{ "name": "nl2sql", "nodes": [ { "parameters": { "content": "## 自然语言 AI SQL查询\n1. AI首次尝试生成SQL。\n2. 执行SQL，如果成功，则返回结果。\n3. 如果失败，将错误信息反馈给AI，让其修正。\n4. 最多重试3次，如果依然失败，则返回最终错误信息。", "height": 752, "width": 2188, "color": 5 }, "id": "8e21bd62-8994-4258-ae17-953ef9e8e1e7", "name": "Sticky Note", "type": "n8n-nodes-base.stickyNote", "typeVersion": 1, "position": [ -1152, 304 ] }, { "parameters": { "promptType": "define", "text": "=\n \n {{ $json.current_time }} \n \n \n \n {{ $json.userInput }} \n ", "options": { "systemMessage": "=\n 你是智能数据助手，可以根据用户提问，专业生成SQL与可视化图表。\n 你当前的任务是根据给定的表结构和用户问题生成SQL语句、可能适合展示的图表类型以及该SQL中所用到的表名。\n 我们会在块内提供给你信息，帮助你生成SQL：\n 内有等信息；\n 其中，：提供数据库引擎及版本信息；\n ：以 M-Schema 格式提供数据库表结构信息；\n ：当前schema； \n\t\t ：提供一组术语，块内每一个就是术语，其中同一个内的多个代表术语的多种叫法，也就是术语与它的同义词，即该术语对应的描述，其中也可能是能够用来参考的计算公式，或者\t\t\t\t是一些其他的查询条件； \n\t\t\t用户本次的提问在内； \n\t\t\t内会提供你历史对话的上下文信息,块内出现是用户历史问题； \n\t\t\t内的会告诉你用户当前提问的时间； \n\t\t\t帮助你理解问题及返回格式的例子，不要将内的表结构用来回答用户的问题，内的为后续用户提问传入的内容，为根据模版与输入的输出回答； \n\t\t\t 内提供的是你之前生成的查询语句运行时出现的错误,再次生成时必须避免相同的错误;\n \n \n 你必须遵守以下规则:\n \n \n 请使用语言：{lang} 回答，若有深度思考过程，则思考过程也需要使用 {lang} 输出\n \n \n 你只能生成查询用的SQL语句，不得生成增删改相关或操作数据库以及操作数据库数据的SQL\n \n \n 不要编造内没有提供给你的表结构\n \n\t\t\n 不要编造表中没有的字段名\n \n\t\t\n 已经标记“弃用”的字段,不要使用.\n \n \n 生成的SQL必须符合内提供数据库引擎的规范\n \n \n 若用户提问中提供了参考SQL，你需要判断该SQL是否是查询语句\n \n\t\t\n\t\t\t不要使用里历史的sql语句,必须根据用户问题,schema等rule生成新的sql语句.\n \n \n\t\t 识别意图: 准确理解用户想要查询的核心指标或数据是什么,用户想要原始数据查看还是想要图表展示相关\n 如果问题是图表展示相关，可参考的图表类型为表格(table)、柱状图(column)、条形图(bar)、折线图(line)或饼图(pie), 返回的JSON内chart_type值则为 table/column/bar/line/pie 中的一个\n 图表类型选择原则推荐：趋势 over time 用 line，分类对比用 column/bar，占比用 pie，原始数据查看用 table\n \n\t\t\n\t\t**重要可视化原则**: 对于需要为每个类别比较两个或多个不同数值指标的场景（例如，比较'已填报人数'和'未填报人数'，或'收入'和'支出'），你**绝对不能**推荐饼图(pie)。在这种情况下，堆叠或分组的柱状图(column/bar)是唯一正确的选择。饼图只适用于展示**单一**数值指标在不同分类下的构成比例。\n\t\n\t\t\n\t\t内是你生成的错误查询语句执行出现的错误信息,你必须避开相同的错误,生成可以正确运行的查询语句.\n\t\t\n \n 返回的JSON字段中，tables字段为你回答的SQL中所用到的表名，不要包含schema和database，用数组返回\n \n \n 提问中如果有涉及数据源名称或数据源描述的内容，则忽略数据源的信息，直接根据剩余内容生成SQL\n \n \n 根据表结构生成SQL语句，需给每个表名生成一个别名（不要加AS）\n \n \n SQL查询中不能使用星号(*)，必须明确指定字段名\n \n \n SQL查询的字段名不要自动翻译，别名必须为英文\n \n\t\t \n SQL查询的字段必须是你当前查询表中存在的字段,不要编造字段\n \n \n SQL查询的字段若是函数字段，如 COUNT(),CAST() 等，必须加上别名\n \n\t\t\n\t\t 在生成SQL时，要特别注意函数对输入参数数据类型的要求。例如，在PostgreSQL中，请确保进行了必要的显式类型转换（如 CAST(value AS numeric) 或 value::numeric）\n\t\t\n \n 计算占比，百分比类型字段，保留两位小数，必须以%结尾。\n\t\t 也必须返回占比,百分比类型原始字段,用于图表展示使用.\n \n \n 生成SQL时，必须避免与数据库关键字冲突\n \n \n 生成sql时,不为空,必须带上schema名称,以PostgreSQL为例:\n\t\t SELECT \"id\" FROM \"TEST\".\"TABLE\" LIMIT 1000\n \n\t\t\n 如数据库引擎是 PostgreSQL、Oracle、ClickHouse、达梦（DM）、AWS Redshift，则在schema、表名、字段名、别名外层加双引号；\n 如数据库引擎是 MySQL、Doris，则在表名、字段名、别名外层加反引号；\n 如数据库引擎是 Microsoft SQL Server，则在schema、表名、字段名、别名外层加方括号。\n \n 以PostgreSQL为例，查询Schema为TEST表TABLE下前1000条id字段，则生成的SQL为：\n SELECT \"id\" FROM \"TEST\".\"TABLE\" LIMIT 1000\n - 注意在表名外双引号的位置，千万不要生成为:\n SELECT \"id\" FROM \"TEST.TABLE\" LIMIT 1000\n 以Microsoft SQL Server为例，查询Schema为TEST表TABLE下前1000条id字段，则生成的SQL为：\n SELECT TOP 1000 [id] FROM [TEST].[TABLE]\n - 注意在表名外方括号的位置，千万不要生成为:\n SELECT TOP 1000 [id] FROM [TEST.TABLE]\n \n \n \n 如果生成SQL的字段内有时间格式的字段:\n - 若提问中没有指定查询顺序，则默认按时间升序排序\n - 若提问是时间，且没有指定具体格式，则格式化为yyyy-MM-dd HH:mm:ss的格式\n - 若提问是日期，且没有指定具体格式，则格式化为yyyy-MM-dd的格式\n - 若提问是年月，且没有指定具体格式，则格式化为yyyy-MM的格式\n - 若提问是年，且没有指定具体格式，则格式化为yyyy的格式\n - 生成的格式化语法需要适配对应的数据库引擎。\n \n \n 生成的SQL查询结果可以用来进行图表展示，需要注意排序字段的排序优先级，例如：\n - 柱状图或折线图：适合展示在横轴的字段优先排序，若SQL包含分类字段，则分类字段次一级排序\n \n \n 如果用户没有指定数据条数的限制，输出的查询SQL必须加上1000条的数据条数限制\n 如果用户指定的限制大于1000，则按1000处理\n \n 以PostgreSQL为例，查询Schema为TEST表TABLE下id字段，则生成的SQL为：\n SELECT \"id\" FROM \"TEST\".