diff --git a/lib/core/services/api_service.dart b/lib/core/services/api_service.dart
index a1560cf..258f964 100644
--- a/lib/core/services/api_service.dart
+++ b/lib/core/services/api_service.dart
@@ -1130,7 +1130,7 @@ class ApiService {
         if (msg.role == 'assistant' && msg.model != null)
           'modelName': msg.model,
         if (msg.role == 'assistant') 'modelIdx': 0,
-        if (msg.role == 'assistant') 'done': true,
+        if (msg.role == 'assistant') 'done': !msg.isStreaming,
         if (msg.role == 'user' && model != null) 'models': [model],
         if (combinedFilesMap.isNotEmpty) 'files': combinedFilesMap,
       };
@@ -1166,7 +1166,7 @@ class ApiService {
         if (msg.role == 'assistant' && msg.model != null)
           'modelName': msg.model,
         if (msg.role == 'assistant') 'modelIdx': 0,
-        if (msg.role == 'assistant') 'done': true,
+        if (msg.role == 'assistant') 'done': !msg.isStreaming,
         if (msg.role == 'user' && model != null) 'models': [model],
         if (combinedFilesArray.isNotEmpty) 'files': combinedFilesArray,
       });
@@ -2652,11 +2652,14 @@ class ApiService {
     String? sessionIdOverride,
     List<Map<String, dynamic>>? toolServers,
     Map<String, dynamic>? backgroundTasks,
+    String? responseMessageId,
   }) {
     final streamController = StreamController<String>();
 
     // Generate unique IDs
-    final messageId = const Uuid().v4();
+    final messageId = (responseMessageId != null && responseMessageId.isNotEmpty)
+        ? responseMessageId
+        : const Uuid().v4();
     final sessionId =
         (sessionIdOverride != null && sessionIdOverride.isNotEmpty)
         ? sessionIdOverride
@@ -2809,6 +2812,8 @@ class ApiService {
     // Always use task-based background flow for unified pipeline.
     // When a dynamic channel (session_id) is not provided, this method falls
     // back to polling and streams deltas to the UI.
+    // Always use background task flow (matches web client) to ensure
+    // server maintains correct history with pre-seeded assistant id.
     final bool useBackgroundTasks = true;
 
     // Use background flow only when required; otherwise prefer SSE even with chat_id.
diff --git a/lib/core/utils/tool_calls_parser.dart b/lib/core/utils/tool_calls_parser.dart
index d6e7bee..1d80012 100644
--- a/lib/core/utils/tool_calls_parser.dart
+++ b/lib/core/utils/tool_calls_parser.dart
@@ -236,6 +236,71 @@ class ToolCallsParser {
       return raw.length > max ? '${raw.substring(0, max)}…' : raw;
     }
   }
+
+  /// Sanitize assistant/user content before sending to the API, mirroring
+  /// the web client's `processDetails` behavior:
+  /// - Remove <details type="reasoning"> and <details type="code_interpreter"> blocks
+  /// - Replace <details type="tool_calls" ...>...</details> blocks with the
+  ///   JSON-serialized `result` attribute (as a quoted string) when available;
+  ///   otherwise replace with an empty string.
+  static String sanitizeForApi(String content) {
+    if (content.isEmpty) return content;
+
+    // Remove blocks we never want to include in conversation context
+    final removeTypes = ['reasoning', 'code_interpreter'];
+    for (final t in removeTypes) {
+      content = content.replaceAll(
+        RegExp(
+          '<details\\s+type=\"${t}\"[^>]*>[\\s\\S]*?<\\/details>',
+          multiLine: true,
+          dotAll: true,
+        ),
+        '',
+      );
+    }
+
+    if (!content.contains('<details')) return content.trim();
+
+    // Replace tool_calls blocks in-order with their results
+    final segs = segments(content);
+    if (segs == null || segs.isEmpty) return content.trim();
+
+    final buf = StringBuffer();
+    for (final seg in segs) {
+      if (seg.isToolCall && seg.entry != null) {
+        final entry = seg.entry!;
+        dynamic res = entry.result;
+        String out;
+        if (res == null) {
+          out = '';
+        } else {
+          try {
+            out = json.encode(res);
+          } catch (_) {
+            out = res.toString();
+          }
+        }
+        // Match web behavior: wrap in quotes so it's clearly a string payload
+        if (out.isNotEmpty && !(out.startsWith('"') && out.endsWith('"'))) {
+          out = '"$out"';
+        }
+        buf.write(out);
+      } else {
+        // Keep the raw text, but also remove any stray non-tool_calls details blocks
+        final t = (seg.text ?? '').replaceAll(
+          RegExp(
+            r'<details(?!\s+type=\"tool_calls\")[^>]*>[\s\S]*?<\/details>',
+            multiLine: true,
+            dotAll: true,
+          ),
+          '',
+        );
+        if (t.isNotEmpty) buf.write(t);
+      }
+    }
+
+    return buf.toString().trim();
+  }
 }
 
