fix: don't persist ephemeral sessions in multi-model chat completions (#17637)

internal/handler/chat_session.go

@@ -127,6 +127,8 @@ type ChatCompletionsRequest struct {
 	LLMID                  string                   `json:"llm_id,omitempty"`
 	PassAllHistoryMessages *bool                    `json:"pass_all_history_messages,omitempty"`
 	PassAllHistory         *bool                    `json:"pass_all_history,omitempty"`
+	StoreHistoryMessages   *bool                    `json:"store_history_messages,omitempty"`
+	StoreHistory           *bool                    `json:"store_history,omitempty"`
 	Legacy                 bool                     `json:"legacy,omitempty"`
 	Stream                 *bool                    `json:"stream"`
 	Temperature            *float64                 `json:"temperature,omitempty"`
@@ -205,12 +207,26 @@ func (h *ChatSessionHandler) ChatCompletions(c *gin.Context) {
 		passAllHistory = *req.PassAllHistoryMessages
 	}
 
+	// Resolve store_history from either alias (default true, mirrors Python)
+	storeHistory := true
+	if req.StoreHistory != nil {
+		storeHistory = *req.StoreHistory
+	}
+	if req.StoreHistoryMessages != nil {
+		storeHistory = *req.StoreHistoryMessages
+	}
+	if !storeHistory && !passAllHistory {
+		common.ErrorWithCode(c, common.CodeDataError, "`pass_all_history_messages` must be true when `store_history_messages` is false.")
+		return
+	}
+
 	// Remove known keys from rawBody; what remains is passthrough kwargs
 	knownKeys := []string{
 		"chat_id", "session_id", "conversation_id",
 		"messages", "question", "files",
 		"llm_id",
 		"pass_all_history_messages", "pass_all_history",
+		"store_history_messages", "store_history",
 		"legacy", "stream",
 		"temperature", "top_p", "frequency_penalty", "presence_penalty", "max_tokens",
 	}
@@ -241,7 +257,7 @@ func (h *ChatSessionHandler) ChatCompletions(c *gin.Context) {
 				req.ChatID, sessionID,
 				req.Messages, req.Question, req.Files,
 				req.LLMID, genConfig, kwargs,
-				passAllHistory, req.Legacy,
+				passAllHistory, storeHistory, req.Legacy,
 				true, streamChan,
 			)
 		}()
@@ -261,7 +277,7 @@ func (h *ChatSessionHandler) ChatCompletions(c *gin.Context) {
 			req.ChatID, sessionID,
 			req.Messages, req.Question, req.Files,
 			req.LLMID, genConfig, kwargs,
-			passAllHistory, req.Legacy,
+			passAllHistory, storeHistory, req.Legacy,
 			false, nil,
 		)
 		if err != nil {

internal/service/chat_channel.go

@@ -321,6 +321,7 @@ func (s *ChatChannelService) HandleIncomingMessage(ctx context.Context, msg Chat
 		nil,
 		map[string]interface{}{"quote": false},
 		false,
+		true,
 		false,
 		false,
 		nil,

internal/service/chat_session.go

@@ -1449,7 +1449,7 @@ func (s *ChatSessionService) ChatCompletions(
 	chatID string, sessionID string,
 	messages []map[string]interface{}, question string, files []interface{},
 	llmID string, genConfig map[string]interface{}, kwargs map[string]interface{},
-	passAllHistory bool, legacy bool,
+	passAllHistory bool, storeHistory bool, legacy bool,
 	stream bool, streamChan chan<- string,
 ) (map[string]interface{}, error) {
 	if ctx == nil {
@@ -1505,7 +1505,7 @@ func (s *ChatSessionService) ChatCompletions(
 				return fail(common.NewCodedError(common.CodeDataError, "Session does not belong to this chat!"))
 			}
 		} else {
-			session, err = s.createSessionForCompletion(ctx, chatID, dialog, userID)
+			session, err = s.createSessionForCompletion(ctx, chatID, dialog, userID, storeHistory)
 			if err != nil {
 				return fail(err)
 			}
@@ -1669,7 +1669,7 @@ func (s *ChatSessionService) ChatCompletions(
 		sendOrCancel(fmt.Sprintf("data:%s\n\n", marshalJSONWithSpaces(wrapper)))
 
 		// Persist session state (matches Python's update_by_id after loop)
-		if session != nil {
+		if session != nil && storeHistory {
 			s.updateSessionMessages(ctx, session, s.getSessionMessagesAsSlice(session), reference)
 		}
 	} else {
@@ -1679,7 +1679,9 @@ func (s *ChatSessionService) ChatCompletions(
 			if chatID != "" {
 				result["chat_id"] = chatID
 			}
-			s.updateSessionMessages(ctx, session, s.getSessionMessagesAsSlice(session), reference)
+			if storeHistory {
+				s.updateSessionMessages(ctx, session, s.getSessionMessagesAsSlice(session), reference)
+			}
 			return sanitizeJSONFloats(result).(map[string]interface{}), nil
 		}
 		ans["id"] = messageID
@@ -1830,8 +1832,7 @@ func (s *ChatSessionService) buildDefaultCompletionDialog(tenantID string) *enti
 	}
 }
 
