关于 cloudflare-vectorize
面向 Cloudflare Workers 开发者的 Cloudflare Vectorize 实现指南,涵盖向量索引、检索和 RAG 工作流,并提供 TypeScript 模板及配置参考。
主要功能
- 01管理向量索引与向量
- 02相似度搜索和元数据筛选
- 03Workers AI 与 OpenAI 嵌入
- 04文档分块与批量导入
- 05提供搜索和 RAG 模板
适用场景
- 01为文档或内容构建语义搜索
- 02实现带上下文检索的 RAG 聊天
- 03按元数据筛选相似向量
安装
复制下面的命令,在终端中运行。
npx skills add https:/ / github.com/ jackspace/ claudeskillz --skill cloudflare-vectorize安装后在 Claude Code 中输入/cloudflare-vectorize确认已加载。
附带文件README.mdSKILL.json
详细介绍
Cloudflare Vectorize
Cloudflare Vectorize 的完整实施指南——这是一个全球分布式向量数据库,可与 Cloudflare Workers 配合构建语义搜索、RAG(检索增强生成)和 AI 驱动的应用。
状态:已可用于生产环境 ✅ 最后更新:2025-10-21 依赖项:cloudflare-worker-base(用于 Worker 设置)、cloudflare-workers-ai(用于嵌入) 最新版本:[email protected]、@cloudflare/[email protected] 节省的 Token:约 65% 避免的错误:8 节省的开发时间:约 3 小时
此技能提供的内容
核心功能
- ✅ 索引管理:创建、配置和管理向量索引
- ✅ 向量操作:插入、更新或插入、查询、删除和列出向量
- ✅ 元数据过滤:每个索引最多支持 10 个元数据索引的高级过滤
- ✅ 语义搜索:使用余弦、欧几里得或点积度量查找相似向量
- ✅ RAG 模式:完整的检索增强生成工作流
- ✅ Workers AI 集成:使用 @cf/baai/bge-base-en-v1.5 原生生成嵌入
- ✅ OpenAI 集成:支持 text-embedding-3-small/large 模型
- ✅ 文档处理:文本分块和批量摄取管道
包含的模板
- basic-search.ts - 使用 Workers AI 的简单向量搜索
- rag-chat.ts - 支持上下文检索的完整 RAG 聊天机器人
- document-ingestion.ts - 文档分块和嵌入管道
- metadata-filtering.ts - 高级过滤示例
关键设置规则
⚠️ 插入向量前必须执行
# 1. Create the index with FIXED dimensions and metric
npx wrangler vectorize create my-index \
--dimensions=768 \
--metric=cosine
# 2. Create metadata indexes IMMEDIATELY (before inserting vectors!)
npx wrangler vectorize create-metadata-index my-index \
--property-name=category \
--type=string
npx wrangler vectorize create-metadata-index my-index \
--property-name=timestamp \
--type=number
原因:必须先创建元数据索引,然后才能插入向量。在元数据索引创建之前添加的向量,无法按该属性进行过滤。
索引配置(之后无法更改)
# Dimensions MUST match your embedding model output:
# - Workers AI @cf/baai/bge-base-en-v1.5: 768 dimensions
# - OpenAI text-embedding-3-small: 1536 dimensions
# - OpenAI text-embedding-3-large: 3072 dimensions
# Metrics determine similarity calculation:
# - cosine: Best for normalized embeddings (most common)
# - euclidean: Absolute distance between vectors
# - dot-product: For non-normalized vectors
Wrangler 配置
wrangler.jsonc:
{
"name": "my-vectorize-worker",
"main": "src/index.ts",
"compatibility_date": "2025-10-21",
"vectorize": [
{
"binding": "VECTORIZE_INDEX",
"index_name": "my-index"
}
],
"ai": {
"binding": "AI"
}
}
TypeScript 类型
export interface Env {
VECTORIZE_INDEX: VectorizeIndex;
AI: Ai;
}
interface VectorizeVector {
id: string;
values: number[] | Float32Array | Float64Array;
namespace?: string;
metadata?: Record<string, string | number | boolean | string[]>;
}
interface VectorizeMatches {
matches: Array<{
id: string;
score: number;
values?: number[];
metadata?: Record<string, any>;
namespace?: string;
}>;
count: number;
}
常见操作
1. 插入与更新或插入
// INSERT: Keeps first insertion if ID exists
await env.VECTORIZE_INDEX.insert([
{
id: "doc-1",
values: [0.1, 0.2, 0.3, ...],
metadata: { title: "First version" }
}
]);
// UPSERT: Overwrites with latest if ID exists (use this for updates)
await env.VECTORIZE_INDEX.upsert([
{
id: "doc-1",
values: [0.1, 0.2, 0.3, ...],
metadata: { title: "Updated version" }
}
]);
2. 使用过滤器查询
// Generate embedding for query
const queryEmbedding = await env.AI.run('@cf/baai/bge-base-en-v1.5', {
text: "What is Cloudflare Workers?"