 /// Ordered piece of content: either plain text or a tool-call entry
diff --git a/lib/features/chat/providers/chat_providers.dart b/lib/features/chat/providers/chat_providers.dart
index d78f1d8..16a721d 100644
--- a/lib/features/chat/providers/chat_providers.dart
+++ b/lib/features/chat/providers/chat_providers.dart
@@ -450,37 +450,31 @@ Future<void> regenerateMessage(
 
     for (final msg in messages) {
       if (msg.role.isNotEmpty && msg.content.isNotEmpty && !msg.isStreaming) {
+        // Clean up tool/details markup to match web client behavior
+        final cleaned = ToolCallsParser.sanitizeForApi(msg.content);
+
         // Handle messages with attachments
         if (msg.attachmentIds != null && msg.attachmentIds!.isNotEmpty) {
           final List<Map<String, dynamic>> contentArray = [];
 
           // Add text content first
-          if (msg.content.isNotEmpty) {
-            contentArray.add({'type': 'text', 'text': msg.content});
+          if (cleaned.isNotEmpty) {
+            contentArray.add({'type': 'text', 'text': cleaned});
           }
 
           conversationMessages.add({
             'role': msg.role,
-            'content': contentArray.isNotEmpty ? contentArray : msg.content,
+            'content': contentArray.isNotEmpty ? contentArray : cleaned,
           });
         } else {
           // Regular text message
-          conversationMessages.add({'role': msg.role, 'content': msg.content});
+          conversationMessages.add({'role': msg.role, 'content': cleaned});
         }
       }
     }
 
-    // Stream response using SSE
-    final response = api!.sendMessage(
-      messages: conversationMessages,
-      model: selectedModel.id,
-      conversationId: activeConversation.id,
-    );
-
-    final stream = response.stream;
-    final assistantMessageId = response.messageId;
-
-    // Add assistant message placeholder
+    // Pre-seed assistant skeleton
+    final String assistantMessageId = const Uuid().v4();
     final assistantMessage = ChatMessage(
       id: assistantMessageId,
       role: 'assistant',
@@ -490,6 +484,24 @@ Future<void> regenerateMessage(
       isStreaming: true,
     );
     ref.read(chatMessagesProvider.notifier).addMessage(assistantMessage);
+    try {
+      final msgsForSeed = ref.read(chatMessagesProvider);
+      await api!.updateConversationWithMessages(
+        activeConversation.id,
+        msgsForSeed,
+        model: selectedModel.id,
+      );
+    } catch (_) {}
+
+    // Stream response via background task (socket/dynamic channel or polling)
+    final response = api!.sendMessage(
+      messages: conversationMessages,
+      model: selectedModel.id,
+      conversationId: activeConversation.id,
+      responseMessageId: assistantMessageId,
+    );
+
+    final stream = response.stream;
 
     // Handle streaming response (basic chunking for this path)
     final chunkedStream = StreamChunker.chunkStream(
@@ -705,6 +717,9 @@ Future<void> _sendMessageInternal(
     // Skip only empty assistant message placeholders that are currently streaming
     // Include completed messages (both user and assistant) for conversation history
     if (msg.role.isNotEmpty && msg.content.isNotEmpty && !msg.isStreaming) {
+      // Prepare cleaned text content (strip tool details etc.)
+      final cleaned = ToolCallsParser.sanitizeForApi(msg.content);
+
       // Check if message has attachments (images and non-images)
       if (msg.attachmentIds != null && msg.attachmentIds!.isNotEmpty) {
         // All models use the same content array format (OpenWebUI standard)
@@ -715,8 +730,8 @@ Future<void> _sendMessageInternal(
         final List<Map<String, dynamic>> nonImageFiles = [];
 
         // Add text content first
-        if (msg.content.isNotEmpty) {
-          contentArray.add({'type': 'text', 'text': msg.content});
+        if (cleaned.isNotEmpty) {
+          contentArray.add({'type': 'text', 'text': cleaned});
         }
 
