From 3f9b09e3063741d2fcf6699ff2b2a77ca0083d53 Mon Sep 17 00:00:00 2001 From: wizardchen Date: Tue, 12 May 2026 19:36:28 +0800 Subject: [PATCH] fix(pipeline): make data-analysis stage opt-in per agent (#1244) The legacy in-pipeline DuckDB SQL data-analysis stage used to run on every quick-answer RAG request whose retrieved chunks included a CSV/Excel file, adding one extra LLM round-trip (~3s) to generate a SQL query that most plain Q&A users never wanted. There was no way to disable it. Introduce a per-agent DataAnalysisEnabled flag (default off), wire it through PipelineRequest, and gate the DATA_ANALYSIS stage on it. Surface the toggle in the agent editor for quick-answer agents with at least one knowledge base attached. Refs: https://github.com/Tencent/WeKnora/issues/1244 --- frontend/src/i18n/locales/en-US.ts | 4 ++++ frontend/src/i18n/locales/ko-KR.ts | 4 ++++ frontend/src/i18n/locales/ru-RU.ts | 4 ++++ frontend/src/i18n/locales/zh-CN.ts | 4 ++++ frontend/src/views/agent/AgentEditorModal.vue | 13 +++++++++++++ .../application/service/session_knowledge_qa.go | 2 +- internal/application/service/session_qa_helpers.go | 6 ++++++ internal/types/chat_manage.go | 6 ++++++ internal/types/custom_agent.go | 7 +++++++ 9 files changed, 49 insertions(+), 1 deletion(-) diff --git a/frontend/src/i18n/locales/en-US.ts b/frontend/src/i18n/locales/en-US.ts index bc3dcfc03..a09f205bc 100755 --- a/frontend/src/i18n/locales/en-US.ts +++ b/frontend/src/i18n/locales/en-US.ts @@ -4007,6 +4007,10 @@ export default { boostLabel: 'FAQ Score Boost', boostDesc: 'Multiply FAQ relevance scores by this factor to rank them higher', }, + dataAnalysis: { + enableLabel: 'Enable Tabular Data Analysis', + enableDesc: 'When the retrieved chunks come from a CSV/Excel file, ask the LLM to generate a DuckDB SQL query before answering. This adds one extra LLM call and several seconds of latency, so only enable it when you actually need SQL-style analysis.', + }, fallback: { fixed: 'Fixed Response', model: 'Model Generated', diff --git a/frontend/src/i18n/locales/ko-KR.ts b/frontend/src/i18n/locales/ko-KR.ts index 1623c0de2..8e3ba269a 100755 --- a/frontend/src/i18n/locales/ko-KR.ts +++ b/frontend/src/i18n/locales/ko-KR.ts @@ -4070,6 +4070,10 @@ export default { boostLabel: 'FAQ 점수 가중치', boostDesc: 'FAQ 관련성 점수에 이 계수를 곱하여 순위를 높입니다', }, + dataAnalysis: { + enableLabel: '표 데이터 분석 활성화', + enableDesc: '검색된 청크가 CSV/Excel 파일에서 온 경우, 답변 전에 LLM에 DuckDB SQL 쿼리를 생성하도록 요청합니다. LLM 호출이 한 번 더 발생하고 수 초의 지연이 추가되므로 SQL 스타일 분석이 실제로 필요할 때만 활성화하세요.', + }, fallback: { fixed: '고정 응답', model: '모델 생성', diff --git a/frontend/src/i18n/locales/ru-RU.ts b/frontend/src/i18n/locales/ru-RU.ts index 544b4d5e3..ab733cabd 100755 --- a/frontend/src/i18n/locales/ru-RU.ts +++ b/frontend/src/i18n/locales/ru-RU.ts @@ -3679,6 +3679,10 @@ export default { boostLabel: 'Коэффициент FAQ', boostDesc: 'Умножение оценки релевантности FAQ на этот коэффициент для повышения ранга' }, + dataAnalysis: { + enableLabel: 'Включить анализ табличных данных', + enableDesc: 'Если извлечённые фрагменты относятся к файлу CSV/Excel, перед ответом LLM сгенерирует SQL-запрос DuckDB. Это добавляет ещё один вызов LLM и несколько секунд задержки, поэтому включайте только при действительной необходимости SQL-анализа.' + }, fallback: { fixed: 'Фиксированный ответ', model: 'Генерация моделью' diff --git a/frontend/src/i18n/locales/zh-CN.ts b/frontend/src/i18n/locales/zh-CN.ts index 1cbd287be..3a94c6220 100755 --- a/frontend/src/i18n/locales/zh-CN.ts +++ b/frontend/src/i18n/locales/zh-CN.ts @@ -4003,6 +4003,10 @@ export default { boostLabel: "FAQ 分数加权", boostDesc: "FAQ 结果的相关性分数乘以此系数，使其排序更靠前", }, + dataAnalysis: { + enableLabel: "启用表格数据分析", + enableDesc: "命中 CSV/Excel 类文件时，先调用大模型生成 DuckDB SQL 进行统计或筛选，再据此回答。会额外增加一次模型调用与几秒延迟，仅在确实需要时开启。", + }, fallback: { fixed: "固定回复", model: "模型生成", diff --git a/frontend/src/views/agent/AgentEditorModal.vue b/frontend/src/views/agent/AgentEditorModal.vue index 270f7b87e..4cfca3238 100644 --- a/frontend/src/views/agent/AgentEditorModal.vue +++ b/frontend/src/views/agent/AgentEditorModal.vue @@ -1064,6 +1064,17 @@ + + +
