diff --git a/frontend/src/views/chat/index.vue b/frontend/src/views/chat/index.vue
index 538b57ee..64acfa0e 100644
--- a/frontend/src/views/chat/index.vue
+++ b/frontend/src/views/chat/index.vue
@@ -609,18 +609,19 @@ const handleMsgList = async (data, isScrollType = false, newScrollHeight) => {
}
if (item.content) {
- if (!item.content.includes('') && !item.content.includes('<\/think>')) {
+ const thinkCloseTag = '';
+ if (!item.content.includes('') && !item.content.includes(thinkCloseTag)) {
item.thinkContent = "";
item.content = item.content;
item.showThink = false;
item.thinking = false;
- } else if (item.content.includes('<\/think>')) {
+ } else if (item.content.includes(thinkCloseTag)) {
// 历史消息中包含完整的 ... 标签,说明 thinking 已完成
item.showThink = true;
item.thinking = false; // 关键:标记 thinking 已完成,使 deepThink 默认折叠
- const index = item.content.trim().lastIndexOf('<\/think>');
+ const index = item.content.trim().lastIndexOf(thinkCloseTag);
item.thinkContent = item.content.trim().substring(0, index).replace('', '').trim();
- item.content = item.content.trim().substring(index + 8);
+ item.content = item.content.trim().substring(index + thinkCloseTag.length);
} else if (item.content.includes('')) {
// 内容包含 但没有 ,说明 thinking 还在进行中(不太可能出现在历史消息中)
item.showThink = true;
@@ -1018,19 +1019,20 @@ onChunk((data) => {
obj.is_fallback = true;
}
- if (fullContent.value.includes('') && !fullContent.value.includes('<\/think>')) {
+ const thinkCloseTag = '';
+ if (fullContent.value.includes('') && !fullContent.value.includes(thinkCloseTag)) {
obj.thinking = true;
obj.showThink = true;
obj.content = '';
obj.thinkContent = fullContent.value.replace('', '').trim();
- } else if (fullContent.value.includes('') && fullContent.value.includes('<\/think>')) {
+ } else if (fullContent.value.includes('') && fullContent.value.includes(thinkCloseTag)) {
obj.thinking = false;
obj.showThink = true;
// Use lastIndexOf to handle edge cases with multiple occurrences,
// consistent with history loading logic (line 280)
- const index = fullContent.value.lastIndexOf('<\/think>');
+ const index = fullContent.value.lastIndexOf(thinkCloseTag);
obj.thinkContent = fullContent.value.substring(0, index).replace('', '').trim();
- obj.content = fullContent.value.substring(index + 8).trim();
+ obj.content = fullContent.value.substring(index + thinkCloseTag.length).trim();
} else {
obj.content = fullContent.value;
}
diff --git a/internal/application/service/chat_pipeline/chat_completion_stream.go b/internal/application/service/chat_pipeline/chat_completion_stream.go
index f18a8edd..9f519568 100644
--- a/internal/application/service/chat_pipeline/chat_completion_stream.go
+++ b/internal/application/service/chat_pipeline/chat_completion_stream.go
@@ -100,29 +100,34 @@ func (p *PluginChatCompletionStream) OnEvent(ctx context.Context,
})
// Start goroutine to consume channel and emit events directly.
- // For non-agent mode, thinking content is embedded in answer stream with tags.
+ // reasoning_content is routed to EventAgentThought (SSE response_type=thinking)
+ // and plain answer text to EventAgentFinalAnswer, matching the Agent pipeline.
// The goroutine monitors ctx.Done() to avoid leaking when the context is cancelled
// and the upstream channel is not closed promptly.
