Refactor: refine chat message to use question only (#20654)

docs/references/http_api_reference.md

@@ -3845,13 +3845,11 @@ Starts a chat completion request. The same endpoint supports three modes:
   - `'Authorization: Bearer <YOUR_API_KEY>'`
 - Body:
 
-  - `"messages"`: `list[object]`
   - `"question"`: `string`
   - `"stream"`: `boolean`
   - `"chat_id"`: `string` (optional)
   - `"session_id"`: `string` (optional)
   - `"llm_id"`: `string` (optional)
-  - `"pass_all_history_messages"`: `boolean` (optional)
   - `"legacy"`: `boolean` (optional)
 
 ##### Request example
@@ -3893,10 +3891,8 @@ curl --request POST \
 
 ##### Request Parameters
 
-- `"messages"`: (*Body Parameter*), `list[object]`
-  The latest user message, or the conversation messages sent to the model when `pass_all_history_messages` is `true`. Either `messages` or `question` is required.
 - `"question"`: (*Body Parameter*), `string`
-  Latest user question. This is equivalent to passing `messages: [{"role": "user", "content": question}]`.
+  The user question.
 - `"stream"`: (*Body Parameter*), `boolean`
   Enables streaming output:
   - `true`: Enable streaming (default).
@@ -3907,8 +3903,6 @@ curl --request POST \
   Optional session ID. If `chat_id` is provided but `session_id` is omitted, a new session will be generated automatically.
 - `"llm_id"`: (*Body Parameter*), `string`
   Optional model override when a specific chat model should be used for this request.
-- `"pass_all_history_messages"`: (*Body Parameter*), `boolean`
-  When `chat_id` and `session_id` are provided, defaults to `false`, so the server uses stored session history and only the latest user message from the request. Set to `true` to replace/use the submitted full `messages` history, and overrides the stored session history.
 - `"legacy"`: (*Body Parameter*), `boolean`
   Defaults to `false`. Enables backward compatibility with RAGFlow v0.23.0 for streaming responses. When set to `true`:
   - Cumulative output: The `"answer"` field in each chunk returns the entire text generated so far, rather than just the new tokens (deltas).

web/src/components/message-item/group-button.tsx

@@ -163,12 +163,14 @@ interface UserGroupButtonProps extends Partial<IRemoveMessageById> {
   content: string;
   regenerateMessage?: () => void;
   sendLoading: boolean;
+  deleteDisabled?: boolean;
 }
 
 export const UserGroupButton = ({
   content,
   messageId,
   sendLoading,
+  deleteDisabled = false,
   removeMessageById,
   regenerateMessage,
 }: UserGroupButtonProps) => {
@@ -211,7 +213,7 @@ export const UserGroupButton = ({
               size="icon-xs"
               className="border-0"
               onClick={onRemoveMessage}
-              disabled={loading}
+              disabled={loading || deleteDisabled}
             >
               <LucideTrash2 />
             </Button>

web/src/components/message-item/index.tsx

@@ -43,6 +43,7 @@ interface IProps extends Partial<IRemoveMessageById>, IRegenerateMessage {
   reference: IReference;
   loading?: boolean;
   sendLoading?: boolean;
+  deleteDisabled?: boolean;
   visibleAvatar?: boolean;
   nickname?: string;
   avatar?: string;
@@ -61,6 +62,7 @@ const MessageItem = ({
   avatar,
   avatarDialog,
   sendLoading = false,
+  deleteDisabled = false,
   clickDocumentButton,
   index,
   removeMessageById,
@@ -152,6 +154,7 @@ const MessageItem = ({
                 removeMessageById={removeMessageById}
                 regenerateMessage={regenerateMessage && handleRegenerateMessage}
                 sendLoading={sendLoading}
+                deleteDisabled={deleteDisabled}
               ></UserGroupButton>
             )}
             {/* Show message content if there's any text besides the download */}

web/src/hooks/logic-hooks.ts

@@ -850,9 +850,6 @@ export const useRegenerateMessage = ({
       if (message.id) {
         removeMessagesAfterCurrentMessage(message.id);
         const index = messages.findIndex((x) => x.id === message.id);
-        // Always pass the truncated history explicitly, even when it is
-        // empty (regenerating the first question), so the backend can
-        // overwrite the session with it via pass_all_history_messages.
         const nextMessages = index !== -1 ? messages.slice(0, index) : [];
         sendMessage({
           // Keep the original id so the question/answer pair id stays

web/src/interfaces/database/chat.ts

@@ -221,6 +221,7 @@ export interface IExternalChatInfo {
 
 export interface IMessage extends Message {
   id: string;
+  awaitingServerId?: boolean;
   reference?: IReference; // the latest news has reference
   conversationId?: string; // To distinguish which conversation the message belongs to
 }

web/src/pages/next-chats/chat-stream/run-stream.ts

@@ -17,7 +17,8 @@ import { useChatStreamStore } from './store';
 export type RunChatCompletionStreamParams = {
   conversationId: string;
   chatId?: string;
-  messages: IMessage[];
+  question: string;
+  files?: IMessage['files'];
   enableThinking?: string;
   enableInternet?: boolean;
   llmSetting?: Variable;
@@ -31,7 +32,8 @@ export type RunChatCompletionStreamResult = {
 export async function runChatCompletionStream({
   conversationId,
   chatId,
-  messages,
+  question,
+  files,
   enableThinking,
   enableInternet,
   llmSetting,
@@ -75,7 +77,8 @@ export async function runChatCompletionStream({
       {
         chatId,
         sessionId: conversationId,
-        messages,
+        question,
+        files,
         enableThinking,
         enableInternet,
         llmSetting,

