fix: render multimodal session content safely

lib/core/screens/chat_screen.dart

@@ -10,6 +10,7 @@ import 'package:speech_to_text/speech_recognition_result.dart';
 import 'package:speech_to_text/speech_to_text.dart';
 
 import '../services/connection_manager.dart';
+import '../utils/message_content.dart';
 import '../utils/responsive.dart';
 
 class ChatScreen extends StatefulWidget {
@@ -295,7 +296,7 @@ class _ChatScreenState extends State<ChatScreen> {
           (msg['toolCallName'] as String?) ??
           '';
       final toolCallId = (msg['tool_call_id'] as String?) ?? '';
-      final content = (msg['content'] as String?) ?? '';
+      final content = messageContentToText(msg['content']);
 
       String toolName = name.isNotEmpty ? name : '';
       if (toolName.isEmpty && content.isNotEmpty) {
@@ -687,7 +688,7 @@ class _ChatScreenState extends State<ChatScreen> {
         continue;
       }
       if (role != 'user' && role != 'assistant') continue;
-      final content = (msg['content'] as String?) ?? '';
+      final content = messageContentToText(msg['content']);
       if (content.isEmpty) continue;
 
       if (currentGroup.isNotEmpty) {
@@ -719,7 +720,7 @@ class _ChatScreenState extends State<ChatScreen> {
 
         final msg = item as Map<String, dynamic>;
         final role = (msg['role'] as String?) ?? 'assistant';
-        final content = (msg['content'] as String?) ?? '';
+        final content = messageContentToText(msg['content']);
         final isUser = role == 'user';
 
         return _MessageBubble(

lib/core/utils/message_content.dart

@@ -0,0 +1,37 @@
+/// Convert Hermes/OpenAI message content into displayable text.
+///
+/// Legacy messages carry a String. Multimodal messages carry a list of typed
+/// parts such as `{type: text, text: ...}` and `{type: image_url, ...}`.
+/// Keep rendering resilient when the gateway adds new part types.
+String messageContentToText(dynamic content) {
+  if (content == null) return '';
+  if (content is String) return content;
+
+  if (content is List) {
+    return content
+        .map(_contentPartToText)
+        .where((part) => part.isNotEmpty)
+        .join('\n\n');
+  }
+
+  return _contentPartToText(content);
+}
+
+String _contentPartToText(dynamic part) {
+  if (part == null) return '';
+  if (part is String) return part;
+  if (part is! Map) return part.toString();
+
+  final text = part['text'];
+  if (text is String && text.isNotEmpty) return text;
+
+  final type = part['type']?.toString() ?? 'unknown';
+  if (type.contains('image') || part.containsKey('image_url')) {
+    return '[Image]';
+  }
+  if (type.contains('file') || part.containsKey('file')) {
+    return '[File]';
+  }
+
+  return '[Unsupported content: $type]';
+}

test/message_content_test.dart

@@ -0,0 +1,43 @@
+import 'package:flutter_test/flutter_test.dart';
+import 'package:hermes_android/core/utils/message_content.dart';
+
+void main() {
+  group('messageContentToText', () {
+    test('keeps legacy string content unchanged', () {
+      expect(messageContentToText('hello'), 'hello');
+    });
+
+    test('joins OpenAI text content parts', () {
+      expect(
+        messageContentToText([
+          {'type': 'text', 'text': 'first'},
+          {'type': 'text', 'text': 'second'},
+        ]),
+        'first\n\nsecond',
+      );
+    });
+
+    test('renders mixed text and image content without throwing', () {
+      expect(
+        messageContentToText([
+          {'type': 'text', 'text': 'caption'},
+          {
+            'type': 'image_url',
+            'image_url': {'url': 'https://example.invalid/image.png'},
+          },
+        ]),
+        'caption\n\n[Image]',
+      );
+    });
+
+    test('handles null and unknown structured content safely', () {
+      expect(messageContentToText(null), '');
+      expect(
+        messageContentToText([
+          {'type': 'custom_part', 'payload': 42},
+        ]),
+        '[Unsupported content: custom_part]',
+      );
+    });
+  });
+}