go func() {
+ thinkingID := fmt.Sprintf("%s-thinking", uuid.New().String()[:8])
answerID := fmt.Sprintf("%s-answer", uuid.New().String()[:8])
- var finalContent string
- var thinkingStarted bool
- var thinkingEnded bool
+ thinkingOpen := false
+
+ closeThinking := func() {
+ if !thinkingOpen {
+ return
+ }
+ eventBus.Emit(ctx, types.Event{
+ ID: thinkingID,
+ Type: types.EventType(event.EventAgentThought),
+ SessionID: chatManage.SessionID,
+ Data: event.AgentThoughtData{
+ Done: true,
+ },
+ })
+ thinkingOpen = false
+ }
for {
select {
case <-ctx.Done():
- if thinkingStarted && !thinkingEnded {
- eventBus.Emit(ctx, types.Event{
- ID: answerID,
- Type: types.EventType(event.EventAgentFinalAnswer),
- SessionID: chatManage.SessionID,
- Data: event.AgentFinalAnswerData{
- Content: "",
- Done: true,
- },
- })
- }
+ closeThinking()
pipelineInfo(ctx, "Stream", "context_cancelled", map[string]interface{}{
"session_id": chatManage.SessionID,
})
@@ -130,18 +135,7 @@ func (p *PluginChatCompletionStream) OnEvent(ctx context.Context,
case response, ok := <-responseChan:
if !ok {
- if thinkingStarted && !thinkingEnded {
- finalContent += ""
- eventBus.Emit(ctx, types.Event{
- ID: answerID,
- Type: types.EventType(event.EventAgentFinalAnswer),
- SessionID: chatManage.SessionID,
- Data: event.AgentFinalAnswerData{
- Content: "",
- Done: true,
- },
- })
- }
+ closeThinking()
pipelineInfo(ctx, "Stream", "channel_close", map[string]interface{}{
"session_id": chatManage.SessionID,
})
@@ -167,43 +161,26 @@ func (p *PluginChatCompletionStream) OnEvent(ctx context.Context,
}
if response.ResponseType == types.ResponseTypeThinking {
- content := response.Content
- if !thinkingStarted {
- content = "" + content
- thinkingStarted = true
- }
- if response.Done && !thinkingEnded {
- content = content + ""
- thinkingEnded = true
- }
- finalContent += content
- eventBus.Emit(ctx, types.Event{
- ID: answerID,
- Type: types.EventType(event.EventAgentFinalAnswer),
- SessionID: chatManage.SessionID,
- Data: event.AgentFinalAnswerData{
- Content: content,
- Done: false,
- },
- })
- continue
- }
-
- if response.ResponseType == types.ResponseTypeAnswer {
- if thinkingStarted && !thinkingEnded {
- thinkingEnded = true
- finalContent += ""
+ if response.Content != "" {
+ thinkingOpen = true
eventBus.Emit(ctx, types.Event{
- ID: answerID,
- Type: types.EventType(event.EventAgentFinalAnswer),
+ ID: thinkingID,
+ Type: types.EventType(event.EventAgentThought),
SessionID: chatManage.SessionID,
- Data: event.AgentFinalAnswerData{
- Content: "",
+ Data: event.AgentThoughtData{
+ Content: response.Content,
Done: false,
},
})
}
- finalContent += response.Content
+ if response.Done {
+ closeThinking()
+ }
+ continue
+ }
+
+ if response.ResponseType == types.ResponseTypeAnswer {
+ closeThinking()
eventBus.Emit(ctx, types.Event{
ID: answerID,
Type: types.EventType(event.EventAgentFinalAnswer),
diff --git a/internal/handler/session/qa.go b/internal/handler/session/qa.go
index ca6ff896..be216c55 100644
--- a/internal/handler/session/qa.go
+++ b/internal/handler/session/qa.go
@@ -700,6 +700,20 @@ func (h *Handler) executeQA(reqCtx *qaRequestContext, mode qaMode, generateTitle
// (Agent mode handles completion in the defer block instead)
if mode == qaModeNormal {
var completionHandled bool
+
+ // Persist reasoning_content into agent_steps so historical reload can
+ // reconstruct the thinking card (same shape as Agent-mode steps).
+ // Accumulate on assistantMessage directly so user-initiated stop also
+ // keeps whatever reasoning had streamed before the cancel.
+ streamCtx.eventBus.On(event.EventAgentThought, func(ctx context.Context, evt event.Event) error {
+ data, ok := evt.Data.(event.AgentThoughtData)
+ if !ok || data.Content == "" {
+ return nil
+ }
+ appendQuickAnswerReasoning(streamCtx.assistantMessage, data.Content)
+ return nil
+ })
+
streamCtx.eventBus.On(event.EventAgentFinalAnswer, func(ctx context.Context, evt event.Event) error {
data, ok := evt.Data.(event.AgentFinalAnswerData)
if !ok {
@@ -903,6 +917,22 @@ func (h *Handler) persistLastRequestState(parentCtx context.Context, reqCtx *qaR
}
}
+// appendQuickAnswerReasoning accumulates streamed reasoning_content from
+// KnowledgeQA (fast answer) into a single AgentStep for history replay.
+func appendQuickAnswerReasoning(msg *types.Message, content string) {
+ if content == "" {
+ return
+ }
+ if len(msg.AgentSteps) == 0 {
+ msg.AgentSteps = types.AgentSteps{{
+ Iteration: 0,
+ Timestamp: time.Now(),
+ ToolCalls: make([]types.ToolCall, 0),
+ }}
+ }
+ msg.AgentSteps[0].ReasoningContent += content
+}
+
// completeAssistantMessage marks an assistant message as complete, updates it,
// and asynchronously indexes the Q&A pair into the chat history knowledge base.
func (h *Handler) completeAssistantMessage(ctx context.Context, assistantMessage *types.Message, userQuery string) {