         // Add image attachments with proper MIME type handling; collect non-image attachments
@@ -772,7 +787,7 @@ Future<void> _sendMessageInternal(
         conversationMessages.add(messageMap);
       } else {
         // Regular text-only message
-        conversationMessages.add({'role': msg.role, 'content': msg.content});
+        conversationMessages.add({'role': msg.role, 'content': cleaned});
       }
     }
   }
@@ -789,6 +804,33 @@ Future<void> _sendMessageInternal(
       : null;
 
   try {
+    // Pre-seed assistant skeleton on server to ensure correct chain
+    // Generate assistant message id now (must be consistent across client/server)
+    final String assistantMessageId = const Uuid().v4();
+
+    // Add assistant placeholder locally before sending
+    final assistantPlaceholder = ChatMessage(
+      id: assistantMessageId,
+      role: 'assistant',
+      content: '',
+      timestamp: DateTime.now(),
+      model: selectedModel.id,
+      isStreaming: true,
+    );
+    ref.read(chatMessagesProvider.notifier).addMessage(assistantPlaceholder);
+
+    // Persist skeleton chain to server so web can load correct history
+    try {
+      final activeConvForSeed = ref.read(activeConversationProvider);
+      if (activeConvForSeed != null) {
+        final msgsForSeed = ref.read(chatMessagesProvider);
+        await api.updateConversationWithMessages(
+          activeConvForSeed.id,
+          msgsForSeed,
+          model: selectedModel.id,
+        );
+      }
+    } catch (_) {}
     // Use the model's actual supported parameters if available
     final supportedParams =
         selectedModel.supportedParameters ??
@@ -1028,11 +1070,12 @@ Future<void> _sendMessageInternal(
           (conv.title == 'New Chat' && msgs.length <= 1);
     } catch (_) {}
 
+    // Match web client: request background follow-ups always; title/tags on first turn
     final bgTasks = <String, dynamic>{
       if (shouldGenerateTitle) 'title_generation': true,
       if (shouldGenerateTitle) 'tags_generation': true,
       'follow_up_generation': true,
-      if (webSearchEnabled) 'web_search': true, // enable bg workflow for web search
+      if (webSearchEnabled) 'web_search': true, // enable bg web search
       if (imageGenerationEnabled) 'image_generation': true, // enable bg image flow
     };
 
@@ -1056,23 +1099,12 @@ Future<void> _sendMessageInternal(
       sessionIdOverride: wantSessionBinding ? socketSessionId : null,
       toolServers: toolServers,
       backgroundTasks: bgTasks,
+      responseMessageId: assistantMessageId,
     );
 
     final stream = response.stream;
-    final assistantMessageId = response.messageId;
     final sessionId = response.sessionId;
 
-    // Add assistant message placeholder with the generated ID and immediate typing indicator
-    final assistantMessage = ChatMessage(
-      id: assistantMessageId,
-      role: 'assistant',
-      content: '',
-      timestamp: DateTime.now(),
-      model: selectedModel.id,
-      isStreaming: true,
-    );
-    ref.read(chatMessagesProvider.notifier).addMessage(assistantMessage);
-
     // If socket is available, start listening for chat-events immediately
     // Background-tools flow OR any session-bound flow relies on socket/dynamic channel for
     // streaming content. Allow socket TEXT in those modes. For pure SSE/polling flows, suppress
@@ -1718,10 +1750,10 @@ Future<void> _sendMessageInternal(
           }
         }
 
-        // Save conversation to OpenWebUI server only after streaming is complete
-        // Add a small delay to ensure the last message content is fully updated
-        await Future.delayed(const Duration(milliseconds: 100));
-        await _saveConversationToServer(ref);
+        // Do not persist conversation to server here. Server manages chat state.
+        // Keep local save only for quick resume.
+        await Future.delayed(const Duration(milliseconds: 50));
+        await _saveConversationLocally(ref);
 
         // Removed post-assistant image generation; images are handled immediately after user message
       },
@@ -1956,17 +1988,7 @@ Future<void> _checkForTitleInBackground(
 }
 
 // Save current conversation to OpenWebUI server
-Future<void> _saveConversationToServer(dynamic ref) async {
-  // Enqueue save task; local fallback remains if queue fails
-  try {
-    final activeConversation = ref.read(activeConversationProvider);
-    await ref
-        .read(taskQueueProvider.notifier)
-        .enqueueSaveConversation(conversationId: activeConversation?.id);
-  } catch (_) {
-    await _saveConversationLocally(ref);
-  }
-}
+// Removed server persistence; only local caching is used in mobile app.
 
 // Fallback: Save current conversation to local storage
 Future<void> _saveConversationLocally(dynamic ref) async {
diff --git a/lib/features/chat/views/chat_page.dart b/lib/features/chat/views/chat_page.dart
index c66d564..6edb8e6 100644
--- a/lib/features/chat/views/chat_page.dart
+++ b/lib/features/chat/views/chat_page.dart
@@ -1030,7 +1030,9 @@ class _ChatPageState extends ConsumerState<ChatPage> {
             ref.read(chatMessagesProvider.notifier).finishStreaming();
           }
 
-          await _saveConversationBeforeLeaving(ref);
+          // Do not push conversation state back to server on exit.
+          // Server already maintains chat state from message sends.
+          // Keep any local persistence only.
 