web/src/pages/next-chats/chat-stream/store.ts

@@ -246,15 +246,28 @@ export const useChatStreamStore = create<ChatStreamState>()(
             // placeholder) were just deleted: without it, the flush would slice
             // off whatever message is now last.
             if (!previous.isStreaming) return state;
+            const messages = previous.messages.slice(0, -1);
+            const question = messages.at(-1);
+            // The server assigns the turn ID for question-only requests.
+            // Keep the local question and streamed answer paired under that ID.
+            if (
+              answer.id &&
+              question?.role === MessageType.User &&
+              (question.id !== answer.id || question.awaitingServerId)
+            ) {
+              messages[messages.length - 1] = {
+                ...question,
+                id: answer.id,
+                awaitingServerId: false,
+              };
+            }
+            messages.push(buildAssistantMessageFromAnswer(answer));
             return {
               sessions: {
                 ...state.sessions,
                 [conversationId]: {
                   ...previous,
-                  messages: [
-                    ...previous.messages.slice(0, -1),
-                    buildAssistantMessageFromAnswer(answer),
-                  ],
+                  messages,
                   lastActiveAt: Date.now(),
                 },
               },

web/src/pages/next-chats/chat-stream/utils.ts

@@ -26,6 +26,7 @@ export function buildQuestionAndPlaceholder(message: IMessage): IMessage[] {
     {
       ...message,
       id: buildMessageUuid(message),
+      awaitingServerId: true,
     },
     {
       role: MessageType.Assistant,

web/src/pages/next-chats/chat/chat-box/single-chat-box.tsx

@@ -103,6 +103,7 @@ export function SingleChatBox({ conversation }: IProps) {
               index={i}
               isLast={i === messages.length - 1}
               removeMessageById={removeMessageById}
+              deleteDisabled={message.awaitingServerId}
               regenerateMessage={regenerateMessage}
               sendLoading={sendLoading}
             />

web/src/pages/next-chats/hooks/use-send-chat-message.ts

@@ -99,13 +99,11 @@ export const useSendMessage = () => {
     async ({
       message,
       currentConversationId,
-      messages: explicitMessages,
       enableInternet,
       enableThinking,
     }: {
       message: IMessage;
       currentConversationId?: string;
-      messages?: IMessage[];
     } & NextMessageInputOnPressEnterParameter) => {
       const sessionId = currentConversationId ?? conversationId;
 
@@ -114,12 +112,8 @@ export const useSendMessage = () => {
       const { ok, aborted } = await runChatCompletionStream({
         conversationId: sessionId,
         chatId,
-        // An explicitly provided list is authoritative, even when empty
-        // (e.g. regenerating the first question must truncate history).
-        messages: [
-          ...(Array.isArray(explicitMessages) ? explicitMessages : messages),
-          message,
-        ],
+        question: message.content,
+        files: message.files,
         enableThinking,
         enableInternet,
         llmSetting: currentDialog?.llm_setting,
@@ -134,14 +128,7 @@ export const useSendMessage = () => {
         notification.error({ message: t('message.requestError') });
       }
     },
-    [
-      conversationId,
-      chatId,
-      messages,
-      failStream,
-      t,
-      currentDialog?.llm_setting,
-    ],
+    [conversationId, chatId, failStream, t, currentDialog?.llm_setting],
   );
 
   // Hand a failed question back to the input box, but only once the box is
@@ -237,21 +224,11 @@ export const useSendMessage = () => {
 
       // Route the question to the conversation it was asked in, not whichever
       // is currently displayed.
-      //
-      // Snapshot the history before appendQuestion writes the question and its
-      // assistant placeholder into the store.
-      const history =
-        useChatStreamStore.getState().sessions[targetConversationId]
-          ?.messages ?? [];
-
       appendQuestion(targetConversationId, questionMessage);
 
       setValue('');
       sendMessage({
         currentConversationId: targetConversationId,
-        // For an existing conversation currentMessages is empty; fall back to
-        // the store's message list instead of sending an empty history.
-        messages: currentMessages.length > 0 ? currentMessages : history,
         message: {
           id,
           content: value.trim(),

web/src/pages/next-chats/hooks/use-send-single-message.ts

@@ -86,7 +86,6 @@ export function useSendSingleMessage({
           ...(storeHistoryMessages === undefined
             ? {}
             : { store_history_messages: storeHistoryMessages }),
-          pass_all_history_messages: true,
         },
         controller,
       );

web/src/services/chat-completion-stream.ts

@@ -19,7 +19,8 @@ import { EventSourceParserStream } from 'eventsource-parser/stream';
 export type ChatCompletionStreamParams = {
   chatId?: string;
   sessionId: string;
-  messages: IMessage[];
+  question: string;
+  files?: IMessage['files'];
   enableThinking?: string;
   enableInternet?: boolean;
   llmSetting?: Variable;
@@ -42,7 +43,8 @@ export function requestChatCompletionStream(
   {
     chatId,
     sessionId,
-    messages,
+    question,
+    files,
     enableThinking,
     enableInternet,
     llmSetting,
@@ -68,8 +70,8 @@ export function requestChatCompletionStream(
     body: JSON.stringify({
       chat_id: chatId,
       session_id: sessionId,
-      messages,
-      pass_all_history_messages: true,
+      question,
+      files,
       reasoning: Number(enableThinking),
       internet: enableInternet,
       ...(temperature === undefined ? {} : { temperature }),