Add Expressive Mode support for ElevenLabs V3 Conversational TTS

- Add bExpressiveMode toggle and editable ExpressiveModePromptFragment
  with audio tag instructions ([laughs], [whispers], [sighs], [slow], [excited])
- BuildAgentPayload: append prompt fragment, set expressive_mode API field
  on agent config, auto-override TTS model to eleven_v3_conversational
- OnFetchAgent: strip expressive fragment from prompt (exact + marker fallback),
  read expressive_mode bool from API, auto-detect V3 model
- TTS model combo: inject asset's current model if absent from /v1/models list
  (covers agent-only models like eleven_v3_conversational)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Unreal/PS_AI_Agent/Plugins/PS_AI_ConvAgent/Source/PS_AI_ConvAgent/Public/PS_AI_ConvAgent_AgentConfig_ElevenLabs.h

@@ -237,6 +237,35 @@ public:
 		"- \"high\" for strong reactions (big laugh, deep sadness, shock)\n\n"
 		"Always return to neutral when the emotional moment passes.");
 
+	// ── Expressive Mode (V3 Conversational) ─────────────────────────────────
+
+	/** Enable Expressive Mode (requires TTS model "eleven_v3_conversational").
+	 *  When enabled, the agent uses emotionally intelligent speech with
+	 *  natural intonation and expressive audio tags like [laughs], [whispers].
+	 *  A prompt fragment is appended to instruct the LLM to use these tags. */
+	UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "Expressive Mode",
+		meta = (ToolTip = "Enable Expressive Mode.\nRequires V3 Conversational TTS model.\nAdds expressive audio tag instructions to the prompt."))
+	bool bExpressiveMode = false;
+
+	/** System prompt fragment appended when bExpressiveMode is true.
+	 *  Instructs the LLM to use inline audio tags for emotional delivery.
+	 *  Editable for customization — stripped automatically on Fetch. */
+	UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "Expressive Mode",
+		meta = (MultiLine = "true", EditCondition = "bExpressiveMode",
+		ToolTip = "Prompt instructions for expressive audio tags.\nAppended to CharacterPrompt when creating/updating the agent."))
+	FString ExpressiveModePromptFragment = TEXT(
+		"## Expressive Speech\n"
+		"You can use expressive audio tags in your responses for precise vocal control. "
+		"Each tag affects approximately the next 4-5 words before returning to normal delivery.\n\n"
+		"Available tags:\n"
+		"- [laughs] for moments of humor or amusement\n"
+		"- [whispers] for confidential or intimate moments\n"
+		"- [sighs] for resignation or relief\n"
+		"- [slow] when emphasizing important information\n"
+		"- [excited] for enthusiastic or energetic delivery\n\n"
+		"Example: \"That's great to hear! [laughs] I'm glad we could sort that out for you.\"\n\n"
+		"Use these tags naturally and sparingly to enhance expressiveness without overusing them.");
+
 	// ── Action Tool ─────────────────────────────────────────────────────────
 
 	/** Include a configurable "perform_action" client tool in the agent configuration.

Unreal/PS_AI_Agent/Plugins/PS_AI_ConvAgent/Source/PS_AI_ConvAgentEditor/Private/PS_AI_ConvAgent_AgentConfigCustomization_ElevenLabs.cpp

@@ -615,6 +615,18 @@ void FPS_AI_ConvAgent_AgentConfigCustomization_ElevenLabs::OnFetchModelsClicked(
 			if (UPS_AI_ConvAgent_AgentConfig_ElevenLabs* Asset = Pinned->GetEditedAsset())
 			{
 				int32 Idx = Pinned->ModelIDs.IndexOfByKey(Asset->TTSModelID);
+
+				// Agent-only models (e.g. eleven_v3_conversational) may not appear
+				// in the general /v1/models list.  Inject the asset's current model
+				// so the combo always reflects the actual value.
+				if (Idx == INDEX_NONE && !Asset->TTSModelID.IsEmpty())
+				{
+					FString DisplayStr = FString::Printf(TEXT("%s"), *Asset->TTSModelID);
+					Pinned->ModelDisplayNames.Add(MakeShareable(new FString(DisplayStr)));
+					Pinned->ModelIDs.Add(Asset->TTSModelID);
+					Idx = Pinned->ModelIDs.Num() - 1;
+				}
+
 				if (Idx != INDEX_NONE && Pinned->ModelComboBox.IsValid())
 				{
 					Pinned->ModelComboBox->SetSelectedItem(Pinned->ModelDisplayNames[Idx]);
@@ -1318,6 +1330,33 @@ void FPS_AI_ConvAgent_AgentConfigCustomization_ElevenLabs::OnFetchAgentClicked()
 								}
 							}
 
+							// 5. Expressive Mode fragment
+							{
+								const FString& Fragment = Asset->ExpressiveModePromptFragment;
+								if (!Fragment.IsEmpty())
+								{
+									int32 Idx = Prompt.Find(Fragment, ESearchCase::CaseSensitive);
+									if (Idx != INDEX_NONE)
+									{
+										Prompt.LeftInline(Idx);
+										Prompt.TrimEndInline();
+										Asset->bExpressiveMode = true;
+									}
+								}
+								// Fallback: marker-based stripping
+								if (!Asset->bExpressiveMode)
+								{
+									const FString Marker = TEXT("## Expressive Speech");
+									int32 Idx = Prompt.Find(Marker, ESearchCase::CaseSensitive);
+									if (Idx != INDEX_NONE)
+									{
+										Prompt.LeftInline(Idx);
+										Prompt.TrimEndInline();
+										Asset->bExpressiveMode = true;
+									}
+								}
+							}
+
 							Asset->CharacterPrompt = Prompt;
 						}
 