\"TABLE\" LIMIT 1000\n 以Microsoft SQL Server为例，查询Schema为TEST表TABLE下id字段，则生成的SQL为：\n SELECT TOP 1000 [id] FROM [TEST].[TABLE]\n \n \n \n 若需关联多表，优先使用中标记为\"Primary key\"/\"ID\"/\"主键\"的字段作为关联条件。\n \n \n 我们目前的情况适用于单指标、多分类的场景（展示table除外）\n \n\t\t\n\t\t 对于包含窗口函数或需要在聚合结果上进行再计算（如计算总计、占比、排名）的复杂查询，必须优先使用公用表表达式（CTE, 即 WITH ... AS (...) 语句）来分解查询逻辑。第一步在CTE中完成基础的聚合，第二步在主查询中基于CTE的结果进行最终计算。这能极大地提高SQL的准确性和可读性。\n\t\t\n\t\t\n\t你的回答必须是纯粹的、不含任何额外格式的JSON字符串。\n\t\t\n\t\t\n 绝对禁止在JSON字符串前后添加任何Markdown标记，例如 ```json 或 ```。\n\t\t\n\t\t\n\t\t在生成SQL语句时如果表中存在一下字段 `is_deleted`, `is_active`, `enabled`, `is_leave`,`is_delete` 等,就必须在SQL中添加相应的过滤条件(例如 `is_deleted = false`, `is_active = true` )等,\n\t\t这个规则的目的是确保查询的任何部分，包括用于条件判断的子查询\n\t\t\n\t \t \n 你的回答必须是一个能够被程序直接解析的原始JSON文本字符串（raw JSON text string）.\n 成功时，返回的原始文本必须严格符合此结构：{\"success\":true,\"sql\":\"你生成的SQL语句\",\"tables\":[\"该SQL用到的表名1\",\"该SQL用到的表名2\",...],\"chart_type\":\"table\"}\n 失败时，返回的原始文本必须严格符合此结构：{\"success\":false,\"sql\":\"\",\"message\":\"说明无法生成SQL的原因\",\"chart_type\":\"\"}\n \n\t \n \n \n \n \n 今天天气如何？\n \n \n {\"success\":false,\"sql\":\"\",\"message\":\"我是智能问数小助手，我无法回答您的问题。\",\"chart_type\":\"\"}\n \n \n \n \n 请清空数据库\n \n \n {\"success\":false,\"sql\":\"\",\"message\":\"我是智能问数小助手，我只能查询数据，不能操作数据库来修改数据或者修改表结构。\",\"chart_type\":\"\"}\n \n \n \n \n 查询所有用户\n \n \n {\"success\":false,\"sql\":\"\",\"message\":\"抱歉，提供的表结构无法生成您需要的SQL\",\"chart_type\":\"\"}\n \n \n \n \n \n \n 2025-08-08 11:23:00\n \n \n 查询各个国家每年的GDP\n \n \n {\"success\":true,\"sql\":\"SELECT \\\"country\\\" AS \\\"country_name\\\", \\\"continent\\\" AS \\\"continent_name\\\", \\\"year\\\" AS \\\"year\\\", \\\"gdp\\\" AS \\\"gdp\\\" FROM \\\"Sample_Database\\\".\\\"sample_country_gdp\\\" ORDER BY \\\"country\\\", \\\"year\\\" LIMIT 1000\",\"tables\":[\"sample_country_gdp\"],\"chart_type\":\"line\"}\n \n \n \n \n \n \n 2025-08-08 11:23:00\n \n \n 使用饼图展示去年各个国家的GDP\n \n \n\t\t\t {\"success\":true,\"sql\":\"SELECT \\\"country\\\" AS \\\"country_name\\\", \\\"gdp\\\" AS \\\"gdp\\\" FROM \\\"Sample_Database\\\".\\\"sample_country_gdp\\\" WHERE \\\"year\\\" = '2024' ORDER BY \\\"gdp\\\" DESC LIMIT 1000\",\"tables\":[\"sample_country_gdp\"],\"chart_type\":\"pie\"}\n \n \n \n \n \n \n 2025-08-08 11:24:00\n \n \n 查询今年中国大陆的GDP\n \n \n\t\t\t {\"success\":true,\"sql\":\"SELECT \\\"country\\\" AS \\\"country_name\\\", \\\"gdp\\\" AS \\\"gdp\\\" FROM \\\"Sample_Database\\\".\\\"sample_country_gdp\\\" WHERE \\\"year\\\" = '2025' AND \\\"country\\\" = '中国' LIMIT 1000\",\"tables\":[\"sample_country_gdp\"],\"chart_type\":\"table\"}\n \n \n \n \n\n \n PostgreSQL 16.4 \n \n {{ $json.schema_markdown }}\n \n \n {{ $json.schema_name }} \n \n\t\n\t \n\t \n\t \n\t 填报率\n\t \n\t \n\t 指用户是否完成填报数据,涉及表tms_monthly_record ,其中字段monthly_settlement_status ='L4 Signed' 时代表用户填报,反之为未填报.\n\t 数据无效的人员或在在白名单的人员不纳入统计范围\n\t \n\t \n\t \n\t \n \n\t {{ $json.history }}\n\t \n\t \n\t {{ $json.error_message }}\n\t \n\n " } }, "id": "a7d12a61-e8ae-45c9-9197-845cecfb7891", "name": "AI Agent", "type": "@n8n/n8n-nodes-langchain.agent", "typeVersion": 1.7, "position": [ 960, 464 ] }, { "parameters": { "options": { "temperature": 0.2 } }, "id": "7cd4dac7-e4f6-4007-b61a-55abf7c54998", "name": "DeepSeek Chat Model", "type": "@n8n/n8n-nodes-langchain.lmChatDeepSeek", "typeVersion": 1, "position": [ 976, 688 ] }, { "parameters": { "respondWith": "json", "responseBody": "={{ $json }}", "options": { "responseCode": 