教你如何用n8n实现一个知识库

   Index Name:    knowledge-base
   Dimensions:    1024          ← 重要！必须匹配Embedding模型输出维度
   Metric:        cosine
   Region:        选择离你最近的区域

   Credential name:  ModelScope API
   Name:             Authorization
   Value:            Bearer ms-your-token-here

   Credential name:  Pinecone API
   Name:             Api-Key
   Value:            your-pinecone-api-key

https://your-n8n-domain.com/webhook/knowledge/upload

curl -X POST https://your-n8n-domain.com/webhook/knowledge/upload \
  -H "Content-Type: application/json" \
  -H "x-api-key: your-custom-api-key-here" \
  -d '{
    "content": "这是一个测试文档内容。",
    "filename": "test-document.txt"
  }'

Conditions > String
  Value 1:  ={{ $json.headers['x-api-key'] }}
  Operation: equal
  Value 2:  your-custom-api-key-here

const text = $input.first().json.body.content || '';
const filename = $input.first().json.body.filename || 'unknown';
const chunkSize = 800;       // 每块800字符
const overlap = 200;          // 重叠200字符，避免语义断裂

function chunkText(text, size, overlap) {
  const chunks = [];
  let chunkIndex = 0;
  const paragraphs = text.split(/\n\n+/);  // 按段落分割
  let currentChunk = '';
  
  for (const para of paragraphs) {
    // 如果当前块+新段落超过限制，且当前块不为空
    if ((currentChunk + para).length > size && currentChunk.length > 0) {
      chunks.push({ 
        content: currentChunk.trim(), 
        index: chunkIndex++ 
      });
      
      // 保留部分重叠内容（直接按字符数截取，支持中英文）
      const overlapText = currentChunk.slice(-overlap);
      currentChunk = overlapText + '\n\n' + para;
    } else {
      currentChunk += (currentChunk ? '\n\n' : '') + para;
    }
  }
  
  // 添加最后一块
  if (currentChunk.trim()) {
    chunks.push({ 
      content: currentChunk.trim(), 
      index: chunkIndex 
    });
  }
  
  return chunks;
}

const chunks = chunkText(text, chunkSize, overlap);

// 返回多个item，每个item代表一个文本块
return chunks.map(chunk => ({
  json: {
    text: chunk.content,
    metadata: {
      source: filename,
      chunkIndex: chunk.index,
      totalChunks: chunks.length,
      uploadTime: new Date().toISOString()
    }
  }
}));

Name:   Content-Type
Value:  application/json

  {{ JSON.stringify({ 
    model: 'Qwen/Qwen3-Embedding-0.6B', 
    input: $json.text, 
    encoding_format: 'float' 
  }) }}

{
  "data": {
    "data": [{
      "embedding": [
        -0.0172576904296875,
        -0.0867919921875,
        ...  // 共1024个浮点数
      ],
      "index": 0,
      "object": "embedding"
    }],
    "model": "Qwen/Qwen3-Embedding-0.6B",
    "object": "list",
    "usage": {
      "prompt_tokens": 12,
      "total_tokens": 12
    }
  }
}

// 解析ModelScope返回的数据
const responseData = JSON.parse($input.first().json.data);
const embedding = responseData.data[0].embedding;

// 获取上游节点的元数据
const metadata = $('文本分块').first().json.metadata;
const text = $('文本分块').first().json.text;

// 生成ASCII安全的ID（Pinecone不支持中文ID）
const safeId = `doc_${Date.now()}_${Math.random().toString(36).substr(2, 9)}_${metadata.chunkIndex}`;

// 返回Pinecone标准格式
return [{
  json: {
    id: safeId,                    // 唯一ID，纯ASCII字符
    values: embedding,              // 1024维向量数组
    metadata: {                     // 元数据（可包含中文）
      text: text,                   // 原始文本
      source: metadata.source,      // 文件名
      chunkIndex: metadata.chunkIndex,
      totalChunks: metadata.totalChunks,
      uploadTime: metadata.uploadTime
    }
  }
}];

Name:   Content-Type
Value:  application/json

  {{ JSON.stringify({ vectors: [$json] }) }}

{
  "upsertedCount": 1
}

Respond With: JSON
Response Body: (使用表达式)

{{ { 
  success: true, 
  message: '文档索引完成', 
  chunks: $('文本分块').all().length, 
  filename: $('文本分块').first().json.metadata.source,
  timestamp: new Date().toISOString()
} }}

{
  "success": true,
  "message": "文档索引完成",
  "chunks": 3,
  "filename": "product-manual.pdf",
  "timestamp": "2025-10-27T12:30:45.123Z"
}

Respond With: JSON
Response Code: 401

{{ { 
  error: 'Unauthorized', 
  message: '无效的API Key',
  timestamp: new Date().toISOString()
} }}

{
  "error": "Unauthorized",
  "message": "无效的API Key",
  "timestamp": "2025-10-27T12:30:45.123Z"
}

curl -X POST https://your-n8n-domain.com/webhook/knowledge/upload \
  -H "Content-Type: application/json" \
  -H "x-api-key: your-custom-api-key-here" \
  -d '{
    "content": "产品特点：支持多语言、实时翻译、智能推荐。\n\n使用方法：登录系统，选择语言，输入内容即可自动翻译。\n\n常见问题：如何切换语言？点击右上角设置按钮，选择目标语言。",
    "filename": "product-guide.txt"
  }'

{
  "success": true,
  "message": "文档索引完成",
  "chunks": 1,
  "filename": "product-guide.txt",
  "timestamp": "2025-10-27T12:30:45.123Z"
}

{
  "error": "Unauthorized",
  "message": "无效的API Key",
  "timestamp": "2025-10-27T12:30:45.123Z"
}

Bad request - please check your parameters
Vector dimension 1024 does not match the dimension of the index 768

Bad request - please check your parameters
Vector ID must be ASCII, but got '产品手册.pdf_0'

// ❌ 错误写法
const id = `${metadata.source}_${metadata.chunkIndex}`;  // 包含中文

// ✅ 正确写法
const safeId = `doc_${Date.now()}_${Math.random().toString(36).substr(2, 9)}_${metadata.chunkIndex}`;