+
+ +

{{ $t('agentEditor.dataAnalysis.enableDesc') }}

+
+
+ +
+
@@ -1850,6 +1861,8 @@ const defaultFormData = { image_storage_provider: '', // 文件类型限制 supported_file_types: [] as string[], + // 数据分析阶段开关（默认关闭，避免在普通问答上多一次 LLM 调用生成 SQL） + data_analysis_enabled: false, // FAQ 策略设置 faq_priority_enabled: true, // 是否启用 FAQ 优先策略 faq_direct_answer_threshold: 0.9, // FAQ 直接回答阈值（相似度高于此值直接使用 FAQ 答案） diff --git a/internal/application/service/session_knowledge_qa.go b/internal/application/service/session_knowledge_qa.go index 5a811800b..36e8e70f4 100644 --- a/internal/application/service/session_knowledge_qa.go +++ b/internal/application/service/session_knowledge_qa.go @@ -179,7 +179,7 @@ func (s *sessionService) KnowledgeQA( AddIf(req.WebSearchEnabled, types.WEB_FETCH). Add(types.CHUNK_MERGE). Add(types.FILTER_TOP_K). - Add(types.DATA_ANALYSIS). + AddIf(chatManage.DataAnalysisEnabled, types.DATA_ANALYSIS). Add(types.INTO_CHAT_MESSAGE). Add(types.CHAT_COMPLETION_STREAM). Build() diff --git a/internal/application/service/session_qa_helpers.go b/internal/application/service/session_qa_helpers.go index b7680b1cf..0240926fe 100644 --- a/internal/application/service/session_qa_helpers.go +++ b/internal/application/service/session_qa_helpers.go @@ -199,6 +199,12 @@ func (s *sessionService) applyAgentOverridesToChatManage( logger.Infof(ctx, "FAQ priority enabled: threshold=%.2f, boost=%.2f", cm.FAQDirectAnswerThreshold, cm.FAQScoreBoost) } + + // Data-analysis pipeline stage (opt-in, default off). + cm.DataAnalysisEnabled = customAgent.Config.DataAnalysisEnabled + if cm.DataAnalysisEnabled { + logger.Infof(ctx, "Data analysis pipeline stage enabled by custom agent") + } } // restrictMentionsToAgentScope filters user-provided @mention targets (KB IDs diff --git a/internal/types/chat_manage.go b/internal/types/chat_manage.go index 164ab0091..bda4ca64a 100644 --- a/internal/types/chat_manage.go +++ b/internal/types/chat_manage.go @@ -42,6 +42,11 @@ type PipelineRequest struct { FAQDirectAnswerThreshold float64 `json:"-"` FAQScoreBoost float64 `json:"-"` + // DataAnalysisEnabled controls whether the in-pipeline DuckDB SQL + // data-analysis stage runs. Off by default to avoid an extra LLM call on + // every RAG request that happens to retrieve CSV/Excel chunks. + DataAnalysisEnabled bool `json:"-"` + // Image / multimodal support Images []string `json:"-"` VLMModelID string `json:"-"` @@ -199,6 +204,7 @@ func (c *ChatManage) Clone() *ChatManage { FAQPriorityEnabled: c.FAQPriorityEnabled, FAQDirectAnswerThreshold: c.FAQDirectAnswerThreshold, FAQScoreBoost: c.FAQScoreBoost, + DataAnalysisEnabled: c.DataAnalysisEnabled, Images: append([]string(nil), c.Images...), VLMModelID: c.VLMModelID, ChatModelSupportsVision: c.ChatModelSupportsVision, diff --git a/internal/types/custom_agent.go b/internal/types/custom_agent.go index 225e3fa03..30ed080f7 100644 --- a/internal/types/custom_agent.go +++ b/internal/types/custom_agent.go @@ -168,6 +168,13 @@ type CustomAgentConfig struct { // When set, only files with matching extensions can be used with this agent SupportedFileTypes []string `yaml:"supported_file_types" json:"supported_file_types"` + // ===== Data Analysis Settings ===== + // Whether to run the legacy in-pipeline DuckDB SQL data-analysis stage when + // the retrieved chunks include CSV/Excel files. This issues an extra LLM + // call to generate a SQL query and is disabled by default because most + // quick-answer / RAG-style agents do not want the added latency. + DataAnalysisEnabled bool `yaml:"data_analysis_enabled" json:"data_analysis_enabled"` + // ===== FAQ Strategy Settings ===== // Whether FAQ priority strategy is enabled (FAQ answers prioritized over document chunks) FAQPriorityEnabled bool `yaml:"faq_priority_enabled" json:"faq_priority_enabled"`