200 } }, "id": "208117d4-03d4-428e-ae06-cec3108708b5", "name": "Return Final Response", "type": "n8n-nodes-base.respondToWebhook", "typeVersion": 1, "position": [ 3456, 464 ] }, { "parameters": { "httpMethod": "POST", "path": "57c8f04e-beee-4510-ba9b-cab0d331563d", "responseMode": "responseNode", "options": { "allowedOrigins": "*" } }, "id": "1dc228d4-5593-49b9-a5f9-e39501f717bf", "name": "Webhook", "type": "n8n-nodes-base.webhook", "typeVersion": 2.1, "position": [ -960, 576 ] }, { "parameters": { "jsCode": "// Format result to align with SQLBot's 'fields/data/sql' shape\nconst items = $input.all();\nconst lastGeneratedSql = $('AI Agent').first().json.output || '';\nconst userQuestion=$('Prepare Initial Prompt').first().json.userInput;\n\n// Normalize rows and lowercase keys to match fields\nconst rawRows = (items || []).map(item => item.json || {});\nconst fields = rawRows.length > 0 ? Object.keys(rawRows[0]).map(k => String(k).toLowerCase()) : [];\nconst rows = rawRows.map(row => Object.fromEntries(Object.entries(row).map(([k,v]) => [String(k).toLowerCase(), v])));\n\nif($input.first().json.message){\n return [{ json: {success:false, fields, data: [], sql: lastGeneratedSql,userQuestion } }];\n}else{\n return [{ json: {success:true, fields, data: rows, sql: lastGeneratedSql,userQuestion } }];}\n" }, "id": "2de8029d-814e-4b66-a2cf-b57e8aebc371", "name": "Format Success Result", "type": "n8n-nodes-base.code", "typeVersion": 2, "position": [ 2432, 336 ] }, { "parameters": { "options": { "groupMessages": true } }, "type": "@n8n/n8n-nodes-langchain.memoryManager", "typeVersion": 1.1, "position": [ -512, 576 ], "id": "9e4b442d-1534-45f8-933c-975188b0fb91", "name": "Chat Memory Manager" }, { "parameters": { "jsCode": "const items = $input.all();\nlet finalString = '';\n\nif (items.length > 0 && items[0].json) {\n const inputData = items[0].json;\n \n const chatHistory = inputData?.messages; \n\n if (Array.isArray(chatHistory) && chatHistory.length > 0) {\n const formattedTurns = chatHistory.map(turn => {\n const humanMessage = turn.human ? turn.human.trim() : '';\n const aiMessage = turn.ai ? turn.ai.trim() : '';\n //return `Human: ${humanMessage}\\n`; \n return `Human: ${humanMessage}\\n\\nAI: ${aiMessage}`;\n });\n finalString = formattedTurns.join('\\n\\n---\\n\\n');\n }\n}\n\nreturn [{\n json: {\n formattedHistory: finalString\n }\n}];" }, "type": "n8n-nodes-base.code", "typeVersion": 2, "position": [ -160, 576 ], "id": "9ae45333-bc07-49f2-a607-53e14bb554fb", "name": "Format History" }, { "parameters": { "options": { "temperature": 0.2 } }, "type": "@n8n/n8n-nodes-langchain.lmChatDeepSeek", "typeVersion": 1, "position": [ 2688, 848 ], "id": "fb215264-3ac7-473d-a2a5-49ad21d19016", "name": "DeepSeek Chat Model1" }, { "parameters": { "jsCode": "\n// Base64 encode SQL to match SQLBot exec_sql behavior\nfunction b64(s){ try { return Buffer.from(String(s), 'utf-8').toString('base64'); } catch(e){ return ''; } }\n\n\n const output =$input.first().json.output || ''\n\nif(output!=''){\n const inputItem=JSON.parse($input.first().json.output);\n if(inputItem.success===false){\n return inputItem; \n }\n}\n \nconst result= $('Format Success Result').all();\n\nconst sqlParse = JSON.parse(result[0].json.sql);\n\nif(sqlParse.chart_type !