});
// Search with metadata filtering
const results = await env.VECTORIZE_INDEX.query(
queryEmbedding.data[0],
{
topK: 5,
filter: {
category: "documentation",
timestamp: { $gte: 1704067200 } // After Jan 1, 2024
},
returnMetadata: 'all',
returnValues: false,
namespace: 'prod'
}
);
3. 元数据过滤运算符
// Equality (implicit $eq)
{ category: "docs" }
// Explicit operators
{ status: { $ne: "archived" } }
// In array
{ category: { $in: ["docs", "tutorials", "guides"] } }
// Not in array
{ category: { $nin: ["deprecated", "draft"] } }
// Range queries (numbers)
{
timestamp: {
$gte: 1704067200, // >= Jan 1, 2024
$lt: 1735689600 // < Jan 1, 2025
}
}
// Range queries (strings) - prefix searching
{
url: {
$gte: "/docs/workers",
$lt: "/docs/workersz" // Matches all /docs/workers/*
}
}
// Nested metadata with dot notation
{ "author.id": "user123" }
// Multiple conditions (implicit AND)
{
category: "docs",
language: "en",
"metadata.published": true
}
4. 命名空间过滤
// Insert with namespace (partition key)
await env.VECTORIZE_INDEX.upsert([
{
id: "1",
values: embedding,
namespace: "customer-123",
metadata: { type: "support_ticket" }
}
]);
// Query only within namespace
const results = await env.VECTORIZE_INDEX.query(queryVector, {
topK: 5,
namespace: "customer-123" // Only search this customer's data
});
5. 列出和删除向量
// List vector IDs (paginated)
const vectors = await env.VECTORIZE_INDEX.listVectors({
cursor: null,
limit: 100
});
// Get specific vectors by ID
const retrieved = await env.VECTORIZE_INDEX.getByIds([
"doc-1", "doc-2", "doc-3"
]);
// Delete vectors
await env.VECTORIZE_INDEX.deleteByIds([
"doc-1", "doc-2"
]);
嵌入生成
Workers AI(推荐——免费)
const embeddings = await env.AI.run('@cf/baai/bge-base-en-v1.5', {
text: ["Document 1 content", "Document 2 content"]
});
// embeddings.data is number[][] (array of 768-dim vectors)
const vectors = embeddings.data.map((values, i) => ({
id: `doc-${i}`,
values,
metadata: { source: 'batch-import' }
}));
await env.VECTORIZE_INDEX.upsert(vectors);
OpenAI 嵌入
import OpenAI from 'openai';
const openai = new OpenAI({ apiKey: env.OPENAI_API_KEY });
const response = await openai.embeddings.create({
model: "text-embedding-3-small", // 1536 dimensions
input: "Text to embed"
});
await env.VECTORIZE_INDEX.upsert([{
id: "doc-1",
values: response.data[0].embedding,
metadata: { model: "openai-3-small" }
}]);
元数据最佳实践
1. 基数注意事项
低基数(适用于 $eq 过滤器):
// Few unique values - efficient filtering
metadata: {
category: "docs", // ~10 categories
language: "en", // ~5 languages
published: true // 2 values (boolean)
}
高基数(避免用于范围查询):
// Many unique values - avoid large range scans
metadata: {
user_id: "uuid-v4...", // Millions of unique values
timestamp_ms: 1704067200123 // Use seconds instead
}
2. 元数据限制
- 元数据索引最多 10 个(每个 Vectorize 索引)
- 元数据最多 10 KiB(每个向量)
- 字符串索引:前 64 个字节(UTF-8)
- 数字索引:Float64 精度
- 过滤器大小:最多 2048 字节(紧凑 JSON)
3. 键限制
// ❌ INVALID metadata keys
metadata: {
"": "value", // Empty key
"user.name": "John", // Contains dot (reserved for nesting)
"$admin": true, // Starts with $
"key\"with\"quotes": 1 // Contains quotes