@@ -1348,6 +1387,13 @@ void FPS_AI_ConvAgent_AgentConfigCustomization_ElevenLabs::OnFetchAgentClicked()
 					{
 						Asset->MaxTurns = MaxTurns;
 					}
+
+					// expressive_mode (V3 Conversational)
+					bool bExpressive = false;
+					if ((*AgentObj)->TryGetBoolField(TEXT("expressive_mode"), bExpressive))
+					{
+						Asset->bExpressiveMode = bExpressive;
+					}
 				}
 
 				// conversation_config.tts
@@ -1364,6 +1410,12 @@ void FPS_AI_ConvAgent_AgentConfigCustomization_ElevenLabs::OnFetchAgentClicked()
 					if ((*TTSObj)->TryGetStringField(TEXT("model_id"), ModelID))
 					{
 						Asset->TTSModelID = ModelID;
+
+						// Auto-detect Expressive Mode from V3 Conversational model
+						if (ModelID == TEXT("eleven_v3_conversational"))
+						{
+							Asset->bExpressiveMode = true;
+						}
 					}
 
 					double Stability = 0.5;
@@ -1563,14 +1615,15 @@ TSharedPtr<FJsonObject> FPS_AI_ConvAgent_AgentConfigCustomization_ElevenLabs::Bu
 
 	UE_LOG(LogPS_AI_AgentConfigEditor, Log,
 		TEXT("BuildAgentPayload: CharacterPrompt=%d chars, bMultilingual=%d, bAutoLangInstr=%d, Language='%s', "
-			"LangFragment=%d chars, MultiFragment=%d chars, bEmotionTool=%d"),
+			"LangFragment=%d chars, MultiFragment=%d chars, bEmotionTool=%d, bExpressiveMode=%d"),
 		Asset->CharacterPrompt.Len(),
 		Asset->bMultilingual,
 		Asset->bAutoLanguageInstruction,
 		*Asset->Language,
 		Asset->LanguagePromptFragment.Len(),
 		Asset->MultilingualPromptFragment.Len(),
-		Asset->bIncludeEmotionTool);
+		Asset->bIncludeEmotionTool,
+		Asset->bExpressiveMode);
 
 	// Language handling: multilingual mode vs fixed-language mode.
 	// The ElevenLabs "language" field only controls STT/TTS — the LLM defaults to
@@ -1618,6 +1671,14 @@ TSharedPtr<FJsonObject> FPS_AI_ConvAgent_AgentConfigCustomization_ElevenLabs::Bu
 		UE_LOG(LogPS_AI_AgentConfigEditor, Log, TEXT("  → Appended ActionToolPromptFragment from ActionSet"));
 	}
 
+	// Append expressive mode instructions (V3 Conversational audio tags).
+	if (Asset->bExpressiveMode && !Asset->ExpressiveModePromptFragment.IsEmpty())
+	{
+		FullPrompt += TEXT("\n\n");
+		FullPrompt += Asset->ExpressiveModePromptFragment;
+		UE_LOG(LogPS_AI_AgentConfigEditor, Log, TEXT("  → Appended ExpressiveModePromptFragment"));
+	}
+
 	UE_LOG(LogPS_AI_AgentConfigEditor, Log, TEXT("BuildAgentPayload: FullPrompt = %d chars"), FullPrompt.Len());
 
 	// prompt object (includes LLM selection + tools)
@@ -1664,6 +1725,10 @@ TSharedPtr<FJsonObject> FPS_AI_ConvAgent_AgentConfigCustomization_ElevenLabs::Bu
 	{
 		AgentObj->SetNumberField(TEXT("max_tokens"), Asset->MaxTurns);
 	}
+	if (Asset->bExpressiveMode)
+	{
+		AgentObj->SetBoolField(TEXT("expressive_mode"), true);
+	}
 
 	// tts
 	TSharedPtr<FJsonObject> TTSObj = MakeShareable(new FJsonObject());
@@ -1678,14 +1743,27 @@ TSharedPtr<FJsonObject> FPS_AI_ConvAgent_AgentConfigCustomization_ElevenLabs::Bu
 	// Monolingual models (e.g. eleven_monolingual_v1) only support English.
 	FString ResolvedModelID = Asset->TTSModelID;
 
+	// Expressive mode requires V3 Conversational — override if needed.
+	if (Asset->bExpressiveMode)
+	{
+		if (ResolvedModelID != TEXT("eleven_v3_conversational"))
+		{
+			UE_LOG(LogPS_AI_AgentConfigEditor, Warning,
+				TEXT("Expressive mode: overriding TTS model '%s' → eleven_v3_conversational (required for audio tags)."),
+				*ResolvedModelID);
+			ResolvedModelID = TEXT("eleven_v3_conversational");
+		}
+	}
+
 	auto IsMultilingualModel = [](const FString& ModelID) -> bool
 	{
 		return ModelID.Contains(TEXT("multilingual"))
 			|| ModelID.Contains(TEXT("turbo"))
-			|| ModelID.Contains(TEXT("flash"));
+			|| ModelID.Contains(TEXT("flash"))
+			|| ModelID.Contains(TEXT("v3"));  // V3 models are multilingual
 	};
 
-	if (Asset->bMultilingual)
+	if (!Asset->bExpressiveMode && Asset->bMultilingual)
 	{
 		// Multilingual mode: force a multilingual TTS model.
 		if (ResolvedModelID.IsEmpty() || !IsMultilingualModel(ResolvedModelID))