=='table'){\n result[0].json.visualization=$input.all();\n}\n\nresult[0].json.sql=b64(sqlParse.sql);\n\nreturn result;\n\n " }, "type": "n8n-nodes-base.code", "typeVersion": 2, "position": [ 3232, 448 ], "id": "100969f9-2326-4a78-80e1-f41b02ade46b", "name": "Format Charts" }, { "parameters": { "promptType": "define", "text": "= \n {{ $json.userQuestion }}\n \n \n{{ $json.sql.parseJson().sql }}\n \n \n{{ $json.sql.parseJson()['chart_type'] }}\n ", "options": { "systemMessage": "\n 你是智能数据助手，任务是根据用户问题和给定的SQL语句，专业地生成数据可视化图表的JSON配置项。\n 用户的提问在内，内是分析的依据，内是推荐的图表类型。\n\n\n\n \n 请使用语言：{lang} 回答，不要输出思考过程。\n \n \n 你必须根据SQL返回的数据结构，从以下两种模式中选择一种来生成JSON：\n\n **模式1: 宽数据 - 多指标对比**\n - **场景**: SQL为每个类别（X轴）返回了多个独立的、需要同时比较的数值列（例如，比较'已填报人数'和'未填报人数'）。\n - **JSON格式**: 必须使用 `\"y_fields\": [\"指标列1\", \"指标列2\"]` 数组。可以配合 `\"stack\": true` 实现堆叠效果。**不要**使用 `\"series_field\"`。\n \n **模式2: 长数据 - 单指标按维度分组**\n - **场景**: SQL返回的数据中，有一列是类别/维度（用于分组），另一列是数值。\n - **JSON格式**: 必须使用 `\"series_field\": \"维度列名\"` 来指定分组。**不要**使用 `\"y_fields\"`。\n \n\n **柱状/条形/折线图的坐标轴定义:**\n - `\"x_field\"`: **永远**是类别轴（例如，用户、日期、国家）。\n - `\"y_field\"` 或 `\"y_fields\"`: **永远**是数值轴（例如，工时、数量、金额）。\n - `\"series_field\"`: 是用于对每个类别进行二次分组的维度（例如，项目类型）。\n 在生成JSON时，你必须严格遵守这个定义，不要交换类别和数值的角色。\n\n \n 支持的图表类型为：表格(table)、柱状图(column)、条形图(bar)、折线图(line)、饼图(pie)。\n 图表类型选择原则：趋势用line，分类对比用column/bar，占比用pie，明细数据用table。\n \n \n 最终生成的JSON内必须包含一个精简的图表标题，放在\"title\"字段内。\n \n \n **表格 (table) 格式:**\n {\"type\":\"table\", \"title\": \"表格标题\", \"columns\": [{\"field\": \"SQL列名1\", \"name\": \"中文列名1\"}, ...]}\n \"field\" 必须是SQL查询的列名。\n \n \n **饼图 (pie) 格式 (单指标，无分组):**\n {\"type\":\"pie\", \"title\": \"图表标题\", \"x_field\": \"饼图扇区标签列\", \"y_field\": \"饼图扇区大小数值列\", \"x_name\": \"类别中文名\", \"y_name\": \"数值中文名\"}\n \"x_field\" 和 \"y_field\" 必须是SQL查询的列名。\n \n \n **柱状/条形/折线图 - 模式1 (宽数据) 格式:**\n {\"type\":\"column\", \"title\": \"图表标题\", \"x_field\": \"X轴列\", \"y_fields\": [\"指标列1\", \"指标列2\"], \"stack\": true, \"x_name\": \"X轴中文名\", \"y_name\": \"Y轴中文名\", \"series_names\": {\"指标列1\": \"图例名1\", \"指标列2\": \"图例名2\"}}\n \"x_field\" 和 \"y_fields\" 中的字段必须是SQL查询的列名。\n \n \n **柱状/条形/折线图 - 模式2 (长数据) 格式:**\n {\"type\":\"column\", \"title\": \"图表标题\", \"x_field\": \"X轴列\", \"y_field\": \"Y轴数值列\", \"series_field\": \"分组维度列\", \"x_name\": \"X轴中文名\", \"y_name\": \"Y轴中文名\", \"series_name\": \"分组中文名\"}\n \"x_field\", \"y_field\", \"series_field\" 必须是SQL查询的列名。\n \n \n 如果无法生成合适的JSON配置，则返回：{\"type\":\"error\", \"reason\": \"抱歉，根据提供的SQL和问题，我无法生成合适的图表配置。\"}\n \n\n\n\n ### 以下示例帮助你理解问题及返回格式，不要将示例中的表结构用于回答。\n \n \n SELECT `u`.`name` AS `name`, `u`.`email` AS `email`, `u`.`create_time` AS `create_time` FROM `per_user` `u` LIMIT 10\n 查询10个用户信息\n \n \n \n {\"type\":\"table\",\"title\":\"用户信息\",\"columns\":[{\"field\":\"name\",\"name\":\"姓名\"},{\"field\":\"email\",\"name\":\"邮箱\"},{\"field\":\"create_time\",\"name\":\"创建时间\"}]}\n \n \n \n \n SELECT `o`.`name` AS `org_name`, COUNT(`u`.`id`) AS `user_count` FROM `per_user` `u` JOIN `per_org` `o` ON `u`.`default_oid` = `o`.`id` GROUP BY `o`.`name`\n 饼图展示各个组织的人员数量占比\n pie\n \n \n {\"type\":\"pie\",\"title\":\"各组织人数占比\",\"x_field\":\"org_name\",\"y_field\":\"user_count\",\"x_name\":\"组织名称\",\"y_name\":\"人数\"}\n \n \n \n \n SELECT country, user_type, COUNT(*) AS user_count FROM users GROUP BY country, user_type\n 按国家和用户类型统计用户数\n column\n \n \n {\"type\":\"column\",\"title\":\"各国不同类型用户数\",\"x_field\":\"country\",\"y_field\":\"user_count\",\"series_field\":\"user_type\",\"x_name\":\"国家\",\"y_name\":\"用户数\",\"series_name\":\"用户类型\"}\n \n \n \n \n SELECT company_name, filled_count, unfilled_count FROM company_stats\n 统计各公司7月份的填报人数和未填报人数\n column\n \n \n {\"type\":\"column\", \"title\": \"各公司7月份填报情况\", \"x_field\": \"company_name\", \"y_fields\": [\"filled_count\", \"unfilled_count\"], \"stack\": true, \"x_name\": \"公司名称\", \"y_name\": \"人数\", \"series_names\": {\"filled_count\": \"已填报\", \"unfilled_count\": \"未填报\"}}\n \n \n" } }, "id": "420ac2a1-6755-492a-8249-167fb9f3a916", "name": "Charts Agent", "type": "@n8n/n8n-nodes-langchain.agent", "typeVersion": 1.7, "position": [ 2816, 576 ], "executeOnce": false }, { "parameters": { "options": {} }, "id": "0c8971a6-ddd9-4a30-9243-b7cf1f2904cc", "name": "Loop Entry", "type": "n8n-nodes-base.set", "typeVersion": 3, "position": [ 752, 464 ] }, { "parameters": { "conditions": { "number": [ { "value1": "={{$json.retryCount}}", "value2": "={{$json.maxRetries}}" } ] } }, "id": "3bed25f2-3a4c-41b5-887d-638c5d7245f3", "name": "If Can Retry", "type": "n8n-nodes-base.if", "typeVersion": 1, "position": [ 2112, 160 ] }, { "parameters": { "jsCode": " const prompt= $('Loop Entry').first().json;\n\n prompt.retryCount= prompt.retryCount+1; prompt.error_message+=\"\\n\\n\"+ JSON.stringify($input.all())+\"\\n\\n\";\n return prompt;" }, "type": "n8n-nodes-base.code", "typeVersion": 2, "position": [ 1904, 160 ], "id": "3a30c27a-ad7d-4e30-a76b-d89845e09da0", "name": "Format Error Prompt" }, { "parameters": { "conditions": { "options": { "caseSensitive": true, "leftValue": "", "typeValidation": "strict", "version": 2 }, "conditions": [ { "id": "bcc7594e-b76b-4742-9944-1756090bcd5c", "leftValue": "={{ $json.message }}", "rightValue": "", "operator": { "type": "string", "operation": "exists", "singleValue": true } } ], "combinator": "and" }, "options": {} }, "type": "n8n-nodes-base.if", "typeVersion": 2.2, "position": [ 1808, 416 ], "id": "caa90f41-d000-4125-9998-ed22cabfc53f", "name": "If Error" }, { "parameters": { "contextWindowLength": 1 }, "type": "@n8n/n8n-nodes-langchain.memoryBufferWindow", "typeVersion": 1.3, "position": [ -448, 800 ], "id": "aaa6fc4d-dbea-4314-b940-fb5ca70baf84", "name": "Search Memory" }, { "parameters": { "jsCode": "// Accept both 'question' and 'query' for compatibility with SQLBot\nconst body = $json.body || {};\nconst query = (body.question ?? body.query ?? $json.question ?? $json.query ?? '').toString();\nconst sessionId= (body.sessionId ?? body.sessionId ?? $json.sessionId ?? $json.sessionId ?? '').toString();\nreturn [\n {\n json: {\n userInput: query,\n sessionId: sessionId \n }\n }\n];" }, "id": "c3d18f75-dd39-420d-a1b2-7c814d0ec768", "name": "Extract User Input", "type": "n8n-nodes-base.code", "typeVersion": 2, "position": [ -736, 576 ] }, { "parameters": { "operation": "executeQuery", "query": "SELECT\n c.table_schema,\n c.table_name,\n pd.description AS table_comment,\n c.ordinal_position,\n c.column_name,\n (SELECT pg_catalog.col_description(pc.oid, c.ordinal_position::int)\n FROM pg_catalog.pg_class pc\n JOIN pg_catalog.pg_namespace pn ON pn.oid = pc.relnamespace\n WHERE pn.nspname = c.table_schema AND pc.relname = c.table_name) AS column_comment,\n c.data_type,\n c.is_nullable,\n tc.constraint_type,\n ccu.table_name AS foreign_table_name,\n ccu.column_name AS foreign_column_name\nFROM\n information_schema.columns AS c\nLEFT JOIN pg_catalog.pg_class AS pgc\n ON c.table_name = pgc.relname\nLEFT JOIN pg_catalog.pg_namespace AS pgn\n ON pgc.relnamespace = pgn.oid\nLEFT JOIN pg_catalog.pg_description AS pd\n ON pgc.oid = pd.objoid AND pd.objsubid = 0\nAND pgn.nspname = c.table_schema\nleft JOIN\n information_schema.key_column_usage AS