}
// ✅ VALID metadata keys
metadata: {
"user_name": "John",
"isAdmin": true,
"nested": { "allowed": true } // Access as "nested.allowed" in filters
}
RAG 模式(完整示例)
export default {
async fetch(request: Request, env: Env): Promise<Response> {
const { question } = await request.json();
// 1. Generate embedding for user question
const questionEmbedding = await env.AI.run('@cf/baai/bge-base-en-v1.5', {
text: question
});
// 2. Search vector database for similar content
const results = await env.VECTORIZE_INDEX.query(
questionEmbedding.data[0],
{
topK: 3,
returnMetadata: 'all',
filter: { type: "documentation" }
}
);
// 3. Build context from retrieved documents
const context = results.matches
.map(m => m.metadata.content)
.join('\n\n---\n\n');
// 4. Generate answer with LLM using context
const answer = await env.AI.run('@cf/meta/llama-3-8b-instruct', {
messages: [
{
role: "system",
content: `Answer based on this context:\n\n${context}`
},
{
role: "user",
content: question
}
]
});
return Response.json({
answer: answer.response,
sources: results.matches.map(m => m.metadata.title)
});
}
};
文档分块策略
function chunkText(text: string, maxChunkSize = 500): string[] {
const sentences = text.match(/[^.!?]+[.!?]+/g) || [text];
const chunks: string[] = [];
let currentChunk = '';
for (const sentence of sentences) {
if ((currentChunk + sentence).length > maxChunkSize && currentChunk) {
chunks.push(currentChunk.trim());
currentChunk = sentence;
} else {
currentChunk += sentence;
}
}
if (currentChunk) chunks.push(currentChunk.trim());
return chunks;
}
// Usage
const chunks = chunkText(longDocument, 500);
const embeddings = await env.AI.run('@cf/baai/bge-base-en-v1.5', {
text: chunks
});
const vectors = embeddings.data.map((values, i) => ({
id: `doc-${docId}-chunk-${i}`,
values,
metadata: {
doc_id: docId,
chunk_index: i,
total_chunks: chunks.length,
content: chunks[i]
}
}));
await env.VECTORIZE_INDEX.upsert(vectors);
常见错误与解决方案
错误 1:插入向量后才创建元数据索引
Problem: Filtering doesn't work on existing vectors
Solution: Delete and re-insert vectors OR create metadata indexes BEFORE inserting
错误 2:维度不匹配
Problem: "Vector dimensions do not match index configuration"
Solution: Ensure embedding model output matches index dimensions:
- Workers AI bge-base: 768
- OpenAI small: 1536
- OpenAI large: 3072
错误 3:无效的元数据键
Problem: "Invalid metadata key"
Solution: Keys cannot:
- Be empty
- Contain . (dot)
- Contain " (quote)
- Start with $ (dollar sign)
错误 4:过滤器过大
Problem: "Filter exceeds 2048 bytes"
Solution: Simplify filter or split into multiple queries
错误 5:对高基数数据执行范围查询
Problem: Slow queries or reduced accuracy
Solution: Use lower cardinality fields for range queries, or use seconds instead of milliseconds for timestamps
错误 6:混淆插入与更新或插入
Problem: Updates not reflecting in index
Solution: Use upsert() to overwrite existing vectors, not insert()
错误 7:缺少绑定
Problem: "VECTORIZE_INDEX is not defined"
Solution: Add [[vectorize]] binding to wrangler.jsonc
错误 8:混淆命名空间与元数据
Problem: Unclear when to use namespace vs metadata filtering
Solution:
- Namespace: Partition key, applied BEFORE metadata filters
- Metadata: Flexible key-value filtering within namespace
Wrangler CLI 参考
# Create index (dimensions and metric cannot be changed later!)
npx wrangler vectorize create <name> \
--dimensions=768 \
--metric=cosine
# List indexes
npx wrangler vectorize list
# Get index details
npx wrangler vectorize get <name>
# Get index info (vector count, mutations)
npx wrangler vectorize info <name>
# Delete index
npx wrangler vectorize delete <name>
# Create metadata index (BEFORE inserting vectors!)
npx wrangler vectorize create-metadata-index <name> \
--property-name=category \
--type=string
# List metadata indexes
npx wrangler vectorize list-metadata-index <name>
# Delete metadata index
npx wrangler vectorize delete-metadata-index <name> \
--property-name=category
# Insert vectors from file
npx wrangler vectorize insert <name> \
--file=vectors.ndjson
# Query vectors
npx wrangler vectorize query <name> \
--vector="[0.1, 0.2, ...]" \
--top-k=5 \
--return-metadata=all
# List vector IDs
npx wrangler vectorize list-vectors <name> \
--count=100
# Get vectors by IDs
npx wrangler vectorize get-vectors <name> \
--ids="id1,id2,id3"
# Delete vectors by IDs
npx wrangler vectorize delete-vectors <name> \
--ids="id1,id2,id3"
性能提示
- 批量操作:每批插入或更新或插入 100-1000 个向量
- 选择性返回:仅在需要时使用
returnValues: true(可节省带宽) - 元数据基数:范围查询所用的已索引元数据字段应保持低基数
- 命名空间过滤:先应用命名空间过滤器,再应用元数据过滤器(优先处理)
- 查询优化:使用 topK=3-10 可获得最佳延迟(较大的值会增加搜索时间)
适用场景
✅ 适合使用 Vectorize 的场景:
- 为文档、产品或内容构建语义搜索
- 实现可检索上下文的 RAG 聊天机器人
- 根据相似度创建推荐引擎
- 构建多租户应用(使用命名空间)
- 需要全球分发和低延迟
❌ 不适合使用 Vectorize 的场景:
- 传统关系型数据(使用 D1)
- 键值查找(使用 KV)
- 大型文件存储(使用 R2)
- 实时协作状态(使用 Durable Objects)
模板位置
所有可用的代码示例都在 ./templates/ 中:
basic-search.ts- 简单的向量搜索实现rag-chat.ts- 完整的 RAG 聊天机器人document-ingestion.ts- 文档处理流水线metadata-filtering.ts- 高级筛选模式
参考文档
详细指南位于 ./references/:
wrangler-commands.md- 完整的 CLI 参考index-operations.md- 索引创建和管理vector-operations.md- 插入、查询和删除操作metadata-guide.md- 元数据索引和筛选embedding-models.md- 模型配置
集成示例
完整的集成指南位于 ./references/:
integration-workers-ai-bge-base.md- Workers AI 集成(@cf/baai/bge-base-en-v1.5)integration-openai-embeddings.md- OpenAI 嵌入集成
官方文档
版本:1.0.0 状态:可用于生产环境 ✅ 节省的 Token:约 65% 避免的错误:8 个主要类别 节省的开发时间:每次实现约 2.5 小时