscript(get, set, data.text ?? "", true);
+        }
+        finalizeLastMessage(get, set, "user");
+      }, [])
+    )
+  );
+
   useRTVIClientEvent(
     RTVIEvent.UserStoppedSpeaking,
     useAtomCallback(