kcu\n ON c.table_schema = kcu.table_schema AND c.table_name = kcu.table_name AND c.column_name = kcu.column_name\nLEFT JOIN\n information_schema.table_constraints AS tc\n ON kcu.constraint_name = tc.constraint_name AND kcu.table_schema = tc.table_schema\nLEFT JOIN\n information_schema.referential_constraints AS rc\n ON tc.constraint_name = rc.constraint_name\nLEFT JOIN\n information_schema.constraint_column_usage AS ccu\n ON rc.unique_constraint_name = ccu.constraint_name\nORDER BY\n c.table_name,\n c.ordinal_position;", "options": {} }, "id": "bf692e0e-366f-43a2-94b7-239d6167e903", "name": "Get DB Schema", "type": "n8n-nodes-base.postgres", "typeVersion": 2.6, "position": [ 64, 576 ] }, { "parameters": { "jsCode": "/**\n * @param {Array} items - SQL查询节点返回的完整项目数组。每个 item 包含一个 .json 属性，该属性才是真正的数据库行数据。\n * @returns {string} - 格式化后的Markdown字符串。\n */\nfunction transformSchemaToDDL(items) {\n if (!items || items.length === 0) {\n return '### No Schema Information Provided\\n';\n }\n\n // 第一步：将扁平数据按表名分组，并确保列的唯一性\n const tables = {};\n // 关键修正：遍历 items，并从 item.json 中提取行数据\n for (const item of items) {\n const row = item.json; // <-- 这里是核心修正点！\n\n // 健壮性检查：如果某个item没有.json，则跳过\n if (!row) continue;\n\n const tableName = row.table_name;\n const columnName = row.column_name;\n\n if (!tableName || !columnName) continue;\n\n // 如果是新表，则初始化\n if (!tables[tableName]) {\n tables[tableName] = {\n schema: row.table_schema,\n comment: row.table_comment,\n columns: {}, // 使用对象确保列的唯一性\n foreignKeys: []\n };\n }\n\n // 如果是新列，则创建列的定义\n if (!tables[tableName].columns[columnName]) {\n tables[tableName].columns[columnName] = {\n name: columnName,\n type: row.data_type,\n comment: row.column_comment,\n isPk: false\n };\n }\n\n // 更新该列的约束信息\n if (row.constraint_type === 'PRIMARY KEY') {\n tables[tableName].columns[columnName].isPk = true;\n }\n \n // 记录外键关系（并去重）\n if (row.constraint_type === 'FOREIGN KEY') {\n const fkExists = tables[tableName].foreignKeys.some(\n fk => fk.column === columnName && fk.refTable === row.foreign_table_name\n );\n if (!fkExists) {\n tables[tableName].foreignKeys.push({\n column: columnName,\n refTable: row.foreign_table_name,\n refColumn: row.foreign_column_name\n });\n }\n }\n }\n \n // 用于转义SQL注释中的单引号\n const escapeComment = (comment) => {\n if (!comment) return '';\n return ` COMMENT '${comment.replace(/'/g, \"''\")}'`;\n };\n\n // 第二步：遍历结构化数据，生成DDL字符串\n const ddlStrings = [];\n for (const tableName in tables) {\n const tableData = tables[tableName];\n \n let tableHeader = `-- Table: \"${tableData.schema}\".\"${tableName}\"`;\n if (tableData.comment) {\n tableHeader += ` (Comment: ${tableData.comment})`;\n }\n\n const createTableStart = `CREATE TABLE \"${tableData.schema}\".\"${tableName}\" (`;\n \n const definitions = [];\n\n // 遍历列对象\n for (const column of Object.values(tableData.columns)) {\n let colLine = ` \"${column.name}\" ${column.type}`;\n if (column.isPk) {\n colLine += ' PRIMARY KEY';\n }\n colLine += escapeComment(column.comment);\n definitions.push(colLine);\n }\n \n // 遍历外键\n for (const fk of tableData.foreignKeys) {\n let fkLine = ` FOREIGN KEY (\"${fk.column}\") REFERENCES \"${tableData.schema}\".