-func (s *ChatSessionService) createSessionForCompletion(ctx context.Context, chatID string, dialog *entity.Chat, userID string) (*entity.ChatSession, error) {
-	newID := utility.GenerateUUID()
+func (s *ChatSessionService) createSessionForCompletion(ctx context.Context, chatID string, dialog *entity.Chat, userID string, saveSession bool) (*entity.ChatSession, error) {
 	name := "New session"
 
 	prologue := "Hi! I'm your assistant. What can I do for you?"
@@ -1848,13 +1849,18 @@ func (s *ChatSessionService) createSessionForCompletion(ctx context.Context, cha
 	refJSON, _ := json.Marshal([]interface{}{})
 
 	session := &entity.ChatSession{
-		ID:        newID,
 		DialogID:  chatID,
 		Name:      &name,
 		Message:   msgJSON,
 		UserID:    &userID,
 		Reference: refJSON,
 	}
+	if !saveSession {
+		// Multi-model comparison sends store_history_messages=false; the
+		// ephemeral session must not be persisted (mirrors Python).
+		return session, nil
+	}
+	session.ID = utility.GenerateUUID()
 	if err := s.chatSessionDAO.Create(ctx, dao.DB, session); err != nil {
 		return nil, err
 	}

internal/service/chat_session_test.go

@@ -930,6 +930,7 @@ func TestChatCompletionsPassesRequestUserIDToPipeline(t *testing.T) {
 		nil,
 		nil,
 		false,
+		true,
 		false,
 		false,
 		nil,
@@ -1010,6 +1011,7 @@ func TestChatCompletionsStreamFinalCarriesDecoratedReference(t *testing.T) {
 		nil,
 		nil,
 		false,
+		true,
 		false,
 		true,
 		streamChan,
@@ -1106,6 +1108,7 @@ func TestChatCompletionsModelIDOverrideUsesModelResolver(t *testing.T) {
 		nil,
 		nil,
 		false,
+		true,
 		false,
 		false,
 		nil,
@@ -1121,6 +1124,69 @@ func TestChatCompletionsModelIDOverrideUsesModelResolver(t *testing.T) {
 	}
 }
 
+// Multi-model comparison sends store_history_messages=false; the ephemeral
+// session must never be persisted, otherwise every model card adds a
+// "New session" row to the session list.
+func TestChatCompletionsStoreHistoryFalseDoesNotPersistSession(t *testing.T) {
+	store := newFakeSessionStore()
+	store.dialogs["dialog-1"] = &entity.Chat{
+		ID:         "dialog-1",
+		TenantID:   "tenant-owner",
+		LLMID:      "chat@factory",
+		LLMSetting: entity.JSONMap{},
+		PromptConfig: entity.JSONMap{
+			"parameters": []interface{}{},
+			"prologue":   "Welcome!",
+		},
+	}
+	store.dialogExists["tenant-owner|dialog-1"] = true
+
+	pipeline := &fakePipeline{
+		resultChan: makeResultChan(
+			AsyncChatResult{Answer: "ok", Final: true, Reference: map[string]interface{}{"chunks": []interface{}{}}},
+		),
+	}
+
+	svc := &ChatSessionService{
+		chatSessionDAO: store,
+		userTenantDAO:  &fakeTenantStore{tenantIDs: []string{"tenant-owner"}},
+		pipeline:       pipeline,
+	}
+
+	result, err := svc.ChatCompletions(
+		context.Background(),
+		"user-1",
+		"dialog-1",
+		"",
+		[]map[string]interface{}{{"role": "user", "content": "hi"}},
+		"",
+		nil,
+		"",
+		nil,
+		nil,
+		true,
+		false,
+		false,
+		false,
+		nil,
+	)
+	if err != nil {
+		t.Fatalf("ChatCompletions failed: %v", err)
+	}
+	if result == nil {
+		t.Fatal("expected a non-nil answer")
+	}
+	if len(store.createCalled) != 0 {
+		t.Fatalf("session must not be created when store_history_messages=false, got %d creates", len(store.createCalled))
+	}
+	if len(store.updateCalled) != 0 {
+		t.Fatalf("session must not be updated when store_history_messages=false, got %d updates", len(store.updateCalled))
+	}
+	if len(store.sessions) != 0 {
+		t.Fatalf("no session should be stored, got %d", len(store.sessions))
+	}
+}
+
 func TestCompletion_EmptyMessages(t *testing.T) {
 	svc := &ChatSessionService{
 		chatSessionDAO: &fakeSessionStore{},