fix message editing

backend/open_webui/main.py

@@ -1784,6 +1784,7 @@ async def chat_completion(
             "user_id": user.id,
             "chat_id": form_data.pop("chat_id", None),
             "message_id": form_data.pop("id", None),
+            "context_message_id": form_data.pop("context_message_id", None),
             "parent_message": form_data.pop("parent_message", None),
             "parent_message_id": form_data.pop("parent_id", None),
             "session_id": form_data.pop("session_id", None),

backend/open_webui/utils/middleware.py

@@ -97,6 +97,7 @@ from open_webui.utils.misc import (
     convert_logit_bias_input_to_json,
     get_content_from_message,
     convert_output_to_messages,
+    merge_assistant_content_into_output_messages,
 )
 from open_webui.utils.tools import (
     get_tools,
@@ -2131,7 +2132,11 @@ def process_messages_with_output(messages: list[dict]) -> list[dict]:
             # Use output items for clean OpenAI-format messages
             output_messages = convert_output_to_messages(message["output"], raw=True)
             if output_messages:
-                processed.extend(output_messages)
+                processed.extend(
+                    merge_assistant_content_into_output_messages(
+                        output_messages, message.get("content")
+                    )
+                )
                 continue
 
         # Strip 'output' field before adding (LLM shouldn't see it)
@@ -2152,10 +2157,12 @@ async def process_chat_payload(request, form_data, user, metadata, model):
     # Load messages from DB when available — DB preserves structured 'output' items
     # which the frontend strips, causing tool calls to be merged into content.
     chat_id = metadata.get("chat_id")
-    parent_message_id = metadata.get("parent_message_id")
+    context_message_id = metadata.get("context_message_id") or metadata.get(
+        "parent_message_id"
+    )
 
-    if chat_id and parent_message_id and not chat_id.startswith("local:"):
-        db_messages = load_messages_from_db(chat_id, parent_message_id)
+    if chat_id and context_message_id and not chat_id.startswith("local:"):
+        db_messages = load_messages_from_db(chat_id, context_message_id)
         if db_messages:
             system_message = get_system_message(form_data.get("messages", []))
             form_data["messages"] = (

backend/open_webui/utils/misc.py

@@ -6,7 +6,7 @@ import uuid
 import logging
 from datetime import timedelta
 from pathlib import Path
-from typing import Callable, Optional, Sequence, Union
+from typing import Any, Callable, Optional, Sequence, Union
 import json
 import aiohttp
 import mimeparse
@@ -275,6 +275,51 @@ def convert_output_to_messages(output: list, raw: bool = False) -> list[dict]:
     return messages
 
 
+def merge_assistant_content_into_output_messages(
+    messages: list[dict], content: Any
+) -> list[dict]:
+    """
+    Preserve tool/reasoning reconstruction from stored `output`, but let an
+    edited assistant `content` replace the final assistant-visible text.
+
+    This keeps the backend-authoritative message chain editable without
+    discarding tool calls or other structured output items.
+    """
+    if not messages:
+        return messages
+
+    content_text = get_content_from_message({"content": content})
+    if not content_text or not content_text.strip():
+        return messages
+
+    merged_messages = list(messages)
+    assistant_indexes = [
+        idx
+        for idx, message in enumerate(merged_messages)
+        if message.get("role") == "assistant"
+    ]
+
+    if not assistant_indexes:
+        return [*merged_messages, {"role": "assistant", "content": content}]
+
+    non_tool_assistant_indexes = [
+        idx
+        for idx in assistant_indexes
+        if not merged_messages[idx].get("tool_calls")
+    ]
+    target_idx = (
+        non_tool_assistant_indexes[-1]
+        if non_tool_assistant_indexes
+        else assistant_indexes[-1]
+    )
+
+    merged_messages[target_idx] = {
+        **merged_messages[target_idx],
+        "content": content,
+    }
+    return merged_messages
+
+
 def get_last_user_message(messages: list[dict]) -> Optional[str]:
     message = get_last_user_message_item(messages)
     if message is None:

src/lib/components/chat/Chat.svelte

@@ -2030,7 +2030,14 @@
 			.map((token) => decodeURIComponent(JSON.parse(`"${token.replace(/"/g, '\\"')}"`)));
 	};
 
-	const sendMessageSocket = async (model, _messages, _history, responseMessageId, _chatId) => {
+	const sendMessageSocket = async (
+		model,
+		_messages,
+		_history,
+		responseMessageId,
+		_chatId,
+		{ contextMessageId = null }: { contextMessageId?: string | null } = {}
+	) => {
 		const responseMessage = _history.messages[responseMessageId];
 		const userMessage = _history.messages[responseMessage.parentId];
 
@@ -2223,6 +2230,7 @@
 				chat_id: $chatId,
 
 				id: responseMessageId,
+				context_message_id: contextMessageId ?? undefined,
 				parent_id: userMessage?.id ?? null,
 				parent_message: userMessage,
 
@@ -2458,7 +2466,8 @@
 					createMessagesList(history, responseMessage.id),
 					history,
 					responseMessage.id,
-					_chatId
+					_chatId,
+					{ contextMessageId: responseMessage.id }
 				);
 			}
 		}