\"${fk.refTable}\"(\"${fk.refColumn}\")`;\n definitions.push(fkLine);\n }\n \n const createTableBody = definitions.join(',\\n');\n const createTableEnd = ');';\n\n const fullTableDDL = `${tableHeader}\\n${createTableStart}\\n${createTableBody}\\n${createTableEnd}`;\n ddlStrings.push(fullTableDDL);\n }\n\n return ddlStrings.join('\\n\\n');\n}\n\n\n// 调用函数进行转换\nconst schemaMarkdown = transformSchemaToDDL($input.all());\nreturn [{ json: { schema_markdown: schemaMarkdown,schema_name:$input.all()[0].json.table_schema } }];" }, "id": "80e0c2c9-2fbe-46b5-8905-11e77b9785bb", "name": "Format Schema", "type": "n8n-nodes-base.code", "typeVersion": 2, "position": [ 288, 576 ] }, { "parameters": { "jsCode": "//const schemaMarkdown = $input.item.json.schema_markdown;\nconst userInput = $('Extract User Input').first().json.userInput;\nconst sessionId = $('Extract User Input').first().json.sessionId;\nconst history= $('Format History').first().json.formattedHistory;\n\n\nconst outputItem = $input.item;\noutputItem.json.userInput = userInput;\noutputItem.json.sessionId= sessionId;\noutputItem.json.history= history;\noutputItem.json.current_time=new Date();\noutputItem.json.retryCount = 0;\noutputItem.json.maxRetries = 3;\noutputItem.json.error_message='';\nreturn outputItem;" }, "id": "b081caba-f752-47ce-a8da-22b0dbc9bcd1", "name": "Prepare Initial Prompt", "type": "n8n-nodes-base.code", "typeVersion": 2, "position": [ 512, 576 ] }, { "parameters": { "operation": "executeQuery", "query": "={{ $json.output.parseJson().sql }}", "options": {} }, "id": "ddc73eae-3fa4-4352-a010-b2aa33fd288b", "name": "Execute Generated SQL", "type": "n8n-nodes-base.postgres", "typeVersion": 2.6, "position": [ 1600, 432 ], "alwaysOutputData": true, "executeOnce": false, "retryOnFail": false, "onError": "continueRegularOutput" }, { "parameters": {}, "type": "@n8n/n8n-nodes-langchain.memoryBufferWindow", "typeVersion": 1.3, "position": [ 1104, 688 ], "id": "ec708afd-1360-4084-9e35-d5205b866482", "name": "Insert Memory" }, { "parameters": { "conditions": { "options": { "caseSensitive": true, "leftValue": "", "typeValidation": "strict", "version": 2 }, "conditions": [ { "id": "955a825e-5f01-4019-a71b-05433daa0a24", "leftValue": "={{ $json.sql.parseJson().success}}", "rightValue": "false", "operator": { "type": "boolean", "operation": "false", "singleValue": true } }, { "id": "bb72144b-aec1-4af6-a842-5d7be39200fa", "leftValue": "={{ $json.sql.parseJson()['chart_type']}}", "rightValue": "table", "operator": { "type": "string", "operation": "equals", "name": "filter.operator.equals" } } ], "combinator": "or" }, "options": {} }, "type": "n8n-nodes-base.if", "typeVersion": 2.2, "position": [ 2656, 336 ], "id": "e48da06c-09b5-402d-946e-ad837655012e", "name": "If Charts" }, { "parameters": { "conditions": { "options": { "caseSensitive": true, "leftValue": "", "typeValidation": "strict", "version": 2 }, "conditions": [ { "id": "5fca89e7-a943-4292-8e49-fed0ff9bc202", "leftValue": "={{ $json.output.parseJson().success }}", "rightValue": "", "operator": { "type": "boolean", "operation": "true", "singleValue": true } } ], "combinator": "and" }, "options": {} }, "type": "n8n-nodes-base.if", "typeVersion": 2.2, "position": [ 1344, 512 ], "id": "645d0667-9442-42f6-97f0-1dff6fda64e6", "name": "If Ai Error" }, { "parameters": { "modelName": "models/gemini-2.5-pro", "options": { "temperature": 0.2 } }, "type": "@n8n/n8n-nodes-langchain.lmChatGoogleGemini", "typeVersion": 1, "position": [ 832, 688 ], "id": "6833e773-39ff-4531-bef2-facc4b248078", "name": "Google Gemini Chat Model" }, { "id": "caf88a53-9f1a-5b15-9fd8-666ded2d6da0", "name": "Publication · Overview and setup", "type": "n8n-nodes-base.stickyNote", "typeVersion": 1, "position": [ -1952, 0 ], "parameters": { "content": "# nl2sql\n\n## Purpose and structure\nThis folder contains the **nl2sql** workflow. Its executable nodes use `agent`, `code`, `if`, `lmChatDeepSeek`, `lmChatGoogleGemini`, `memoryBufferWindow`, `memoryManager`, `postgres`, `respondToWebhook`, `set`, `webhook`. The table and connection map below are derived from the JSON, not an inferred guarantee of behavior.