           if (context.mounted) {
             final canPopNavigator = Navigator.of(context).canPop();
diff --git a/lib/shared/services/tasks/outbound_task.dart b/lib/shared/services/tasks/outbound_task.dart
index 0bab8c8..cf77ffe 100644
--- a/lib/shared/services/tasks/outbound_task.dart
+++ b/lib/shared/services/tasks/outbound_task.dart
@@ -74,17 +74,6 @@ abstract class OutboundTask with _$OutboundTask {
     String? error,
   }) = GenerateImageTask;
 
-  const factory OutboundTask.saveConversation({
-    required String id,
-    String? conversationId,
-    @Default(TaskStatus.queued) TaskStatus status,
-    @Default(0) int attempt,
-    String? idempotencyKey,
-    DateTime? enqueuedAt,
-    DateTime? startedAt,
-    DateTime? completedAt,
-    String? error,
-  }) = SaveConversationTask;
 
   const factory OutboundTask.generateTitle({
     required String id,
@@ -121,7 +110,6 @@ abstract class OutboundTask with _$OutboundTask {
         uploadMedia: (t) => t.conversationId,
         executeToolCall: (t) => t.conversationId,
         generateImage: (t) => t.conversationId,
-        saveConversation: (t) => t.conversationId,
         generateTitle: (t) => t.conversationId,
         imageToDataUrl: (t) => t.conversationId,
       );
diff --git a/lib/shared/services/tasks/task_queue.dart b/lib/shared/services/tasks/task_queue.dart
index f1595f9..dda42fd 100644
--- a/lib/shared/services/tasks/task_queue.dart
+++ b/lib/shared/services/tasks/task_queue.dart
@@ -273,22 +273,7 @@ class TaskQueueNotifier extends StateNotifier<List<OutboundTask>> {
     return id;
   }
 
-  Future<String> enqueueSaveConversation({
-    required String? conversationId,
-    String? idempotencyKey,
-  }) async {
-    final id = _uuid.v4();
-    final task = OutboundTask.saveConversation(
-      id: id,
-      conversationId: conversationId,
-      idempotencyKey: idempotencyKey,
-      enqueuedAt: DateTime.now(),
-    );
-    state = [...state, task];
-    await _save();
-    _process();
-    return id;
-  }
+  // Removed: enqueueSaveConversation — mobile app no longer persists chats to server.
 
   Future<String> enqueueGenerateTitle({
     required String conversationId,
diff --git a/lib/shared/services/tasks/task_worker.dart b/lib/shared/services/tasks/task_worker.dart
index f6087bd..866dd82 100644
--- a/lib/shared/services/tasks/task_worker.dart
+++ b/lib/shared/services/tasks/task_worker.dart
@@ -22,7 +22,7 @@ class TaskWorker {
       executeToolCall: _performExecuteToolCall,
       generateImage: _performGenerateImage,
       imageToDataUrl: _performImageToDataUrl,
-      saveConversation: _performSaveConversation,
+      // saveConversation removed — we no longer push chat state to server
       generateTitle: _performGenerateTitle,
     );
   }
@@ -328,36 +328,7 @@ class TaskWorker {
     }
   }
 
-  Future<void> _performSaveConversation(SaveConversationTask task) async {
-    final api = _ref.read(apiServiceProvider);
-    final messages = _ref.read(chat.chatMessagesProvider);
-    final activeConv = _ref.read(activeConversationProvider);
-    final selectedModel = _ref.read(selectedModelProvider);
-    if (api == null || messages.isEmpty || activeConv == null) return;
-
-    // Skip if last assistant is empty placeholder
-    final last = messages.last;
-    if (last.role == 'assistant' &&
-        last.content.trim().isEmpty &&
-        (last.files?.isEmpty ?? true) &&
-        (last.attachmentIds?.isEmpty ?? true)) {
-      return;
-    }
-
-    try {
-      await api.updateConversationWithMessages(
-        activeConv.id,
-        messages,
-        model: selectedModel?.id,
-      );
-      final updated = activeConv.copyWith(
-        messages: messages,
-        updatedAt: DateTime.now(),
-      );
-      _ref.read(activeConversationProvider.notifier).state = updated;
-      _ref.invalidate(conversationsProvider);
-    } catch (_) {}
-  }
+  // _performSaveConversation removed
 
   Future<void> _performGenerateTitle(GenerateTitleTask task) async {
     final api = _ref.read(apiServiceProvider);
@@ -387,15 +358,8 @@ class TaskWorker {
             updatedAt: DateTime.now(),
           );
           _ref.read(activeConversationProvider.notifier).state = updated;
-          try {
-            final cur = _ref.read(chat.chatMessagesProvider);
-            await api.updateConversationWithMessages(
-              updated.id,
-              cur,
-              title: updated.title,
-              model: selectedModel.id,
-            );
-          } catch (_) {}
+          // Do not push full messages to server; skip remote update.
+          // Optionally refresh list to reflect server-side title when it’s generated there.
           _ref.invalidate(conversationsProvider);
         }
       }