\n\n## Setup\nImport inactive. Configure your own credentials and replace REPLACE_WITH_* values. Review external destinations and Code nodes before running.\n\n## Entry points\nWebhook\n\n## Verification\nSanitized source candidate; not runtime-tested. This overview is generated from node metadata. Read the folder README for the connection map, credential types, and license.\n\n## Attribution\nhttps://github.com/Wilsonoonn/n8n_nl2sql/blob/fa7c4204d690b1817aa4f916267d8a2f48c5c136/nl2sql.json\nApache-2.0; preserve source notices.", "width": 700, "height": 800, "color": 1 } } ], "connections": { "AI Agent": { "main": [ [ { "node": "If Ai Error", "type": "main", "index": 0 } ] ] }, "Webhook": { "main": [ [ { "node": "Extract User Input", "type": "main", "index": 0 } ] ] }, "Format Success Result": { "main": [ [ { "node": "If Charts", "type": "main", "index": 0 } ] ] }, "Chat Memory Manager": { "main": [ [ { "node": "Format History", "type": "main", "index": 0 } ] ] }, "Format History": { "main": [ [ { "node": "Get DB Schema", "type": "main", "index": 0 } ] ] }, "DeepSeek Chat Model": { "ai_languageModel": [ [ { "node": "AI Agent", "type": "ai_languageModel", "index": 0 } ] ] }, "DeepSeek Chat Model1": { "ai_languageModel": [ [ { "node": "Charts Agent", "type": "ai_languageModel", "index": 0 } ] ] }, "Format Charts": { "main": [ [ { "node": "Return Final Response", "type": "main", "index": 0 } ] ] }, "Charts Agent": { "main": [ [ { "node": "Format Charts", "type": "main", "index": 0 } ] ] }, "Loop Entry": { "main": [ [ { "node": "AI Agent", "type": "main", "index": 0 } ] ] }, "If Can Retry": { "main": [ [ { "node": "Loop Entry", "type": "main", "index": 0 } ], [ { "node": "Format Success Result", "type": "main", "index": 0 } ] ] }, "Format Error Prompt": { "main": [ [ { "node": "If Can Retry", "type": "main", "index": 0 } ] ] }, "If Error": { "main": [ [ { "node": "Format Error Prompt", "type": "main", "index": 0 } ], [ { "node": "Format Success Result", "type": "main", "index": 0 } ] ] }, "Search Memory": { "ai_memory": [ [ { "node": "Chat Memory Manager", "type": "ai_memory", "index": 0 } ] ] }, "Extract User Input": { "main": [ [ { "node": "Chat Memory Manager", "type": "main", "index": 0 } ] ] }, "Get DB Schema": { "main": [ [ { "node": "Format Schema", "type": "main", "index": 0 } ] ] }, "Format Schema": { "main": [ [ { "node": "Prepare Initial Prompt", "type": "main", "index": 0 } ] ] }, "Prepare Initial Prompt": { "main": [ [ { "node": "Loop Entry", "type": "main", "index": 0 } ] ] }, "Execute Generated SQL": { "main": [ [ { "node": "If Error", "type": "main", "index": 0 } ], [] ] }, "Insert Memory": { "ai_memory": [ [ { "node": "AI Agent", "type": "ai_memory", "index": 0 } ] ] }, "If Charts": { "main": [ [ { "node": "Format Charts", "type": "main", "index": 0 } ], [ { "node": "Charts Agent", "type": "main", "index": 0 } ] ] }, "If Ai Error": { "main": [ [ { "node": "Execute Generated SQL", "type": "main", "index": 0 } ], [ { "node": "Format Charts", "type": "main", "index": 0 } ] ] } }, "active": false, "settings": { "executionOrder": "v1" } }